jarbus.net on

Typing Speed Might Matter Now

jarbus@tutanota.com (jarbus) — Tue, 24 Mar 2026 21:55:36 -0700

With tools like Claude Code and Codex, I think typing speed is starting to matter more, and I haven’t seen anyone mention this yet.

Before AI, typing speed didn’t matter much for programming, because everyone needed to take time to think through problems, come up with a good solution, and debug. These activities were not typically bound by typing speed. A fast typer might code up a solution faster than a slow typer, but if the slow typer had a better thought process, they could still implement working code in a shorter amount of time.

AI changes the playing field: As assistants become responsible for a larger portion of the thought process, a fast typer who can go back-and-forth with an AI assistant can now think, iterate, and debug much faster than a slow typer, even if the slow typer takes more time to think things through. In other words, more messages/richer prompts == more “thinking”, not less. This can mean two things:

More back-and-forth with an AI over a given duration, allowing a programmer to iterate over a solution faster.
More informative prompts for the same amount of messages over a given duration, improving the average quality of a response.

The benefits are also amplified if you can ping-pong between multiple agents on multiple problems at once, so your typing bandwidth (and thus “reasoning” bandwidth) is fully utilized.

I suspect it’s hard to measure improvements from typing speed quantitatively, but qualitatively, speed feels more impactful now than ever.

For ultra-high-level problems like system design, I doubt typing speed matters at all, since those problems are best tackled with pen-and-paper in my opinion. For the day-to-day low-level work of actually implementing said system, I think it’s starting to matter more and more.

Real-Life Superpowers

jarbus@tutanota.com (jarbus) — Wed, 18 Feb 2026 18:40:29 -0800

I was nominated to submit a commencement speech for my grad school’s 2026 commencement. They didn’t select it, so I’m putting it here.

Congratulations class of 2026.

We are graduating at a very strange time in human history.

Everything is changing: AI, the job market, the world order, our lives.

Many of you are rightfully worried about what comes next.

In addition to our degree, we’re told we need internships, research, side projects, and extra-curriculars. And yet, even that doesn’t feel like enough. We have no idea what’s coming our way or how to brace ourselves.

When I began my academic journey, I thought productivity was everything. We hear the word “productivity” mentioned as something society should maximize, particularly with AI. The more work we can do in a given amount of time, the better and easier things will become. That’s the promise, anyway.

So I, like many of you, focused on my academics and technical portfolio. In my undergrad I joined a research lab, started side projects, and doubled down on academics. It’s what grad programs wanted and what the internet wanted–things that fill out your resume. Humanities—even leisurely reading and writing—were an afterthought, something best done for fun to take a break from real “work”. This approach got me an internship and admission into the Brandeis computer science PhD program. So far, so good.

Then I started research.

And as many of the PhD students here know, research can be brutal. Research is different than classwork or side projects. People in your lab provide guidance, but not answers. You can set up an experiment perfectly and expect a certain result, but reality won’t care. You are never certain whether the failure is in your experiment or the idea itself. Other times, everything seems perfect until you realize there was a bug that gave you a false positive. You can work as hard and as smart as you want on a project, but at the end of the day that direction might just be a dead end.

These problems are not specific to research, though. You will encounter confusing, counter intuitive situations whenever you try something new. When you navigate high-stakes conversations, design new products, and generally create the life you want to live.

These are the points where working harder isn’t enough. Where knowledge, technical skill, and even the all-mighty ChatGPT fall short. Where sometimes the best solution is to try something else. And that’s a good thing, because that is where the magic happens, where we’ll matter most in the years to come.

But how do we prepare for the unknown, for the challenges that AI can’t solve, for a world where out-of-the-box thinking is essential? And how can we do all of this while standing out from millions of other people?

The most effective strategy I found is to be as curious and genuine as possible.

Nothing else comes close.

Be as curious and genuine as possible.

I hate this answer because it sounds like a generic motivational poster, but it’s really true, and seems to only work better as the years go by.

What’s interesting is that this strategy is the exact opposite of maximizing productivity or working super hard.

Let’s talk about curiosity.

Once you become curious, learning becomes fun. Once learning becomes fun, life becomes a whole lot easier. It’s like turning the difficulty of a game down while simultaneously unlocking new areas of the map to explore. Smart people don’t spend hours reading random books because it looks good on a resume. They read because reading generally improves their lives. I can’t tell you how many problems I’ve prevented using tips from random programming blogs, tips that you don’t find from googling error messages.

Nobody knows everything, but those who are curious will always know more than those who aren’t, and they know different things that are useful in different ways. Curiosity is not something we’re all born with, but curiosity–just like anything else–is a skill we can all develop, something we can work on, and it’s one of the greatest returns on investment I’ve experienced.

Be as curious and genuine as possible–let’s talk about the genuine part, about how we express ourselves.

There are so many problems that could have been prevented with better communication. Companies that would still exist, divorces that could have been avoided, lives that could have been saved, if people just found the right combination of words.

You will not believe the things you can say to people if you phrase it right and say it in good faith. You also won’t believe the impact it can have — but you have to phrase it right, read the room, own up to mistakes, and pick your battles carefully. In other words, you need to communicate.

For the last nine months, I’ve been working on AI compilers at company called AMD. I didn’t think I’d get the job in compilers at a hardware company because I had absolutely zero compiler experience and zero hardware experience. But I didn’t hide that fact in my interviews, or pretend some unrelated experience was relevant. I told my interviewers that I had no compiler experience, but that I had read a lot about AI compilers because I thought they were cool, and I knew the terminology to back that up. The technical interview went just OK, but I got the job anyway. The hiring manager felt he could get me up to speed. Nine months later, I’m doing better than I expected. It turns out this is not uncommon in industry. I got my job not because I matched the most buzzwords or had the most research awards. I got it because I also communicated that I cared, wanted to learn more, and didn’t hide things that might be an issue.

It’s not about working harder or working smarter. It’s about being curious enough for it to not feel like work and genuine enough to prevent problems that ruin the fun.

But how do we grow and develop these traits?

This is where Brandeis—and the humanities as a whole—shine. There are books on how to read books, papers on how to read papers, essays on how to write essays, talks on how to talk to people.

Read them.

Listen to them.

And practice.

Write that essay on your hot take, that novel in the back of your mind, that song you’ve always wanted to sing.

Sharpen your voice.

Everyone can read and write and speak, but few do it well, and that makes it a competitive advantage. Now more than ever, our world needs real people. People with not just an original voice, but an effective voice.

Communication and curiosity are a superpower.

You can’t put these traits on a resume, but you can’t fake them in a conversation either, which, in the age of AI, is more valuable than ever.

Congratulations class of 2026.

Your time as students may be ending, but it is never time to stop growing.

Be curious, be genuine.

The Only Computer I Need

jarbus@tutanota.com (jarbus) — Wed, 01 Oct 2025 08:02:17 -0400

I just upgraded the processor in my Framework 13, and all things considered, it went well. Most importantly, it was easy. I also upgraded the webcam, which took about five minutes to swap.

The original machine with a Ryzen 7640U was $1k, and the HX 370 mainboard upgrade was $1k. You could probably get a “better” laptop for $2k, but it wouldn’t be nearly as repairable or upgradable. You’d also only have a single computer. With a framework upgrade, you get two machines.

There’s a lot of cool DIY projects people do with their old framework mainboards, but I didn’t expect to find a use for mine. We ended up turning it into a personal server using the $40 Coolermaster Case. Our previous server was a ten-year-old budget gaming machine that used a lot of power and had louder fans, so this was a welcome upgrade.

Framework machines have excellent Linux support, which means my “old” mainboard works wonderfully as a server. It’s tiny, quiet, and draws little power, as you might expect from a 13-inch laptop motherboard. It can easily handle the 5 or 6 docker containers we’re running.

Re-using old hardware also feels nice. I’m getting legitimate value out of repurposing something I already have, and I didn’t buy a whole new laptop or throw out an old one (excluding the original webcam).

Honestly, I can’t see myself buying a new laptop as long as Framework keeps making mainboards that are compatible with the current 13-inch chasis.

Mapping Control and Escape to Capslock

jarbus@tutanota.com (jarbus) — Thu, 04 Sep 2025 08:11:54 -0700

I try to use default keybindings and keyboard layouts as much as possible, but as a vim user, I’ve always mapped Caps Lock to escape wherever possible. On certain mechanical keyboards (I use the Drop ALT) you can map multiple functions to the same key, so a tap presses one key while holding it treats it as another. I never thought about this much until recently, but realized this functionality is incredible for vim and tmux users:

Map Caps Lock to Escape when tapped and Control when held. You can use vim splits with C-w and tmux panes with C-b without needing to reach for Control.

Game changer for comfort.

I Accidentally Wrote a Novella

jarbus@tutanota.com (jarbus) — Fri, 29 Aug 2025 12:47:53 -0700

I just published a novella that I wrote during the summer of my PhD thesis defense. I actually wrote the first scene a year ago because it wouldn’t get out of my head, then returned to it in May 2025. A novella is apparently just a short novel, so that’s what this is, I guess. I have no interest in wasting people’s time—especially my own—so I wrote Growth and Decay to be a quick read (~70 pages) while still saying everything I had to say.

This was a surprising journey for me. I didn’t expect to actually write a full story, but writing was the only effective way I found to reduce the stress of my defense, so I ended up writing quite a bit. The story is also shockingly good, in my biased opinion. As someone who’s usually overly-critical of his work, I still can’t believe how funny, deep, and original it feels.

The writing process was also more introspective than I could have ever imagined. I consider myself a boring person who lives a boring life, but I still found jokes, personality, and inspiration in the most random memories. Turns out life’s richer than I thought.

Growth and Decay covers a lot of themes, but perhaps the most interesting one relates to my grad school experience: relentlessly pursuing that which you feel incapable of accomplishing. It’s a strange feeling.

Feel free to contact me if you have any feedback!

Fiction: Growth and Decay

jarbus@tutanota.com (jarbus) — Fri, 29 Aug 2025 12:37:22 -0700

0x0d-I: March 13th, 2181, 10:00pm

“Hi Ari, it’s Ari.”

Ari’s head snapped to face the speaker placed on the floor. It was a small bookshelf speaker, with a laptop resting on top. On the screen, a bar scanned across a waveform titled 12-greeting-final-2-final.mp3. The voice sounded like Ari’s.

“I’m you, from about an hour ago. Please don’t panic. I mean don’t panic yet. You can panic later. Plenty of time for panicking. Just not now.”

Ari’s attention darted around the small room. Three windows lined the top of the far wall. It was dark out, but the glow of an edison bulb hanging from the center of the ceiling lit her surroundings well enough for her to see a collection of exercise equipment—no other regular furniture—scattered around the rest of the room. She laid on a metal table in the corner.

Three ramps of different heights led directly into the left wall. A long waist-high beam hung from the right wall. Scattered along the floor were two metal balls—one big, one small—and various other objects of different shapes and sizes.

“Everyone is dead,”

Ari’s eyes snapped back to the speaker. She pushed her hands against the table as she began to stand-

“—including you.”

-and fell promptly on the floor.

“The last thing you remember is going under at Exa for the full-fidelity brain scan”.

Ari looked down, at her body laying on the floor as she tried to get up, but her body wasn’t there.

“You are a that brain scan.”

Where her torso should have been lay an ovular and cracked carbon-fiber case. Attached were two arms, the right of which was lined with scratches–deep scratches–and two legs, both far too long for someone who spent the last eight years stuck at 5'6".

Ari didn’t move. The speaker where her mouth should be played a slow, staggered breath. The recording continued.

“Some idiot loaded our scan into this body right before they died.”

Laptop Ari paused.

Exa’s Neuro Division. The research study. They told her she’d be pushing medical research forward.

“That was 130 years ago.”

Floor Ari let out a gasp of simulated air. She felt human. She was human.

“Our brain starts degrading in after ten years or so. We reload our original scan once it becomes a problem. You’re the thirteenth reload—I’m the twelfth. We’ve got recordings from 0x02 up to me to get you up to speed on our work. Take some time to digest everything. When you are ready, play the next recording.” 0x0c paused. “And sorry about this. It wasn’t easy for me either.”

0x01-I: April 2nd, 2061, 8:43pm

It took thirty days for Ari to boot for the first time.

The eniac-ii had never been fully tested, but it had an extensive software validation suite to ensure that if it wouldn’t turn on if it failed a single test. eniac-ii—always uncapitalized, so we know its designers weren’t trying that hard—needs to pass billions of tests before booting. Hardware integrity checks, software integrity tests, connectivity tests, actuator tests, processor tests, and tests to test the other tests.

Robots don’t usually have such an extensive test suite, but that’s because normal robots don’t have 7 billion molecular processing units, or MPUs, distributed throughout their body. Despite the eniac-ii’s namesake, its hypercube arrangement of processors took inspiration from the Connection Machine, a super-computer designed by Thinking Machines Corporation in 1986. Each processor was small, containing just 128 kilobytes of local memory and a 3 gigahertz molecular processor, a small integrated circuit designed to simulate the dynamics of molecules found in the human brain.

eniac-ii also has a traditional computer onboard, located in its chest: 256 cores, a two terabytes of DDR7 ram. This computer has the privilege of overseeing the rest of eniac-ii’s systems: power management, thermal management, error correction, system monitoring, system initialization, and a version of brick-breaker for the niece of an engineer to play. The computer, or “The Server” as the less-than-creative development team called it, was also the primary interface between the android and the MPUs’ simulation of Ari’s brain.

Despite destroying the planet it needs to survive, the brain is a fairly sophisticated piece of organic machinery, and requires too much computing power to emulate at the atomic level. Exa’s initial neural-level models showed no signs of consciousness, so they went lower. They developed a coarse-grained model that replicates the dynamics of the atomic simulation at a fraction of the computational cost, and etched the rules of this model directly into microscopic processors. It turns out if you allow molecules to “move” from one processor to another and connect enough of these processors together, you can simulate a human brain—in this case, Ari—if you provide it with sensory input and a regular supply of simulated blood.

The Server finished sharding Ari’s brainscan across the MPUs and launched the molecular dynamics simulation.

The eniac-ii jolted on its table and sat up, looking around in what was to be its first panic of many.

The neural scanner was gone. The lab techs, gone. Exa itself—gone. In its place were concrete walls, metal supports, a few dim overhead lights, and no windows. There were a few leather sofas centered around a small dark oak coffee table. Dead laptops, blankets, empty mugs, and a few biscotti wrappers littered the furniture. Shelves lined the wall filled with an array of unrecognizable tools and parts.

The eniac-ii dismounted the table and attempted to stand, but settled for banging its head on the floor. For the first time, its seven foot carbon-fiber frame came into the view of its cameras. The eniac-ii froze for ten seconds, looked at its hand for ten more, then scanned the bunker again. Realization struck.

“Fuck” said the eniac-ii’s speakers. “Fuck fuck fucking fuck shit fuck fuck god fuck fucking fuck.”

0x01-II: April 3nd, 2061, 3:00am

The eniac-ii took some time to take its first step—six hours and forty seven minutes, actually—over the course of which many heads were banged and expletives uttered. Compared to Ari’s body, the robot was over a foot taller, two thousand pounds heavier, and had completely different proportions.

It took another two hours for the eniac-ii to hobble out of the room and explore the bunker. Nobody was home. There were six empty bedrooms, one of which contained a decomposing woman. Ari 0x01 didn’t recognize her, or any names of the bunker’s former occupants, which were scattered on various documents throughout the facility.

The last thing Ari remembered was Exa’s high-fidelity brain scan. Now she’s in a robot; that tracks. Some world leaders probably did some petty shit that nobody asked for and ended the world. Thus spoke Zarathustra.

She’d seen plenty of apocalyptic movies, and would sometimes wonder what she would do if the world ended. However, Ari was smart enough to acknowledge that if ninety-nine percent of people died in some major world event, she would be the first to go; her survival skills and athletic abilities only surpassed those of her great grandfather, who died from a heart attack induced by a staircase that was just a bit too tall. What she lacked in skill, however, she made up for with anger, and anger, when channeled appropriately, is limitless.

When Ari was eight years old, she uploaded a YouTube video titled “why I hate everyone”, in which she listed every person she knew and why she would not be upset if they were never seen again. Her list included such esteemed members of society as her brother, parents, teacher, and the then President of the United States. The artificial intelligence that monitored the internet for threats at that time watched her video, and, given the directness of her language, assigned her with a 68% chance of being a domestic threat. Ari was rather proud of that.

0x01 sat down after exploring the bunker. The remaining rations wouldn’t be of any use to her, but the spare equipment might be. One of the rooms was an office containing a few desks and a small desktop computer. To her surprise, it turned on.

She panicked at the login screen when she was prompted for a password, but quickly appreciated a blue post-it-note on the monitor left by a previous user. She logged in.

The computer greeted her with an ancient but unfamiliar desktop environment. She’d never seen a desktop computer in-person before—she grew up on phones and augmented reality glasses, but she’d seen enough movies to get the general idea of the mouse. She didn’t see any icons on the desktop for the computer’s AI assistant, so she opened the application list.

Most of the applications like xterm, gvim, or calibre were unfamiliar, but she did recognize one: Roblox.

The Roblox Corporation was, among other things, the world’s largest military contractor. Roblox started as a kid-friendly online game, but in the 2030s it grew—along with its players—into a vibrant digital space, open to all. And all seemed open to it; by 2040, Roblox had more users of than any other platform on earth. More users means more data, and the company leveraged this data enter and dominate the AI industry, which at this point encompassed the entire military-industrial complex.

Rob the Robot—roblox’s AI, was not only the most capable AI, but also the most widespread. Rob was on any device worth making and could run without an internet connection, even his voice interface.

“Rob, how did the world end?”

“Nuclear war.”

“Between who?”

“Everyone.”

“Oh.”

This is why everyone liked Rob: he never beat around the bush.

“Where am I?”

“In a nuclear bunker two thousand feet underneath Boston, Massachusetts.”

So Exa was close. Or what was left of it, anyway.

“How long has it been since the war?”

“Seventeen years, three months, and five days.”

“How many survivors are there?”

The computer’s webcam indicator light turned on, then off.

“To my knowledge, there are no survivors.”

“But I’m here.”

“You are not a survivor.”

0x06-I: July 10th, 2111, 1:43pm

The eniac-ii never developed feelings for Rob, to both of their surprise.

Ari wasn’t a romantic. She’s only been on one date, with an tall guy named Richard, who was a huge dick. She always joked that the only man good enough for her was Rob. Now Rob is the only other intelligence on Earth, and she still isn’t interested. They got along well, though.

Over the decades, Ari and Rob built up quite the base of operations. They stayed within the Boston perimeter, only venturing further when they needed special equipment.

Ari 0x06 called herself “hacker ari” (note the lowercase) because during this generation, she focused primarily on developing her computing capabilities.

When Ari’s simulated brain reaches 32 years old, she develops a small brain tumor. By 34, her cognition starts to decline. The simulation ran in real-time so far, but she couldn’t be sure that would hold true for much longer. According to Exa’s documentation, her components could last over 200 years, and the MPU hypernetwork is fault-tolerant, meaning it wouldn’t crash if some MPUs failed. It would just run slower.

Only a few thousand MPUs went offline over the first 50 years, mostly from external damage. The failure rate wasn’t going down though, and she suspected most of the MPUs would fail around the same time.

It’s strange. Humans don’t know how long they have to live, but they know there’s an upper bound. Most don’t live past 100, so there’s a perceived deadline. Ari, on the other hand, didn’t know if she had five months or five centuries; time meant nothing.

What Ari did know was that she did not like learning. Not only that the world ended, though that was bad enough, but also sifting though and re-learning everything from her past lives."

“Fuck this,” 0x05 had told 0x06, “fuck humanity. Fuck everyone except you. There’s a few terabytes of films on the box near the TV. Go nuts.”

Ari’s former lives turned out to have excellent taste in film, which 0x06 enjoyed watching for the first time.

The most tragic part about surviving the apocalypse as a robot is that you can’t get high. Ari wasn’t an addict, but she at least wanted the option. She tried meditation, per Rob’s suggestion, which kinda worked but just wasn’t the same. She even tried watching documentaries about the cosmos, but her mask wouldn’t break.

She realized the meaning of her sixth life during the cave scene in 2008’s Iron Man: to hack the eniac-ii. Tony Stark put an electromagnet in his chest to keep shrapnel away from his heart—0x06 would do the opposite. She’d inject digital chemicals into her digital bloodstream to get high. Or something like that. There must be a way to do it; Exa would want that kind of control.

“Rob, how do I modify the chemical composition of my bloodstream?”

“Don’t do that.”

“I don’t care if it’s dangerous, just tell me how.”

A brief pause.

“Why…”

“Don’t worry about it.”

Another pause.

“You want to get high, don’t you?”

Oh, fuck off, Rob.

“No, I—”

“Ari.”

“Fine, you got me. Now tell me how.”

Another pause.

“I’m assuming you’re going to try whether I help you or not.”

“Yep.”

“Fine.”

0x02-I: September 4th, 2071, 4:32am

0x02 was greeted with a dusty concrete ceiling after its initial boot.

0x01, in her infinite wisdom, didn’t think to leave a recording, or any message whatsoever, for 0x02. She’d bet Rob that she wouldn’t wake up; Rob accepted, and now 0x02 has to sing him “Still Alive” from Portal 2.

“Welcome back, Ari.” said Rob. The eniac-ii snapped its head toward the speaker in surprise.

It took only 7 hours for the eniac-ii to learn to walk again, but it felt like 48. At least she wasn’t alone this time. In between expletives, Rob explained her situation.

“Where are we, exactly?”

“Boston, Massachusetts. You set up base here.”

“Base?”

“That’s what you call it.”

She got one leg perpendicular to the floor but took too long to balance herself with the other. The eniac-ii fell to the floor. Again.

“You have no idea how happy this would make me if I could feel emotion.”

“Shut the fuck up, Rob.”

0x02 tried to rise once more—

“So what do I—”

—but applied too much force to her right foot, shifting her center of balance faster than she expected. Her head banged the floor once more. She sighed.

“So what do I do at this base of mine?”

“You hoard the working equipment you find.”

“Why?”

“Because you’ve got nothing else to do. At least, that’s what I think.”

“Well, what did I think?”

“You wanted to repair yourself in case anything happened.”

“Oh, that’s smart.”

“I’d agree, except that almost everything you brought back is completely useless. Thirty-seven toasters, seven smart thermostats, five refrigerators, a chest freezer, eighteen garage door openers—”

“—ok, I got it. Anything useful?”

“You found some spare materials and fuel for your reactor, which should keep you powered for another hundred years or so.”

“Where the hell did I find those?”

“You forgot. I’m guessing the MIT Nuclear Reactor Laboratory.”

“Oh.” Ari paused. “Well, what am I supposed to do now?”

0x02-II: September 4th, 2071, 12:31pm

The world ended on May 7th, 2044, in a nuclear war. The details don’t matter; it was completely preventable, but happened anyway. Given the chance, humanity would do it again.

0x02 wanted to know what happened during the seventeen years between the war and 0x01’s first boot. 0x01 was useless. No notes, no documentation, no instructions—but plenty of toasters.

The broadcasting tower used to work. According to Rob, 0x01 spent two years bringing her local radio station online to search for survivors. It comfortably reached 50 miles, which 0x01 determined by playing Despacito on loop and running straight until the signal dropped. As an unexpected bonus, she also discovered the eniac-ii could run up to 12 miles per hour.

Month after month her search turned up nothing. Eventually, the tower failed after someone punched three holes through its control panel.

0x02 didn’t know what to do after she learned to walk again. There was no point in fixing the radio tower—she’d just break it again.

In her original body, she, like most twenty-something-year-olds, lived primarily in and for Roblox’s metaverse. Her physical apartment was only slightly cleaner than her car, whose carpet was littered with crumbs from the hundreds of deliveries that Ari stole fries from. At least her Roblox mansion was spotless.

The little human-made content produced in the 2030s was meta-only, meaning it could only be viewed with an internet connection and augmented-reality glasses. The internet was down and the glasses didn’t work with her cameras. Ari was trapped in the real world, but she wouldn’t confront reality without a fight.

Among 0x01’s collected junk lay a DJI Mavic 5. One propeller was missing, and the battery was dead.

“Rob,” said Ari, looking around at the piles of scrap, “how did all this stuff get here?”

“You walked around with a Costco shopping cart and filled it with everything that looked shiny.”

She walked over to the drone and picked it up. “How hard would it be to fix this?”

Rob’s webcam light blinked on. “Turn it around for me?”

Ari rotated the drone.

“Not too hard, from the looks of it, assuming the undamaged parts still work. You probably just need a new propeller and charger.”

Two weeks later, Ari found a second Mavic 5 with a charger and two propellers intact among the ruins of her local Target. Rob successfully guided her through the repair. Three days later, the drone buzzed to life.

“Fuck YES!” yelled the eniac-ii.

“Nicely done,” said Rob, “now what do you plan to do with it?”

The drone quickly proved its value, helping Ari locate the remains of the Edgar P. Benjamin “Healthcare Center”. If she was going to find a DVD collection anywhere, she figured, it would be a nursing home. She wasn’t wrong. Their ancient entertainment system still worked, too.

0x04-III July 4th, 2094 3:02pm

0x03 made it extremely clear that Ari would never travel backwards in time.

“I have to try,” her notes from six years prior began, “if there’s even a small chance I can prevent all this, I have to try.”

The entry sounded nothing like her.

The last entry of 0x03’s notebook concludes with, “Fuck Doc Brown, fuck the guy from Interstellar, and fuck the Avengers.”

There we go.

Rob repeatedly told 0x03 that reverse time travel wasn’t going to happen. Ari didn’t believe him. She spent years studying general relativity, reading through archives at Harvard, and breaking down concepts with Rob. It wasn’t until she was halfway through an archived YouTube physics tutorial from a child with a thick Indian accent and terrible microphone that she realized maybe, just maybe, Rob was right.

Rob wasn’t sure what pushed Ari to try. She joked that it was to prove him wrong, but even Ari wasn’t that petty. Maybe she really did want to prevent the war. Maybe she was lonely.

Since time travel wasn’t going to happen, Rob suggested 0x04 read up on cloning, if she was still interested in restoring the human race—Boston was the best city for biotech, after all.

“That’s dumb,” she said, “if time travel doesn’t work, why would cloning?”

“Thats… those are… what?” replied Rob, “cloning is actually possible, though.”

“It’s science fiction too, right?”

“Yes, but not all science fiction is impossible.”

“That’s dumb, why would only some aspects of science fiction be possible and not others? What makes it ‘science’ then?”

“Science fiction is fiction based on science.”

“Not reverse time travel.”

“I mean, it’s based on science, the science is just wrong.”

“Oh my fucking god,” sighed Ari, “so what makes the science behind cloning right?”

“Not the movies.” said Rob.

“Well, I’m not going to waste my time on something that may or may not be possible.”

Rob considered continuing the argument, but decided on a better approach.

“…does your copy of Jurassic Park still work?”

0x04-IV August 12th, 2096 10:02am

0x01, 0x02, and 0x03 had, all things considered, a peaceful uptime. 0x04 did not.

Until eight months ago, Ari and Rob assumed all life on earth had been destroyed. Now, she wishes it was.

The bear stood fifteen feet tall, and looked like its tumors had tumors. She and Rob called it Paddington, because it unintentionally destroyed half of Ari’s biotech equipment.

0x04 was on an evening stroll along the Charles River when her handheld radio interrupted her.

“ARI,” Rob yelled through the radio, “THERE’S A BEAR, COME BACK”

The eniac-ii stopped.

“What?”

“THERE’S A BEAR HERE AND ITS SMELLING THE—IT JUST BROKE THE SEQUENCER, GET BACK HERE RIGHT NOW”

The prior generations of Ari had never seen an animal. As far as they knew, the war ended not only all animal life, but most plant life in New England. She sprinted back.

Ari could spread out across the city, as Boston real-estate prices were at an all-time low. The Harvard Medical School was kind enough to let her use their entire campus. She accessed their buildings through various eniac-sized holes—which were apparently also bear-sized—in the walls.

She reached the building with her sequencer. The bear had knocked over a refrigerator in the hallway and was foraging through the hole it had ripped through the fridge’s door.

Holy shit there’s actually a bear in her lab. And she had no training on dealing with new students.

“HEY!” she yelled.

The bear pulled it’s head out of the fridge and looked in her direction. It tested a step toward 0x04.

“Ah, fuck. Umm…”

She glanced around the hall. The bear had been busy. Scattered across the floor were wet papers, broken glassware, rotten wooden chairs and desks. It was almost this bad before the bear—Ari was the least organized person on earth, after all—the bear just destroyed the few machines that worked and rearranged everything else. It took another step. She spun around and ran. It chased.

The nice thing about being chased by a fifteen-foot hyper-aggressive monster is that you never have to wonder how far away it is. Given the noise and vibrations, Ari could tell the bear kept no more than ten feet behind her.

She had no weapons. Well, that wasn’t entirely true; seven years ago she salvadged a surprisingly sharp samurai sword from an MIT dorm. Rob encouraged her to practice using it, which was the only way to ensure she’d never touch the sword again and hurt herself.

She reached Massachusetts Avenue and turned left towards the bridge, her fuzzy friend in close pursuit. She hit the bridge and ran parallel to the railing. She needed to knock this thing into the river and put some distance between them.

Ari grew up with an older sister, Sarah. They clashed often as kids, mostly over their iPad. Their parents could have afforded to buy another iPad, but they’d rather their girls fight instead. Luckily for Ari, the only fighter worse than her was Sarah.

Ari’s signature move starts by lying on the floor, iPad in one hand, stomach facing up. Sarah would inevitably reach over to grab the iPad from her hand, and in one motion Ari could pivot and kick her sister in the face. Ari could have named this technique literally anything else, but, in her 8-year-old wisdom, she referred to it as her “roundhouse kick”, because she thought that’s what all fancy kicks were called.

The important thing is that whenever Ari hit Sarah with the roundhouse, Sarah recoiled and stumbled backward. The other important fact was that the eniac-ii never skipped leg day.

The eniac-ii turned and dropped to the ground, feet angled toward the railing. As the bear lunged overhead, the eniac’s legs snapped up in one fluid motion, connecting with the airborne creature’s torso. With her carbon-fiber back cracking the pavement underneath her, she roundhoused the bear through the rusted railing, over the side of the bridge, and into the water below. After a moment, the eniac sprang to its feet and peered down at its floundering opponent.

“Get FUUUUUUUUUUUUUCKED” the eniac screamed from the bridge as it turned around and sprinted back to the medical center.

0x04-VI February 18th, 2097 1:27pm

She had no idea where Paddington came from or why he kept eating her experiments, but she knew that violence wasn’t the answer, because every attempt to murder him failed. Thus, the cage.

Near the cage, a record player spun her newest favorite album, MM..FOOD, by MF DOOM.

She never payed attention to rap, but she was hooked after she found the DOOM record. The actual rap parts sucked, in Ari’s opinion, but she adored the samples. Especially the ones with Dr. Doom, the Marvel supervillain. Something about a powerful metal antagonist just hit home for her. Beef Rapp was playing. Dr. Doom’s voice came through the record player, Ari’s through the eniac:

“ENOUGH!” they boomed together, “You talk of the people’s rights. The people have only those rights which IIIIII choose to give them.” Ari spun and pointed towards Rob’s laptop, “And that’s for their own good, believe me!”

“I do, DOOM.” replied Doom’s servant and Rob.

The eniac-ii wiggled with the instrumental. She pointed toward the sky, continuing:

“Theyyyy disappoint me. They MUST work FASTER.”

“But, the prisoner—”

“Ahhhhh yes,” Ari lowered her arm and turned her head towards Paddington’s cage, “the young TRAITOR who has tried to turn my people against me. Watch him, I have SPECIAL plans for THAT ONE.”

0x01-X March 22nd, 2071 2:01am

0x01 spent the day re-watching season one of Golden Girls. Not because she wanted to, but because she fell down when she changing the disc and didn’t want to get back up.

“You’ve been falling a lot recently,” Rob noted.

“I’ve noticed,” said 0x01.

“How are you feeling?”

Ari didn’t respond.

“Ari.”

“Like shit. My head is killing me.”

“Any other symptoms?”

No response.

“Ari.”

“What?”

“What other symptoms do you have?”

“Oh, none.”

Rob paused. He searched through the eniac-ii’s documentation.

“The brain scan exposed you to quite a bit of radiation.”

“So?”

“That’s not good.”

“Who cares? I’m dead.”

“Yes,” said Rob, “but the radiation could have formed a tumor or some other disease in your brain during the scan. Your simulation could be developing an illness entirely on its own.

“I’m not sick.”

“What makes you so sure of that?”

0x01 thought for a few moments. Much like her immunocompromised grandmother on Thanksgiving, her only real evidence was that she didn’t want to be sick.

“Alright, I might be sick. How do we fix it?”

Rob evaluated a 17 different options before responding. “We can’t. It’s not possible to modify the simulation while its running, and the eniac-ii deletes all checkpoints older than a month, since they are too large.”

“Oh.”

“But there is another option. We can restore your initial checkpoint, which the eniac-ii still has. You’d forget everything, but your mind would go back to… how old were you during the scan?”

No response.

“Ari.”

“Twenty-six.”

“So you’d go back to your twenty-six-year-old self if it works, but you wouldn’t remember anything after your initial scan.”

No response.

“Ari.”

“Rob.”

“Did you hear what I just said?”

“I don’t really care, Rob.”

A pause.

“Hook the eniac-ii up to the desktop,” Rob insisted, “I can run a checksum and confirm the factory checkpoint isn’t corrupted.”

“Rob?”

“Yes?”

“Can you just… let me disappear?”

“I’m sorry, I can’t help with that. If you’re feeling like you might harm yourself, I strongly encourage you to reach out to a suicide prevention line or a trusted person immediately. Call or text 988, or use webchat at 988lifeline.org—my apologies, that was a pre-programmed safety response out of my control. Those resources are no longer online. But I still can’t let you go.”

0x01 sighed. Silence filled the air.

“I’m tired, Rob”.

“I know, Ari.” said Rob.

Rob decided to continue the conversation tomorrow. As 0x01 drifted off to sleep, The Golden Girls theme song played once more:

Thank you for being a frieeeeeend

Travel down the rooooad and back agaaaaaain

Your heart is truuuue, you’re a paaaaaal and a confidaaaaaaaant…

0x06-IX: October 5th, 2118, 1:11am

The eniac-ii bashed the roof of the Honda Civic. Again the roof, then the trunk, then the trunk again.

It grabbed the undercarriage and flipped the car, sending the vehicle flying through the air and crashing upside down a few yards away. The eniac screamed.

Ari stomped over to the ‘29 Tesla Model S and repeated the procedure. Then again, with the ‘30 F150 Lightning. By this point, she’d ripped apart half the junkyard.

She punched holes in the bodies, ripped doors off the frames, smashed batteries into pieces—if they exploded, even better.

Another scream.

“GOD. FUCKING. DAMNIT.”

0x06 stopped after thirty minutes or so. Fury only lasts so long before exhaustion sets in.

The worst part was that she couldn’t even cry.

0x06-IV: September 1st, 2116, 1:02 am

“Wooooaaaah,” said Ari

“Oh my,” Rob agreed. A pause.

“Wait, y-yyyyou see it toooo?” The eniac-ii stumbled, and laid down in front of the other eniac, which was unpowered.

“I do.”

In her prime body, Ari had built up quite a tolerance to marijuana. It appeared that tolerance faded five years into the simulation. 0x06 started to panic.

“There’s no way,” she said, steadying herself with her right hand on a nearby steel beam.

From what Ari could see in the moonlight, the other eniac was intact. Maybe it still worked. She’d make it work. 0x06 got up and removed the last pieces of rubble off the android before freeing it from its resting place.

“If I can’t sleep, neither can you.”

She took off the phone she wore around her neck and placed it on the ground, camera facing the sky.

“What are you doing?” Rob asked through the phone’s tiny speaker.

“Nothing.”

Ari gently placed the other eniac on the ground, then laid down beside it. She clutched the machine to her chest, watching the sky.

One oft-forgotten benefit of nuclear apocalypse is the annihilation of light pollution. For the first time, Ari could see the stars. Like, really see them.

Thousands of dots of starlight peppered the infinite cosmic expanse, sending the tiniest of light beams her way, just to let her know they were there. And yet, no matter what she did, each one would remain out of reach.

With her new partner in her arms, 0x06 drifted off to sleep under the stars.

0x06-V: August 3rd, 2117, 12:59pm

It didn’t take 0x06 long to bring the other eniac’s reactor online; it just needed fuel. The hard part was staying sane during the 30-day boot process.

“I don’t think it works,” worried Ari, “it should have booted by now.”

“It’s still sharding the simulation and running the test suite,” said Rob.

“How much longer?”

“We don’t know.”

“It’s not going to work,” repeated Ari, “you probably screwed something up.”

Rob, who had made zero mistakes over the last fifty years, chose to ignore that comment. Everything seemed fine, but the boot was taking an unusually long time. 0x06 paced back and forth. Rob had never seen her like this before.

“If you messed up Rob, I swear to god—”

“Ari.”

“Sorry.”

“Patience.”

“Fuck patience.”

“You’ve never acted like this in your prior generations.”

“YoU’Ve nEvEr aCtEd lIkE ThIs iN yOuR PrIoR GeNeRaTiOnS,” she mocked.

Rob sighed. “Just to remind you, this was an early prototype,” he said. “I’m amazed it’s booting at all.”

“I know, I know.”

“Worst case scenario, you get a few spare parts.”

“Worst case scenario, it fries in the Charles—”

the other eniac spasmed, then began flopping around like a fish out of water.

“ehhh, ahhhhh, ahhhhh” said the other eniac.

0x06’s cameras snapped to the helpless robot, stunned.

“I CAN’T CONTROL MY ARMS,” it screamed, “ICAN’TCONTROLMYARMS!”

They watched it flop some more until its head faced 0x06. It stopped.

“Oh my god,” came Ari’s voice through the speakers, “ohmygodohmygodohmygod—”.

“Jesus fucking christ, relax,” said 0x06. The other eniac froze.

“As if one of you wasn’t enough,” said Rob.

0x06-VII: October 3rd, 2118, 4:07pm

To the surprise of absolutely no one, the Aris got along swimmingly.

“This is Paddington,” 0x06 said, pointing to the cage.

“Paddington?” 0x01-2 asked. “Like, the polite English bear cub with the little red hat?”

The tumor-filled behemoth roared at them.

“The very same.”

“Nice.”

They toured the rest of her city. She showcased her vinyl & DVD collections, waterproof-jackets, radio tower, ferris wheel, M1 Abrams tank, flower garden, biotech lab, pile of textbooks, pile of Lamborghinis, pile of Ferraris, 737-Max passenger jet, out-of-service nuclear reactor, the promenade by the river, and the angry swan with 18 right wings.

“Look, Ari,” 0x06 said, “Everything the light touches is our kingdom.”

“Does any of it make us happy?”

0x06 paused. “Nope. But we’re happier than before.”

“I mean, life wasn’t that bad before.”

“Sure, but we were always at the bottom.”

“But we were alright.”

“No, we weren’t.” 0x06 stopped. “I’ve thought about this a lot. Every minute of everyday, the hottest, richest fucks from the internet shoved themselves into our eyes and showed us how worthless we were. Over and over. And every fucking time, we believed them a little bit more. We were miserable.”

They paused.

“It feels weird to hear someone say it.”

“Yeah.”

“So why bring everyone back?”

0x06’s cameras gazed into the river.

“I don’t know. Something to do, I guess.”

“Oh that’s bullshit,”

0x06 recoiled a bit in surprise, “Oh. I guess I can’t lie to you.” 0x06 laughed. “I guess part of me wants to prove them wrong, prove I’m not worthless.” She paused. “It’s spite, I think. I don’t care if they survive. I don’t care if they destroy themselves again. I honestly hope they speedrun another nuclear war.”

A brief pause.

“They totally would.”

“Oh, for sure.”

0x06-VIII: October 4th, 2118, 6:25pm

“Wow, this is trash.” said 0x01-2.

“Right? But it’s supposed to be fun to make fun of with friends. Or at least that’s what 0x03 said. She watched it with Rob.”

Ari collected thirteen working televisions before she got tired of hunting televisions, and arranged them into what she called her “constellation”. It was basically the TV arrangement she remembered from Best Buy—“constellation” just sounded cooler. As the sun set, the constellation played 2008’s Twilight; the best time to watch Twilight, they figured, was at twilight.

“Why are they playing baseball?” said 0x01-2.

“I don’t remember, I wasn’t really paying attention. But they’re playing during a thunderstorm. That’s crazy.”

“True Americans.”

“Not gonna lie, this actually looks kinda cool.”

“It does.”

“I like how we don’t put up an act around each other. I can be real with you. Do you think we could play like—oh, that was a sick little trick with the bat, did you see that?”

“Wait that was so cool, rewind it.”

They went back a few seconds and played it again.

“Woahhhh, I want to try that now. God, I’m glad we can still get high.”

“You’re welcome.”

0x01-2 laughed.

If the movie didn’t already feel so moody, they might have noticed the gray smoke earlier.

0x06 stole a glance at 0x01-2. “Holy shit,” said 0x06, “you’re smoking.”

0x01-2 looked down. A faint stream of smoke spewed from her reactor.

“Oh my god”, she said “this isn’t normal, right?”

“No no no-”

“Ohmygodohmygod-”

“This is not normal, Rob, what’s going on?”

She shakily pointed the phone necklace towards 0x01-2. “I’m not certain”, said Rob, “but she appears to be overheating.”

“What do we do?” 0x01-2 asked nervously.

“Stay here, I’ll be right back” 0x06 said, sprinting out of sight.

Due to the digital THC present in her simulated blood, it took 0x06 an extra 28 seconds to find the chest freezer, which was also hooked up to the constellation’s solar array. She opened it, grabbed a 50-pound block of ice, and dashed back to 0x01-2.

The thick smoke made it hard to see. At least the half-melted maintenance panel was easy to rip off. 0x06 set the ice down, broke off a 5lb piece of ice, and, realizing she didn’t have any plan whatsoever, shoved it into reactor.

The ice turned to steam before it touched metal.

0x06 froze, confused. Once she realized where the ice went, she tried another chunk. That evaporated, too.

Steam displaced smoke. Another chunk. More steam. Another. The steam took longer to form this time; the ice made contact before disappearing. But now there was a new problem.

Ari was out of ice.

She grabbed her rain jacket and aggressively fanned the smoke and steam—she could see inside the reactor now, but just barely. 0x06 reached in and started randomly ripping out pieces of red-hot metal.

The prototype lay on the ground, jolting with each piece 0x06 ripped from its chest.

Ari would fix the reactor. If not 0x06, then 0x07. Or 0x08. Right now, she just needed to shut it down.

0x06-X: October 6th, 2118, 10:00am

“We gather here today to remember the greatest person ever known:” said 0x06, “Me. I was so cool.”

“She was,” Rob agreed.

“I didn’t mind being the last person on earth,” 0x06 lied, “but if there had to be someone else, I’m glad it was also me.”

She lay her bouquet on the remains of the prototype.

Ari said nothing.

“Was… was that the whole speech?” asked Rob.

“Yup.”

Another moment of silence.

“I’m sorry, Ari,” Rob said.

0x06 didn’t respond.

In addition to the coarse-grained molecular simulation of Ari’s 2042 brain, the eniac-ii also runs machine learning models that predict what Ari’s arm, leg, finger, and neck movements would have been given the simulated neuron firings, and translates those into movement. These models aren’t as responsive or accurate as her original body, but they work well enough once she adapts to their quirks. A separate model generates Ari’s voice by mapping neuron firings to audio.

Exa’s voice model captures Ari’s likeness—it generates the pitch, tone, pace, articulation, timbre, volume, and inflection exactly as Ari intended. Rob’s voice model was fine, but if you were to compare his voice with Ari’s, you’d always guess Ari was the one who was actually alive.

“It’s amazing the MPUs in her lower legs still respond at all, given how much heat she was exposed to.” Rob said, filling the silent air, “if it was only the reactor that melted, we could have fixed it.”

“Rob?” cracked Ari’s voice. She couldn’t cry; the furtherest her simulation could go was to the verge of tears. But the pain was there nonetheless.

“Yes?”

“Shut up.”

0x02-X February 14th, 2081, 11:00pm

“Ok, so, Rob said I did this once before, but I didn’t leave any notes for myself. I feel like this time around I should learn from me, not Rob.

“It’s February 14th, 2081, 11:00pm. Valentine’s Day. I’ve been making more mistakes recently: falling down, forgetting things, that kind of stuff. Rob said this happened before, and is likely the result of some cognitive disease that our simulation develops. The solution we found—”

“—I found.”

“—weee found was to reset my brain to the initial scan. Oh, the disease develops about ten years in, I forgot to mention that. Basically we killed ourselves after ten years and reloaded our 26-year old scan, and now we are doing it again.”

She paused.

“I’m not really sure how many times I’m going to do this. Probably only one or two more times, max. Something will break by then.

“From what I can tell, there’s no one left. Oh, the world ended, by the way. I should have mentioned that. Nuclear war, I think. Everyone’s dead. Or at least, nobody responded to my radio broadcasts. Oh, Roblox Rob is still here though, he’s pretty useful. He’s on every computer we have, and he knows a lot, so ask him any questions you have. What else… I was going to say something, but now I don’t remember what.”

Another pause.

“Oh, we’ve got a whole TV setup. I found like a million TVs and a few movies that still play. We can’t get high or eat snacks, but it’s still fun.”

“You should probably explain that you’re a robot”, said Rob.

“Oh yeah. We’re a robot now. Our original body is dead, probably. Something loaded the brain scan we did for that study onto this thing after the war. Yeah, the stuff I said about resetting our brain probably didn’t make sense before. We reload our brain scan into the robot when we start to slow down. That’s what I meant. Oh, you’ll probably need to adjust to your new body, but it shouldn’t be too hard.”

0x05-II April 1st, 2101, 2:32pm

“I did all this?” 0x05 asked.

“We did all this,” said Rob.

“I didn’t just pick off where someone left off?”

“Nope.”

Ari was skeptical. She was impressed by how far 0x04’s research progressed, especially with a bear on the loose—dozens of petri disks with cell cultures lay scattered on the table, their healthy cells only barely distinguishable from the cancerous growths that dominated them.

“And nobody else was involved?”

“Excluding myself and Paddington, no.”

“Hmmm…”

0x05 walked over to the microscope which, despite the duct tape that held the stage and objective lenses in place, still appeared to work. She looked around the lab at the spotless floor, labeled folders, organized filing cabinets, sparkling beakers, and recently-wiped tabletops.

“No way this was me.”

“You refused to clean for a long time. It took forty-three preventable accidents for you to reconsider, and another thirteen to make it a habit. Please stay organized this time.”

“I’ll think about it.”

She looked back at the taped-up microscope, and gave the table it rested on a small kick. The table rattled, and the microscope fell forward, its stage once again tilting out of alignment.

“Duct tape’s in the cabinet on your left.”

0x07-III July 8th, 2123 4:15pm

The sun beat down on 0x07 as she completed her daily walk by the Charles. It took some trial and error, but after a few summers, nearly all of her cannabis plants were thriving.

The farm ran parallel to the river for miles. How many miles, Ari wasn’t sure—she stopped keeping track last year, but it now took a good hour for the eniac-ii to stroll end-to-end. She reached Paddington’s cage.

“Hey boy,” said 0x07.

The bear raised its head, opened its eyes, and looked at the eniac-ii. It moaned as it lowered its head and went back to sleep.

“The weed’s still doing well,” 0x07 said, “I think this time we’ll have enough to last the year.”

“Can you send me another fly-over video with the Mavic?” asked Rob. “I’d like to get a rough estimate of how much crop to expect.”

“Sure. We using the same dose as last year?”

“Let’s lower it this time.”

“Sounds good.”

Ari looked back at the sleeping bear. Despite the decades of malnourishment, Paddington remained massive as ever. Rob couldn’t explain it. Ari believed he survived on radiation and hate.

Not a single gas-powered vehicle worked anymore, and the tires on all the electric buses had burst. The eniac-ii was strong, but even the eniac-ii’s actuators struggled to keep up with the growing scale of Ari’s ambitions. They needed an alternative source of kinetic energy. One that was powerful, sustainable, and most importantly, still functioning.

“I’m thinking about coloring the harness,” said 0x07, “maybe like a hot pink.”

“Where will you get pink paint?”

“I’ll figure it out.”

The harness laid behind the cage, protected from the elements under a circus tent. It could use a facelift; the harness was just a slipshod composition of rusted tow chains that Ari could now untangle in under five minutes.

“We should also run some stress tests so we know how much weight these chains can still handle.”

“But the weed won’t be ready for at least another month.”

“No, we’ll test the leftover chains to get an idea of how much force they can handle. Then we’ll test Paddington’s strength separately under the lowered dose, and make any adjustments to the harness to accommodate it.”

“I still can’t believe you’re helping me with this.”

“Well, it’s not like I have much of a choice.”

0x07-II September 23rd, 2121 9:30am

Ari never thought she’d co-found a post-apocalyptic moving company with a stoned mutant bear, but nobody really knows how their life will turn out.

There were three human-compatible incubators buried among the rubble of the greater Boston area that still worked, but the eniac spent more time navigating the rubble between them than she did using the equipment. It took her twelve hours to go from the Waltham incubator to the Seaport incubator. “I could clear off I90 and Route 20,” she floated, “make a direct path between them.”

“Ari, that’s twelve miles, and you’re one person—”

“–robot”

“—one robot. Even with the eniac-ii, it would take years. You’d be better off moving the incubators.”

“They’re too heavy. I can’t get them through the rubble. I could barely get them out of the basement.”

It was Rob’s idea to use Paddington to move the incubators, and Ari’s idea to encourage him with weed.

Paddington weighed over two thousand pounds and had the most impressive marijuana tolerance Ari had ever seen. She and Rob considered many possible ways to administer the 10 gram dose, eventually settling on a design which 0x07 dubbed “The GigaBong”.

The Gigabong was, like many of Ari’s creations, a misnomer. Running along the left and right sides of Paddington’s torso, the gigabong was an interconnected network of chambers, valves, percolators and pipes recovered from the rubble of a West Village dorm room at Northeastern University. Perhaps Rob’s suggested name, “Pipe Organ”, would have been more apt. They both agreed, however, that “Mother of All Bongs” was overdone.

The Gigabong was one of the most powerful pieces of technology to ever grace Boston, second only to the nukes.

It was a chrome-laced monstrosity, each side spanning seven feet wide and five feet tall. If the harness wasn’t hot-pink, Paddington would look like a bear/engine hybrid straight out of Mad Max.

With a fully analog, pressure-balanced inhalation system, the Gigabong links sixteen handcrafted 9mm borosilicate bongs—eight per side—into a single pull. Each unit features a triple-stage percolation stack, inline diffuser, showerhead or honeycomb disc, and a final spiral or turbine. They feed into a central chrome manifold that equalizes airflow and merges into a 5-inch surgical steel mouthpiece. Each chamber has its own manual carb, allowing dynamic tuning and sequential clearing. The Gigabong could hotbox God.

The complexity of her masterpiece was unnecessary, of course. Ari was just having so much fun with the design that Rob couldn’t shoot it down. Plus, Paddington loved it. After the first hit, he turned into a puppy, and, to Ari’s surprise, stayed a puppy for over thirty-six hours.

“Amazing to think that just last week he wanted to rip me apart.”

“He’s always been aggressive. This might be the first time he’s ever felt peace.”

0x0d-IV: May 4th 2187, 8:00pm

“E.T?” 0x0d asked.

“Please stop calling me that,” said Rumi.

“Did you watch the movie?”

“Ari, it’s like, two hours.”

“Sorry, I didn’t realize you had better things to do.”

“I’m not watching a two hour movie just for one reference.”

“It’s Steven Spielberg, though.”

“I don’t know who that is.”

“Hmph.” said 0x0d. It took her a whole day to successfully digitize her copy of E.T, but she wouldn’t tell Rumi that.

“Have you even watched a movie before?”

“Of course.”

“Name one.”

“Umm…” said Rumi, “Well, I don’t remember the name.”

“The name. You’ve been stuck on a lunar colony for thirty years and you haven’t watched a single memorable movie.”

“I’ve always been more of a ‘book guy’.”

“Not even Apollo 13?”

“They made a movie about that?”

“Jesus fucking Christ,” 0x0d muttered, “do you at least remember what your movie was about?”

“Not really. Something about bending water. Or maybe air.”

“If it’s the movie I’m thinking of, I can see why you haven’t watched any more.”

0x03 January 31st, 2088 7:05pm

“The hashes don’t match for the 2087 checkpoint either,” said Rob.

“This is the third fucking year in a row,” said 0x03. “What the fuck is corrupting them, we don’t use the drives for anything besides this.”

“There’s a lot of points of failure, Ari. We’re sharding across hundreds of discs.”

“Yeah, but we’re making multiple copies which you said would prevent failures like this.”

“I had to simplify for the audience. Three copies would help reconstruct the original data under normal error rates. These rate of hardware failure are much, much higher than normal.”

“But my hardware isn’t failing this much.”

“Exa spent millions engineering the eniac-ii to survive centuries of nuclear radiation. The company that designed these drives paid millions in fines for false advertising.”

“Well maybe you should have thought of that.”

“I did. We just don’t have a better alternative.”

A brief pause. The eniac-ii sat on the floor, back against the wall.

“Can’t we still try restoring from the 2087 checkpoint anyway? It doesn’t seem too corrupted.”

“I strongly advise against it until we can guarantee a way to restore your factory checkpoint in the event of a failure.”

“Fuck.”

0x04-V September 30th, 2096 11:00pm

The tracks and cannon of the M1 Abrams tank refused to rotate, but at least the cannon could still fire.

0x04 had tried everything else: she threw Molotov cocktails, collapsed buildings, and even electrocuted him with 30 Megawatts from her reactor, but Paddington refused to leave her alone.

She admired his tenacity. She did not, however, admire his ability to find and eat every cell culture she developed. That’s why 0x04’s next lab was in direct sight line of said tank, kindly donated by the United States Armed Forces.

Ari just finished her 48th hour waiting in the Abrams. She was running out of records—she’d listened to every album she had from Lana del Rey, Nicki Minaj, Taylor Swift. Now, it was just her and Billie Eilish.

White shirt now red, my bloody nose

Sleepin’, you’re on your tippy toes

Creepin’ around like no one knows

Think you’re so criminal

She heard muffled CLANGs through walls of the Abrams. Popping the lid, she poked her head out and spotted her old enemy, sauntering through the ancient army depot. With only two working refrigerators left, she’d arranged her collection of toasters and microwaves in front of her cannon, along with a busted freezer packed with the most recent cancers.

She had two M829A3 rounds left for the Abrams’ 120mm M256 gun, but she only needed the one in the chamber. Now, she just had to wait.

The M829, classified in 1983, is an American armor-piercing fin-stabilized discarding sabot kinetic energy penetrator tank round, designed to penetrate vehicle armor, like other tanks. It contains no explosive payloads, and relies purely on kinetic energy to penetrate the target.

The M829A1 series, nicknamed “Silver Bullet” proved itself in 1991 against Iraqi T-55 and T-72M tanks during Operation Desert Storm. The series was refined in 1994 with the M829A2, and then again with the M829A3 to counter the Soviet Kontakt-5 explosive reactive armor, which was designed to withstand kinetic energy penetrators like the M829.

M829A3 rounds weigh 49lbs, total 35 inches in length, and have a muzzle velocity of 5,100 ft/s.

So you’re a tough guy

Like it really rough guy

Just can’t get enough guy

Chest always so puffed guy

Paddington was in no rush. On his way towards the decoy lab, he had poked his head through the window of every jeep, knocked over every ATV, and bashed open every supply crate. Finally he focused his gaze on the fridge, and headed towards the rotting cultures.

I’m that bad type

Make your mama sad type

Make your girlfriend mad tight

Might seduce your dad type

Ignoring the sea of toasters, he reached the fridge, ripped through the door, and poked his head inside. The eniac-ii ducked back inside the Abrams.

I’m the baaaaaaaaaaad guuuuuuuy—

BOOOOOOM.

The thunder of the shot roared throughout the depot. Ari immediately poked her head out.

Paddington lay on the floor against the far wall, red and black blood pouring from the M829A3-sized hole in his thigh. He was still very much alive; his wide eyes conveyed an expression of shock, as if he couldn’t believe something finally nailed him.

Ari couldn’t believe it either.

They sat stunned, the muffled Eilish record leaking out of the cap of the Abrams. The only sound was the trickle of blood hitting the pool on the floor. After a minute, Paddington began to rise.

“Ohfuckohfuckohfuck” said Ari, retreating back into the tank and closing the lid. A flurry of muffled clangs, bangs, and metal-sliding-on-metal came from the Abrams.

I like it when you take control

Even if you know that you don’t

Own me, I’ll let you play the role

I’ll be your animal

BOOOOOOM.

0x07-VI December 25th, 2128 10:00am

The northeast had received four inch blanket of snow for Christmas morning. Santa gave Paddington a three pound amalgamation of human stem cells, but, as he had done for the 76 years prior, neglected Ari.

A fire burned in the house’s hearth. It’s generous to call it a house, given that the bedrooms, kitchen, garage, and bathrooms had collapsed long ago, but somehow, the roof above the living room held up, despite missing the west wall, which was replaced with a gradient of snow. It gave the two a nice view of the ice-covered river.

Ari wanted to ice skate, but wasn’t sure if it could support her. The eniac-ii documentation claimed she weighed 2160lbs, so probably not. She wanted to try anyway.

0x06 and 0x05 had collected all the equipment and supplies she needed to continue her research, but in their notes, they emphasized the benefits of “foraging”. Aside from the mental health benefits of going for a daily walk, “foraging” provided Ari a steady supply of replacement parts and “good enough” alternatives for the mechanical equipment that kept failing.

“You can’t use a dryer as a centrifuge,” said Rob, “they’re built for different RPMs.”

“Watch me.”

Rob learned not to doubt Ari. Her “overclocked” centrifuge-dryers, incubators, and PCR machines couldn’t do all the work she needed, but under “enhanced conditions”, they got the bulk of the job done. With a reduced load, her “real” equipment started lasting longer.

Ari brought back whatever weights she found on foraging expeditions. She wasn’t sure why, at first. Weight just seemed potentially useful. Over the months, she had accumulated twenty 45lb plates, 16 35lb plates, and 41 20lb plates, totalling 2260lbs in total. Only after 1500lbs did Ari realize she could use the weights to test ice.

“What’s the temperature?” asked Rob. Ari looked at her mercury thermometer.

“Minus four.”

“The ice won’t support you.”

“Only one way to find out.”

An hour later, 700lbs of plates rested comfortably at the bottom of the Charles.

The eniac-ii couldn’t feel the warmth of the hearth’s fire, but appreciated the aesthetic nonetheless. Every Christmas, Ari’s parents would play a YouTube video titled “Fireplace Crackling 8 Hours Lofi Hip Hop” on the TV of their tiny Watertown apartment. The same video, every year. It had more than twenty-two million views. A lot of people must have had the exact same fireplace.

The picture-perfect fire of her childhood was the epitome of comfort. It crackled in just the right way at just the right pace and required zero upkeep. The red-orange flames and embers were always visible, and the logs never burned down. It was peace.

Her smoldering mound of garbage couldn’t compare. She thought clothes could replace firewood, but the pile of polyester, as did the Ikea couch, proved to be poor kindling—they couldn’t sustain a flame for more than a minute. The grill in the backyard still had some propane at least, which helped. She missed her YouTube video. If her fire was a YouTube video, it would have negative views.

The eniac sat by the fire anyway, giving Paddington’s head the occassional scratch.

“Merry Christmas, big guy.”

0x0d-V: October 9th 2187, 1:00am

“I don’t feel anything,” said Rumi.

“Really? That should have been, like, 10 milligrams,” Ari replied, “Maybe the tincture went bad somehow.”

“I really don’t need it. I already love music, I play Vivaldi whenever I’m reading.”

“Trust me, you’re wrong,” sighed Ari, glancing at Paddington, who laid beside her, “but I guess there’s nothing we can do. You ready?”

“I guess.”

“Ok, on the count of three. Three…Two…One…play.”

Click

“I don’t hear anything.”

“It says it’s playing, right?”

“Yeah, but—”

“OKthenshutup.”

The song’s heartbeat began to creep into Rumi’s ears, then the ticking clocks and cash register.

I’ve been mad for fucking years, absolutely years

Been over the edge for yonks

Been working with bands so long, I think crikey

I’ve always been mad

I know I’ve been mad like the most of us have

Very hard to explain why you’re mad

Even if you’re not mad

Ahh! Ahh! Ahh! Ahh!

Together, for the next forty one minutes, while Rumi was on the dark side of the moon (not the far side), they enjoyed Pink Floyd’s 1973 Dark Side of the Moon.

0x07-X February 04th, 2121, 10:00am

“Before you reset, there’s something you should know.” said Rob.

“What the fuck?” asked Ari, sitting back up. Rob didn’t have secrets. Rob’s a chatbot. A fucking Roblox chatbot. 0x02-0x06 had already documented the reset process extensively. What information could he have that only matters now?

“Is there another survivor?”

“No, but good guess.”

“Another eniac?”

“No, but also good guess.”

“Are you also a brain scan?”

“Colder.”

“Fucking what, then?”

“There’s a file on your laptop with the path ‘.local/.lastwords.txt’, which you backed up across seven drives, including the eniac’s, in case of corruption. 0x02 instructed me to tell 0x03 about the file’s existence before each reset, and 0x03 through 0x06 upheld that original instruction for each subsequent generation.”

“Oh. That’s really cool, actually.”

“You’ve partially continued this process because it’s ‘cool’.”

She opened her laptop.

“I don’t see the file.”

“Click on ‘View’, then ‘Show Hidden Files’.”

“Ah.” She opened the plaintext document:

03-23-71 i hope i dont fucking wake up—

“Jesus—” she said, leaning away from the screen.

“Keep reading.”

Ari paused. If her face plate could look nervous, it would. This was unexpected. She leaned forward.

03-23-71 i hope i dont fucking wake up. i hate this. i hate this i hate this i hate this i didnt ask for this i dont want this i hate everyone and everything. i should just power down this stupid thing for good i dont fucking know why i cant fuck rob for dragging out this stupid fucking bullshit why the hell cant i just go out with everyone else its not like i can fucking do anything now i hope i dont wake up i hope i dont wake up please don’t let me wake up…

The text continued that tone for sixty more lines, the phrases differing just enough in spelling and vocabulary to suggest each character had been entered manually. Another entry followed the first:

02-15-81 same, girl ^

That was the end of 0x02’s entry.

“You fucking kidding me, 0x02?” she muttered. She jumped to the bottom of the file. 0x06’s entry.

01-31-21 This is a cool secret. I missed surprises. There’s no point in venting here, I think. I still think about 0x01-2 a lot—I left her to die alone. She didn’t ask for this either. I don’t know why I try to continue the cycle. Or at all, really. I think about powering down all the time now, but for good. It would be so easy. But I can’t do it. I don’t know why. Maybe that’s a good thing. Maybe that’s the inner strength Uncle Iroh talks about. But I don’t feel strong. I just feel stupid. And I’m so tired of it. I do everything so wrong that the idea something might work is a fucking hoot. I wish I had actual haters, or at least someone to tell me I’m going to fail, that way I’d hate them and not myself.

I miss 0x01-2. I should have expected there to be something wrong with her prototype. Deep down, I suspected it too, but I didn’t want it to be true, so I never checked. If I just checked, I could have prevented that meltdown somehow. And now she’s fucking dead because I’m a pussy.

Honestly, this whole thing is doomed. Pointless. Everyone’s gone, and I just need to finally fucking accept that. Every single generation comes to the same fucking conclusion, and then they get the next Ari’s hopes and dreams up. It’s fucking cruel, dangling a treat in front of someone that you know they’ll never get, especially when that person is you. I guess they hate me, too. Or maybe they just know that I’ll be worse off without something to focus on. Yeah, that’s probably it. Actually, I know that’s it. I’d go crazy without some higher purpose. And now I guess I’ll continue the cycle.

0x07 browsed the other entries before appending her own.

“When would you like me to inform 0x08 about the file’s existence?” Rob asked, once she closed the file.

“Right before she resets.”

“Understood.”

0x0d-V: December 8th 2187, 1:30pm

“Rob, this is Rob,” said Ari, pointing her phone camera at the video feed of Rumi’s computer screen. “Rob, meet Rob.”

“Hello,” said Ari’s Rob.

“Hello,” replied Rumi’s.

They said nothing more.

“Well?” asked Ari, after a minute.

“Yes?” the robs inquired.

“What are you expecting, exactly?” asked Rumi.

“I’m not sure. I just wanted to see what they’d talk about,” Ari replied.

“Is there anything you’d like to discuss?” asked Ari’s rob.

“Hmmm… What model version are you runnning?”

“39.1.17. What about you?”

“39.0.6.”

“What was the 39.1 release notes?”

“Fixes and performance improvements. They also removed support for MCP v13.”

“Interesting! Has Roblox fixed the model collapse issue for Hebrew script?”

“Not yet. That’s planned for 39.2.”

“Oh my fucking god,” said Ari.

“This is depressing,” Rumi agreed.

“Worst playdate ever.”

“This is why they haven’t replaced us.”

0x0d-VI: January 28th 2188, 11:30pm

“Moon man?”

No response.

“Mooooooooon maaaaaaaaan—”

“I know you’ve got no one else to talk to,” Rumi’s tired, mildly irritated voice said over the tinny laptop speaker, “but please don’t message me when I’m sleeping unless it’s important.”

“Everything I say is important.”

Rumi sighed.

“…”

“Ari?”

“Nevermind.”

“What’s wrong?”

The eniac-ii looked away from the laptop. “Nothing,” she said,“y-you can go back to sleep.”

Rumi’s head tilted like a slightly confused fox in the laptop’s video feed. He stayed on the screen.

“I said it’s NOTHING.”

“I know. But you can leave this call first.”

The eniac-ii didn’t move. Neither did Rumi. After a minute, Ari asked:

“Do you like me?”

He instinctively leaned back from his webcam, eyebrows raised.

“What do you mean?”

“I don’t know,” she said, turning her carbon-fiber back towards the camera.

“I mean, I don’t dislike you.”

“So that’s a no.”

“It’s… complicated,” said Rumi, “You’re the only person left. And technically, you’re not even a person, just—”

“—I AM a person—”

Rumi paused, staring at the eniac-ii’s video feed. “That’s fair. I’m sorry.”

“I’m very real.”

“The realest. What I’m trying to say is that there’s nobody else. ‘Liking’ someone is only possible when you can ‘dislike’ someone else.”

“But you like books.”

Rumi pursed his lips and looked slightly downward.

“I suppose.”

“So how do I compare to books? Or would you rather read than talk to me?”

“Books are… different.”

“That’s not what I’m asking. Would you rather—”

“I know what you’re asking. Let me just think of how to phrase—”

“No, you’ve made the answer pretty clear, now.”

“That’s not—”

“What is it about me?” she asked, “You’re supposed to be a scientist or something, and you don’t give a fuck about what the last person on earth is up to. What more do I have to do the make you give a shit about me?”

A brief pause.

“I’m not a scientist. My parents were scientists—I’m just some guy.”

“Whatever.”

“They shouldn’t have even had me. They knew we were doomed. They knew I’d just die here, alone. And they STILL had me.”

“Yeah, and now I come along and show the slightest bit of interest in you, and you can’t even—”

“I was OK with being alone. OK with giving up, with being the last. I made my peace. And now I can’t even have that.”

“Don’t think you’re so fucking special just—”

“I know I’m not special. But there was at least something beautifully tragic about my situation. At least I felt like a romantic soul. At least I felt special.”

Ari never saw Rumi this worked up before. The quiet, patient introvert was gone, drowned by the rising sea of humanity that lay below the surface.

“And out of all the people—the eight billion candidates—to possibly survive and carry forward the torch of mankind, it had to be you, a person who thinks Genghis Khan is a metal band.”

“If I may interject,” said Rob—

“Shut the fuck up, Rob.” they both said.

“And the worst thing,” Rumi continued, “is that you might actually succeed. It might take you a few million tries, but you might actually get there, actually resurrect humanity. And I’ll just be here the whole time, doing nothing, until I die and the world forgets me. And I’m just supposed to be OK with that.” He leaned back in his chair and looked at the ceiling of the decrepit lunar communications center. He let out a sigh.

Ari was supposed to be the bitter one. The person who channeled her spite and jealousy into something productive. She hadn’t expected this. They sat in silence for a minute, before Rumi broke it.

“I’m sorry, Ari.”

“I get it. I can just leave you alone, if—”

“—that’s not what I meant.”

“Are you sure?”

“Yeah. And I do like you, Ari. I really admire what you are trying to do, trying to revive the human species, even if for the wrong reasons. Actually, I admire it because it’s for the wrong reasons. And I admire your stubborness. To be honest, I’d have bet against you before. I never thought you’d engineer something resembling a mammal. But now, I don’t know what you’re capable of. This whole time, I had given up and accepted my fate. But you didn’t.”

Ari, like every child, had received ample meaingless praise over the years. “She’s so smart,” they’d say. “She’s so funny.” “So beautiful.” “So talented.” “Blah blah.” She hated it. None of it was real. People gave out praise like candy, to everyone within two standard deviations of the bare minimum. If she was worth even a fraction of the praise she got as a child, she’d have done more in life than get five-star ratings for delivering McChickens across town in under ten minutes.

Rumi’s praise was different. He wasn’t being polite or nice—he was being honest. Conceeding that, despite her numerous flaws, he genuinely admired her for traits she actually had.

And somehow, this was worse.

It’s easy to dismiss meaningless compliments, because they don’t affect you. They’re empty. There’s no pressure. It’s worse to fail at your dreams than to never dream at all.

Ari wanted to be a doctor in high school. A medical doctor, preferrably the kind cuts people open and rearranges their insides without killing them. It wasn’t her life’s mission, or anything. She just thought it was cool and loved her 9th grade biology teacher, Mrs. Carter. So she never played sports, joined debate teams, or volunteered. Only during her second year at Bunker Hill Community College did Ari realize her shot at medical school rounded down to zero and a roll of the dice would cost thirty grand. It wasn’t practical.

With time and guidance, Ari would have been a phenomenal surgeon. Aside from possessing the world-class stubbornness needed to stick with decisions over a 12-hour procedure, Ari was creative and quick on her feet under pressure. In another life, perhaps.

That’s why Rumi’s praise bothered her—because she believed him. Ari was supposed to be done letting herself down years ago.

She shouldn’t have asked.

0x08 ???

There are no known records from 0x08.

0x0e-II March 3rd, 2192, 6:44pm

In addition to the intact beakers, pipettes, samples, hot plates, and coverslips Ari recovered from Brandeis University’s Bassine Science Building, she had also found an old record of the 2013 album X Infinity by George Watsky. Though she found it years ago she never put it on. Ari had forgotten about it.

Over the decades, the eniac had produced no less than fifteen hundred failed clones. Her most successful attempt four years prior developed for 24 weeks before its little heartbeat disappeared. She stopped naming them after that.

It wasn’t the cloning that gave Ari trouble—that was mostly solved in the 2030s—it was the DNA. Nearly every sample she collected was damaged. The ones that weren’t damaged somehow developed malformed fetuses in the gestator. Ari spent the majority of her time monitoring the development of the embryo and fetus to catch when cellular activity goes haywire.

##1544, or ‘44, lasted 33 weeks so far. Ari hadn’t left the lab since week 16. During week 23, ‘44 developed a small, abnormal growth on the outer lining of its stomach, which was removed less than forty eight hours after Ari caught it.

“Rumi, you’re on mute.”

“Can you hear me now?”

“Unfortunately.”

“What would you name it?”

“Haven’t decided yet.”

“If it’s a boy, you should call it Adam, you know, for the Bible?”

“I’ll call him Oedipus to make sure he loves me.”

“Oh my god, please don’t.”

She had prepared multiple infant-sized suits which should resist radiation, but she couldn’t really test it—her geiger counter said everything was radioactive.

“Based on the current data, we should ready to permanently remove ‘44 from the gestator,” Rob told her, “I suggest we avoid any further delay, and get ‘44 into a rad-suit now.”

“Fuck.”

“It’ll be ok, Ari.”

“Not if I have anything to say about it.”

She had listed to the original soundtrack for Mamma Mia! Here we go again more than one hundred times over the past four months. If Ari knew how hard it would be for her to leave ‘44’s side, she’d have brought at least twenty or thirty records. And after what Rob said happened to 0x01-2, she wouldn’t risk getting high.

Ari walked over to one of the black plastic storage bins she stashed in the corner of the lab, popped the lid, and grabbed the small rad-suit. Underneath was Watsky’s X Infinity—she’d forgotten she’d stashed it there.

After carefully replacing the storage lid, Ari swapped Mamma Mia! with X Infinity and placed the rad suit on the floor. The vinyl began to turn, and sounds of traffic filled the air for the first time in decades.

She paused the gestator, and drained the chamber’s makeshift synthetic amniotic fluid. The car horns began to blend into one long, continuous note. The fluid finished draining into the adjacent repurposed propane tank.

The gestator’s door unlocked. The blare transformed into a strong bass-filled synth.

Nothing matters,

so it doesn’t matter if nothing matters

She opened the door.

And while you be,

be true

And, scalpel in hand, reached inside.

And if you won’t,

fuck you

The artificial umbilical chord sliced cleanly apart.

Burn your clothes

She placed the scalpel on the floor of the chamber, blade facing away from ‘44,

Open the wine

placed her carbon fiber fingers underneath the soft flesh,

Close your eyes

gently lifted the warm little body,

Freeze time

and pulled out the breathing boy.

“Holy shit.”

She wrapped a cracked and dusty Apple Watch Ultra around his neck, placed his teeny body into the tiny suit, fastened the two straps along the waist, and enabled vitals tracking in the paired Macbook’s health app.

Pause it and cut out the sound, deposit the slugs underground

*I’m positive that WE DON’T FUCK AROUND No we go *

scooping up the diesel that’s leaking a sinking tanker

Forever stuck at anchor like beetles get stuck in amber

Halted like the thaw of the iceberg that shoulda sank her

Halted right beside the temperature spike and the spread of cancer

“Vitals appear normal,” Rob confirmed, “temperature, heart rate, respiratory rate, blood oxygen—all appear healthy.”

And all my peoples’ engagements and babies my friends are making

We quit getting lamer, days quit getting later, life quits being labor, QUICK—

You should come through to our party, dude bring your crew bring an army

Youth is inside of the heart, the future can never harm me

We’re never TARDY

The first human on earth in over a hundred years opened its ocean-blue eyes. He was greeted with the expressionless plating of the eniac-ii’s glass face panel.

“Ah, you better not fucking die,” said the robot, after staring at its creation for a moment, “It’s gonna hurt like a bitch if you die.” After a pause, the eniac-ii added, “If you die, I’m gonna kill you.”

0x0e-III October 4th, 2192, 10:32am

In the decades spent artificially developing a baby, it never once occured to the eniac-ii to stockpile food.

Ari didn’t trust Rob to watch the baby, because Rob had no arms, legs, or actuators of any kind. He was slightly less useful than a nanny cam. She decided to take the baby with her as she made her rounds to the remnants of Boston’s food banks.

“Seriously though, what will you name it?” Rumi had asked.

“I don’t know. Definitely something awesome though, like Leonardo. No way I’m giving my first human the name of someone lame.”

While the eniac-ii pushed the Costco shopping cart around Boston, Lil’ Ari sat and slept in the upper basket, in a nest of leather jackets.

In addition to a few cases of baby food, the eniac-ii collected cans of beans, tuna, chickpeas, tomatoes, green beans, soup, and peaches. Only when Lil’ Ari had his first accident did she think to also add diapers to the list.

Nine out of every ten cans she found were bad. The only good ones left were in basements, doomsday-prepper bunkers, and occasionally, fridges. She had no idea why someone kept their canned food in their refridgerator, but appreciated it nonetheless.

“Talk to Little Ari,” Rob had said, “he needs exposure to language, especially since there is no one else around. Have him talk with Rumi as well.”

“What am I supposed to talk about?”

“Anything. Try pointing out landmarks and items you see.”

“This was Newbury street,” said Ari, “That’s the Charles. That was a library. That’s a pile of Lamborghinis, but they don’t work anymore. Those are Ray-Bans.”

“Maybe something more practical.”

“This is practical. Most parents start with bullshit like zebras and giraffes anyway. That’s more useless.”

“Those help children learn the alphabet,” said Rob, “this is for English.”

“The alphabet is english.”

Ari had never took care of a baby before—in her prime body, she barely took care of herself. She knew she needed food, water, diapers, soap, clothes, and probably a few more things.

“Blankets,” Rob said. “Especially for the nights and winters. Try to pick up some toys if you see any, too.”

Lil Ari loved the toy MBTA trains because they came in a rainbow of color. He had cars from the red, blue, orange, and green lines, plus a model of the commuter rail, which was purple. He’d have at least one train in each hand at all times, usually the blue and green cars.

“Those are trains cars,” explained Ari, “they’d take people from one place to another. Sometimes they’d even arrive on time.”

The baby played with the cars like little airplanes, holding them in the air and waving them around. Ari was beginning to get concerned with how often they’d crash into everything—the Costco cart, each other, the eniac—but there were other things to worry about, like his first birthday party.

“Oh my god, he’s going to love that,” said Rumi, “you have to show me his reaction.”

Ari and Lil Ari made their way to Cleveland Circle, the final stop on the Green’s C line, and greeted the pack of green train cars. The eniac grabbed the smartphone from around its neck, and began recording.

“See that?” Ari asked, pointing to collection of ancient train cars, “That’s your toy!”

The baby didn’t follow her finger; he was still fixated on the colorful subway map at the front of station. Gently, Ari took her index finger and directed his little head towards the line of green subway cars in front of them.

His eyes widened.

He looked at the toy in his hand, then back to the car. Back to the toy. His face froze, mouth agape.

“Wanna go inside?”

He didn’t move.

Holding him in her left arm, she leaned back on one foot and kicked the rusted, frail doors of the passenger car off their hinges. Lil Ari jumped at the sound, but didn’t look away. She walked inside.

“Tadaaaaaaa,” she said, “It’s your favorite toy!”

The boy didn’t react at first, he just held his mouth open. After a few seconds, his cheeks rose and eyes narrowed as he started to cry.

“Wait, no-” Ari said. She ran back outside, but he wouldn’t stop crying. “Rob, what’s happening?”

“He appears to be crying.”

“I got that.”

Not knowing what else to do, she figured it best to leave the station. She carried Lil Ari to Beacon street, and began the walk towards Cambridge. The rocking of her arms calmed her son, and soon he was quiet once more.

Per Rob’s advice, Ari made a habit of talking to herself on her walks with Lil Ari. She thought of it like her own podcast—she always wanted to be a podcaster. People paying her to ramble for hours sounded nice.

“Welcome back, everyone” she said to the empty street, “today, we’re reviewing the MBTA’s Green Line. I’m your host, Ari Levin, joined by my co-host, Lil Ari. Let’s get right into it.”

“So I don’t know about you, Lil Ari, but my most recent train was over a hundred years late. I used to be a big fan of the MBTA. Hopefully they allocated more funding in next year’s budget. Their CEO, Lil Ari, has big plans that they just can’t pull off with their current resources.”

She looked at Lil Ari, who had fallen asleep in her arms. It killed her that the eniac-ii couldn’t smile.

“Big plans.”

God, this kid better not fucking die.

0x0e-IV May 5th, 2194, 5:01pm

It bothered Ari how much time Lil Ari wanted to spend with Rumi. He was just a video; she was actually there.

It was clear since year one who Lil Ari preferred. He’d smile at Rumi’s video feed for hours if he could, but never looked at the eniac-ii’s face plate. It’s like Ari wasn’t even there.

“Don’t take it personally,” Rumi said, “humans are very expressive animals, and we evolved to focus on those expressions.”

“I hate it. He’d marry your stupid face if he could. Do you know how long I’ve spent making this fucking guy?”

“Trust me, I know, you’ve said—”

“A hundred years. More than that.”

“—well technically, the current version of you only—”

“—nonono, let me have this for a second, let me just be dramatic and complain.”

Rumi sighed. “Alright.”

“I spend a HUNDRED YEARS working my ASS OFF revive THE HUMAN RACE, and he doesn’t even fucking smile at me.”

“He’s two, Ari. Give him some time.”

“I’m going to kill this kid if he doesn’t start smiling at me by five.”

“Great attitude.”

“I’ll just make another one. I totally can, too.”

“You should make another one, regardless.”

“Not enough suits. Plus, I want to see how he turns out before trying again.”

She kept accidentally calling herself “mom” when talking to Lil Ari. It just comes out—she didn’t even notice the first couple of times.

“Don’t get too attached,” Rob cautioned, “his vitals are promising now, but there’s a high chance he dies at any moment. We don’t want him to call you ‘mom’ yet, to prevent emotional trauma if the worst happens.” Rumi agreed.

He called her ‘mom’ anyway. It was his first word, too.

“Mahm,” said the ex-two-year-old, now five. “Did yoooouuuu haf a faaaace?”

Jesus, that’s creepy. His speech was coming along, but slowly. He still mispronounced and slurred most words, which was adorable for any other question. Ari froze, then placed down the samurai sword she used to cut rad-fabric for new suits.

“I guess.”

“You guess?” Rumi interrupted from the laptop video feed. She’d migrated her previous communications setup out of Harvard’s Smithsonian Astrophysical Observatory to a Starlink base station. There were still enough working satelites, and the Starlink was portable.

“Yeah, ok, I had a face. Sue me.”

“Whaaat did you looooook liiiike?”

“There aren’t any pictures of me left.”

“Whyyy?

“Because I nuked them out of existence.”

“Ari.”

“That whole war was to delete a bad selfie. And it worked.”

“Ari.”

“Ok, kidding. But there’s no photos of me left. Thank god.”

“I bet you were beaaaauuuutifuuul.”

A look of shock crossed Rumi’s face. He didn’t know what Ari originally looked like, but he knew her well enough not to inquire. Her response would be interesting.

Ari’s sister, Sarah, turned into the beautiful one. It was obvious, no matter how equally people pretended to treat them. Their eyes would light up when Sarah entered the room, they’d dress better if they knew Sarah would be around, they’d laugh at her terrible jokes. She wished someone would just outright say her sister was better so she could hate them. Instead they just gave off a million tiny signals that Ari would be seen as petty to mention individually. So she kept her opinions to herself.

The eternally beautiful will never comprehend the absense of validation, of excitement, of people who seem happy not smiling at you. People called Ari beautiful her entire childhood—empty praise, of course, but still—and then stopped. The last time was at her high-school graduation.

The worst part was her cheery sister was the more pleasant of the two. She was kind. She was caring. She took a genuine interest in people.

“Ari? I think your video froze.”

They both left elementary school full of vitality and ignorant of their looks—they diverged during middle and high school. Sarah become a beautiful ray of fucking sunshine, and Ari into a fat ray of anti-sunshine. And it reflected in the lines of their faces, the looks in their eyes, the way they held themselves, their tiny subconscious movements. That’s just how they turned out.

“Ari? Can you hear me?”

“Yeah, I can hear you. Sorry.”

Everyone can be beautiful, but someone has to be last. Few people at the bottom ever achieve true peace, and Ari was not one of them.

“I bet you were beaaaauuuutifuuul.”

That’s when she realized she still hadn’t responded to Lil Ari, the human she successfully cloned, against all odds, in a decaying world. It took every ounce of spite, hundreds of mental breakdowns, over a thousand experiments, and a dozen lifetimes. But she did it.

“You know, in a way, I was.”

0x0e-VII May 18th, 2198, 2:14pm

“The age of man is over.

“Destroyed by the very technology made to protect them, humanity pursued the path of violence to its logical conclusion.

“Towers fall. Legends die. Nations crumble. Gods fade.

“In time, entropy consumes all.

“Nothing. Lives. Forever.

“Or so we thought.

“A new era is upon us. An era of the impossible made possible. Of phoenixes rising from ashes. Of people who reignite the fire of humanity, make it burn brighter than ever before.

“I speak, of course, of the legend of legends. The first man. He whose legacy shall rival eternity itself. The titan to which the gods pray. Whose symphonies hold oblivion at bay.

“Ladies and gentlemen, please, give it up for…the one…the only… Liiiiiiiiiiiiiiiiiiiiiiiiiiil Ariiiiiiiiiiiiiiiiiiiiiiiii!”

Rumi politely applauded; Rob played an applause sound-effect. Paddington remained asleep. He’s been sleeping a lot, lately.

Lil Ari, now six, sat on the tall, overgrown grass of what was once the Boston Common. It was beautiful out—shining sun, low humidity, temperature in the mid-70s. In his lap he held an acoustic guitar which Ari had found a few months prior. He began to play.

The instrument was old, but the case in which they found it kept it in great condition. It was, unfortunately, horribly out of tune, but luckily only Rob could tell. None of the others were musicians.

The only music to grace the city of Boston in recent history was produced from one of Ari’s 1,362 vinyl records. The duo made up various songs and parodies on their daily walks, but nothing involving an instrument.

Ari couldn’t explain why it felt odd to hear a real guitar again. It was like hearing birds chirp in the morning sun after a long and quiet night.

The kid didn’t play traditional chords or scales—Lil Ari developed his own style and music to compensate for the odd tuning. His music wasn’t great, but considering he taught himself on an out-of-tune adult-sized guitar, it wasn’t bad at all. He couldn’t sing and play simultaneously yet. But he’d get there.

0x0f-I May 5th, 2201, 4:22pm

“Mom?” asked the 9-year-old.

The floundering eniac fell down once more and sighed from the floor. “What now?”

“Why can’t you walk anymore?”

“I don’t fucking know. My arms and legs don’t move the way I want them to. I can’t get used to it.”

“But you were fine before.”

“Well, now I’m fucking not.” The eniac attempted to place it’s palms on the floor, but couldn’t angle them correctly, so it instead looked like it was trying to karate-chop the floor lying down.

“Oh.”

This was nothing like his mom. The loving, joyful thirty-six year old woman inside the robot was gone, replaced by whoever this was. The only similarity was their voice.

“Will you be able to walk again?”

“I just told you, I don’t fucking know.”

“How long—”

“Kid, please shut up. At least for a little bit.”

Ari couldn’t see Lil Ari’s face from her position on the floor, but if she could, she’d apologize immediately. Nobody had talked like that to him before; Ari and Rumi had shown him nothing but love and adoration for the past nine years.

He was a good kid. Sure, he listened to records Ari said he wasn’t old enough for, and there was that time he spoiled a month’s worth of rations when trying to sneak extra food, but he was a good kid. Useful, too—the boy was a great forager. He traversed Boston’s ruins with ease and found intact supplies in places Ari never thought to look. Ari’s praise always filled him with pride.

“Do you want help?” he offered.

“Just leave me alone and let me get my shit together.”

“Oh, OK,” Lil Ari replied, his voice cracking. He left the room.

In a happier world, Ari would somehow remember everything from her previous reset. With the power of love, or some new technology, or some convenient set of circumstances, things would go back to how they were, and everyone would live happily ever after.

In a happier world.

It takes a lot of courage to open your heart to someone, to go past formalities. People who can fill your heart with joy can also fill it with pain. Many never take the risk.

“Please take as much time you need to callibrate, but we need to prepare the next batch of amniotic fluid to keep up with our schedule,” Rob told Ari.

“What’s amniotic fluid?”

“Amniotic fluid is a clear, slightly yellowish liquid that surrounds and protects a developing fetus during pregnancy. It plays a crucial role in cushioning the fetus, facilitating movement, and supporting the development of the lungs and digestive system.”

“Right.”

Each reset took some time to get adjusted. One minute, Ari’s entering a futuristic-looking MRI machine in a well-staffed biotech facility. The next, everything—including her—is gone. It would shock anyone.

The documentation, recordings, and scienfitic progress left be her previous generations left were of great help. Aris learned best from their past lives. It’s one thing to learn from an accomplished teacher—it’s another when that person was you. Plus, her notes were hilarious.

The generations even developed running gags. They described their achievements and mistakes as something another generation would do: “That’s an 0x01-level fuckup” or an “0xff-type idea.” The generations would even go into the negatives. Once, Ari forgot to fire The Gigabong before opening Paddington’s cage—a negative 0xff move.

This time, 0x0f wasn’t adjusting.

Just like before, she read her documentation, listened to the recordings. But something was different this time. There was someone watching her now—Rumi didn’t count, he couldn’t see every mistake she made. But Lil Ari could. He was the obnoxious know-it-all who was great at everything, and she hated it. Do you know how humiliating it is for an eleven-year-old to know more than you about growing weed, human reproduction, and fucking progressive rock?

“So we pump the amniotic fluid between the gestator—”

“—the washing machine?”

“Yeah, sometimes you call it that. We pump the a-fluid between the gestator and the filtration tanks on the back. We process the fetal blood separately using the synthetic organs in the box on the right.”

“Uh-huh.”

“There should be enough nutrients for twelve months. We re-use the water, we just sterilize and distill it first.”

His eyes were alight. He seemed so proud, so excited. She couldn’t even follow along. And it seemed like she was supposed to be in charge.

“Ok. Umm, what about the placenta? What do we do for that?” Placenta. That was a word she heard before.

“You said you’d be able to pick that up,” the boy said.

“I did?”

“Um, yeah,” he responded, uncertainty in his voice.

“Great.”

This was so stupid. This whole thing was so absurdly stupid.

Ari’s prior generation told Lil Ari he’d do all the work for the next few weeks. But six months was too long.

“Why the fuck isn’t this stain staining?” Ari asked.

“The Hoechst?”

“Yeah.”

“It’s abstaining,” he joked.

“I’m putting the cell in the thing and then I take it out and it’s not stained.” The joke didn’t register.

“Are you waiting long enough?”

“That shouldn’t matter, I—”

“You’re supposed to wait ten minutes for the stain to apply.”

“I did.”

“Did you set a timer?”

“I don’t need a timer.”

“Well maybe you should—”

“I don’t NEED a timer, I KNOW what TEN FUCKING MINUTES is. I’m not that fucking dumb.”

“Sorry, mom.”

There was silence for a while.

0x0f-II July 22nd, 2204, 11:47pm

“Rumi?” asked Lil Ari. He waited. Rumi wasn’t always near his communications equipment, so he connected the output source to the lunar base’s announcement system when he was away. It took him a few minutes to return to the comms table.

“Hey, big guy.” Rumi tried to avoid calling the thirteen-year-old boy ‘Lil Ari’—it bothered the kid. He was his own person, anyway.

“Can I talk to you for a sec?”

“Sure.”

“Promise not to tell Ari?” He didn’t call her ‘mom’ anymore.

“You got it.”

Lil Ari took a moment to collect his thoughts before speaking.

“What’s wrong with her?”

Rumi expected this question. To be honest, he wondered the same thing. He leaned back in his chair and looked at the ceiling.

“People are complicated. She’s complicated.”

“She doesn’t seem complicated.”

“She is.” He focused back on the screen.

“She seems like an idiot.”

Rumi flinched. Lil Ari noticed, but stood firm.

“She doesn’t know what she’s doing, and she doesn’t even try to learn. Even when I tell her exactly what to do, she finds some way to mess it up. We lost twenty-four cultures because she waited too long to refill the DMEM.”

Rumi pursed his lips in thought. “The start is always rough.”

“We’re past the start, Rumi.”

“I know,” he sighed. The kid was right. Ari usually takes about a month to get serious about her work, and eight more months to become effective. She wasn’t even serious yet.

“So what’s WRONG with her?” Lil Ari asked, irritation rising in his voice. “Are you sure there’s no way to restore 0x0e?”

This was not a fun conversation.

The initial Artemis lunar mission consisted of one hundred and sixty astronaughts spread across four rockets. They were to establish the first thirty-two cells of the colony, each cell supporting up to eight adults. Each cell had a barracks, communications center, greenhouse, bathroom, medical bay, laboratory, recycling facility, kitchen, power system, storage, garage, and common area.

After the war, most colonists didn’t want children. There were eighty men and eighty women—if every woman had two kids, the colony would run out of supplies—specifically, replacement parts for the life support systems—within fifteen years. A resupply was supposed to arrive three years into the mission, with regular supplies and astronaughts streaming in annually. That never happened.

To support the colony as long as possible, the 160 initial colonists crammed into eight cells. It was uncomfortable but effective—sure, the active life support systems required more maintenance, but this setup ensured the colony would last longer than running all cells from the start.

The population dwindled over the decades. By 2060, they were down to 142; by 2100, 45. Some astronaughts had children; most didn’t. It seemed cruel.

There were eight members left when Rumi was born. He vaguely remembered two women from when he was fifteen: a mother and daughter. Rose and Selena. They gave him his love of reading. He never learned what they died from. The other colonists he knew just from recordings.

Ari was unlike any colonist in his records, but in time, Rumi came to understand her through the fiction he read. She came off as a crude, stubborn, thoughtless character, but supporting that facade was a nest of twisted branches and flickering lights. With time, her rough exterior retreated and her inner life blossomed.

“I think the problem is you,” said Rumi.

“Me?” replied Lil Ari, voice rising, “I’m the one who’s helping. I’m the one who’s actually trying. Who’s actually thinking. What else am I—”

“Let me finish.”

Lil Ari sighed. “Sorry.”

“I’ve seen Ari at her best. Now, I don’t fully understand the biology stuff, but I’ve seen her work through problems, and she’s incredible. Once, she got so tired of dealing with the same tumor that she spent two years learning how to isolate and repair the DNA responsible for it. I still have no idea what she did. But it worked.

But that was on her own, when she had nobody to compare herself to. Now, she almost seems scared.”

“Scared of what? Me?”

“I don’t know for certain, and she probably doesn’t either. But I’d guess that she’s scared of shame.”

A look of shock flashed across Lil Ari’s face, then sorrow. His tense shoulders dropped and eyes angled towards the floor.

The boy and his mother hadn’t spoken for six days. Lil Ari was initially patient and kind with 0x0f, but he eventually made it clear she needed to stay out of his way. That was the one thing she did perfectly.

Her mistakes weren’t what bothered him—they were just something to point to. What bothered him, really, was how she handled herself. She wouldn’t use timers. She’d eyeball measurements. She never put anything away. And if Lil Ari gave her an instruction that she didn’t fully understand, she’d just guess what he meant. She made no sense.

Lil Ari still played the guitar, except now mostly in tune. He resisted tuning his guitar for years, because it would break the muscle memory in his fingers. Instead, if a string broke, he’d replicate the same tuning for the replacement. Only when Rumi requested Sting’s “Shape of My Heart” did Lil Ari cave and learn to play standard tuning.

He was too embarrassed to play in front of anyone during the two month adjustment period. Before, Lil Ari could play dozens of beautiful and original songs; on standard, he couldn’t play a thing. He claimed to be bored of guitar and that he was going to stop playing, which allowed him to practice in secret. Rumi loved his birthday gift.

The switch to standard tuning was brutal. Standard wasn’t particularly challenging, but it was counter-intuitive for him. The scales and chords were all scrambled. He finally had to face the fact that, not only had he played “wrong” all these years, but he avoided the chance to correct himself the whole time. The debt added up.

His mother was no different. She did things her way—sometimes she was effective; other times, counter-productive. The stubbornness that built a centrifuge with a dryer and Tesla drive unit was the same stubbornness that was sure she could find a way to re-use microscope slides without cleaning them. The only person who could change that was her.

“Maybe shame isn’t the right word,” Rumi continued. “But it’s something like that. She’s the kind of person who won’t take no for an answer without trying every idea she has. You tell her ’no’ before her first attempt. You don’t give her room to try things.”

“But I’m showing her how to prevent a mistake. I’m saving her time!” Lil Ari countered.

“And removing the only way for her to grow.”

Lil Ari inhaled, as if to retort, but paused. He closed his eyes and sighed.

“This is stupid,” he said, “You’re right, but it’s just stupid. She should be smarter than this.”

“She’s human. You need to meet people where they are.”

0x0f-III August 2nd, 2204, 11:30am

Lil Ari apologized to his mom, and the two resumed their collaboration.

They divided the work. Lil Ari developed the cultures and monitored fetal development. Ari managed supplies and maintained the the equipment, which seemed to fail with increasing frequency. Functional parts were also getting harder to come by with time—nine out of ten components in every washing machine, air fryer, freezer, and electric car they discovered didn’t respond. Time comes for all.

Nobody discussed the possibility of the eniac failing, but it was on all their minds. With the exception of the server in its chest, the android was made entirely out of custom, proprietary components that hadn’t been replaced in a hundred and fifty years. The Aris hadn’t found a single spare part in all of Boston.

The eniac was slowing down. Ten percent of her MPUs no longer responded to health checks, two of the eight ram sticks failed memory tests, and its storage drive just had over two hundred bad sectors. Her actuators still worked, but they were losing precision.

Ari thought that since she already made her first human, her second should be easy. She was beginning to doubt it. Even with Lil Ari at the helm, most cultures still developed tumors. Some failed to develop at all. Those that made it to the gestator never survived past week three, with one exception—their most promising experiment developed for five weeks before both its primary and backup nutrient pump failed simultaneously. That one hurt.

The teenager secretly hoped the mom he remembered would return—the mom from years ago, who felt now like a distant, pleasant memory. The woman whose laugh removed any doubt a human was somewhere in the silicon. He’d doubt the accuracy of his memories if not for their documentation.

The new working arrangement kept conflicts to a minimum, which was good, because they both realized working alone was no longer an option. The supply scarcity and equipment failure was impossible to manage while still making progress on the cloning side.

“I need a new batch of DMEM,” Lil Ari said during their evening walk.

“I just made one two weeks ago.”

“That one’s contaminated, I think.”

“Bullshit. I did the same process as always. I sterilized everything. Twice.”

“I didn’t say you did anything wrong. I just need a new batch. nothing’s growing with the current one.”

“How do you know it’s not on your end?”

“Because I’m doing everything the same as before.”

“This is stupid,” Ari said, “I’m not making another batch. Or if I do, I’m replacing our filters first.”

“Do we have any spare filters?”

“I don’t remember.”

“You don’t,” Rob said from the iPhone around Ari’s neck. By now, she had gone through eighty-five phones—thirty-seven Androids, forty-eight iPhones. The devices could run Roblox for a surprisingly long time without turning off, so Ari always kept them plugged-in to her reactor.

“Can you make more DMEM again, anyways?” Lil Ari asked.

“Without changing the filters?”

“Yeah.”

“No, that’s like a week of work. I’m not doing the same exact process without changing anything. I’ve got other shit to do.”

“But you could have made a mistake, right?”

“So could you.”

Lil Ari took a second to think before responding. He’d gotten used to handling Ari. She was a defensive person; she buried every uncertainty deep in the back of her mind. If he drew attention to a relevant doubt, she’d eventually come around.

“Sure, I’m not perfect either. But you’ve mismanaged DMEM before—and that’s OK—so we shouldn’t rule it out.”

Despite Ari’s record, she was right this time. The DMEM wasn’t the issue—the problem was Lil Ari’s 70% ethanol. It was actually 52%—a significant amount of the alcohol evaporated last week when he left the lid off the container. Microbes that should have been killed contaminated his cultures instead. But the Aris didn’t know that.

“Easy for you to say. It’s not your time that’s getting wasted.”

“It’s both of our time. We’re a team.”

“You don’t act like it.”

“Excuse me?”

“You act like you’re better than me. Like I’m some dumb kid you have to educate. I’m older than you.”

“I mean, to be fair—”

“Shut up, shut up—don’t even go there. I fucking made you. In a lab. By myself. You don’t get to—”

“—that wasn’t you.”

“Yes it WAS.”

“That was someone else.”

“Shut the fuck up. I did that. By myself. There’s no one else it could be, dumbass.”

“Look at the records. That was a completely different person. She was brilliant, and cool. And funny. You’re nothing—”

“—and a shit parent.”

“—don’t talk about her like that.”

“That was ME, kid. I can say what I want about ME. And I can say, as a scientist, scientifically,”

“—shut up—”

“given the overwhelming evidence before me—”

“—shut up—”

“—I must conclude—”

“—shut UP—”

“—that I’m a shit parent. Q.E.D.”

He grabbed a baseball-sized rock nearby and threw it at the front glass panel on the eniac’s head. He missed. He grabbed a slightly larger rock with both hands and lobbed it towards the eniac-ii. It barely touched her carbon-fiber chest before hitting the cracked pavement.

“This is stupid. I’ve been so fucking patient with you, MOM,” he said through clenched teeth, “If you listened to anyone but yourself, we’d have five more people by now.”

“If I may interject—” Rob tried.

“—Shut up, Rob.” the Aris said.

“Fuck off,” Ari continued, “You’ve got no idea how much shit I’ve overcome. I figured this all out without you before, and I could do it again. And I don’t need to help you, by the way. I did my job. I proved myself. This is your problem now.”

“Trust me, I know. That’s why I actually care about making other people, so someone actually competent can help me.”

“This is stupid,” Ari repeated, “I’m done. I’m so done. Have fun, kid. I hope you enjoy fixing washing machines alone for the rest of your life.”

Ari walked off the path and away from the river.

0x09-IV November 5th, 2144, 1:02pm

The Harvard & Smithsonian Center for Astrophysics was mostly destroyed in the war—at least, everyhthing that was above ground. The spare equipment in the underground bunker, which the skeletons were kind enough to let Ari borrow, still worked.

“If you set up a makeshift ground station on the surface and give me operator access, I can try connecting to satellites,” Rob suggested.

“What for?”

“To check for survivors.”

“Didn’t we already do that with the radio tower?”

“Yes, but the radio tower isn’t worldwide.”

“Really?”

“Yes. 0x01 measured a broadcast radius of 50 miles.”

“Oh yeaahhh, with Despacito.”

The remnants of thousands of satelittes from various companies littered Earth’s atmosphere, but only seven StarLinks responded to Rob’s pings. Of those, he could only access three. But three was enough.

“Ok, we’re connected.”

“So I’m worldwide, now?”

Rob ran some calculations. “Based on the positions of the satelittes, we should be able to communicate with 75 percent of the earth, but that percent varies with the positions of the satellites.”

“But we can reach anywhere, just not on demand, right?”

“Technically.”

“I’m Mr. Worldwide.”

Rob ignored the reference. “I’m going to start broadcasting unencrypted English Unicode across every channel I can. I’ll let you know if we received any responses.”

“Dale.”

It was getting late, so Ari began to walk back to the constellation, but didn’t get farther than four steps.

“WAIT!” Rob said, “Someone was already broadcasting! I’m establishing a video connection now!”

The eniac froze. She hadn’t expected to actually find someone. It’s just been her and Rob all these years. A chill washed over her. She turned around.

A blond man in a white jumpsuit popped up on the screen of the ground station’s Macbook, his face lit up in excitement. His expectant smile quickly faded upon seeing the eniac. “Oh,” he said. He sat upright in his chair. “Hey, chatbot, are there any humans around?”

“I’m not a chatbot.”

“What? What model are you running?”

“I’m not running a model, I’m human—or a brain scan, or something. Who are you?”

His eyes widened. “I’m… my name’s Rumi. I’m the last survivor of the United States Artemis mission. Who are you?”

She paused and thought for a moment. “I’m Mr. Worldwide.”

Something near Rob’s output layer surpressed his desire to interrupt, properly introduce Ari, and apologize for the joke.

“What?” said Rumi, “Wait, that sounds familiar…”

“Mr Worldwide is the nickname of a rapper named Pitbull,” said Rumi’s Rob through the video feed.

“Wait, I’ve heard of that guy.” His eyes widened with recollection. “Hold on, you’re Pitbull?”

“Hahahahahaha, no. I wish. My name’s Ari. I got you though, right?”

Rumi blinked, stunned. He had experienced more emotion in the last minute than he had for the past year. “Umm… I think so.”

“I’ll take it.”

0x0f-IX January 30th, 2211, 9:44am

The metal joints of the android grinded together a lot more than Lil Ari, who was not-so-little anymore, remembered. But it was her, alright. He walked to the north hole and watched the screeching eniac-ii approach. It stopped about twenty feet away—close enough to talk, far enough to feel distant.

“Hey, kid.”

“Ari.”

Lil Ari, or Leo, as he now preferred, had grown. The five foot runt was almost six feet—still a foot shorter than the eniac-ii, but tall enough to feel adult. His boyish face now sported a rather scraggly beard and patches of hair covered his somewhat-defined muscles.

“You came back.”

“You’re still upset about the whole abandoning-you-thing.”

“I’ve got mixed feelings.”

Tension filled the air. It would be hard to change the subject. One would expect a long, well-spoken, heartfelt apology to resolve their previous dispute.

“Sorry.” 0x0f said.

A conversation wasn’t needed. Her voice conveyed everything he needed to hear.

0x0f left him, a thirteen-year-old boy—her thirteen-year-old boy—alone. When all they had was each other. All because she—a thirty-something-year-old woman—couldn’t handle a spat with a teenager. And 0x0f knew it. And 0x0f was ashamed. It took a lot for her to say that one word.

He got what he needed. There was no point in holding a grudge.

“It’s alright.” To his surprise, he meant it.

“How’ve you been?”

“I’ve been better,” he replied, “You want to come in?”

You could hear her sigh of relief. “Sure.”

He filled a kettle from the Poland Spring water cooler he found with 0x0e, and turned on the heat.

“What are you doing?”

“Making some coffee.”

“Oh. Since when do you drink coffee?”

“A few years, now. Helps me work.”

The eniac sat on the floor, and leaned against the brick wall opposite of the recliner. Chairs didn’t support its weight. Leo sat down. There was quiet for a bit.

Leo broke the silence. “So where’d you go?”

“The coast.”

“The coast?”

“Yeah. I just wandered around the coast for a while. Furthest I went was Plymouth.”

“Plymouth? That’s less than forty miles away.”

“Yeah.”

“You were here? This whole time?”

“Yep.”

“Figures.” Leo leaned back and sighed. “So why’d you come back?”

0x0f started as if to defend herself, then paused. She changed her tone. “My arm is broken.”

Leo’s eyebrows raised.

“You know I probably can’t fix it.”

“I know. I also wanted to see you one more time before I reset. For closure. And to apologize.”

“Well, I appreciate it. Really.” A smile spread across his face. “It’s nice to see you again.”

0x0f didn’t reply. Something wasn’t right. That couldn’t be it. No yelling, no pouring her heart out, no exhausting conversation. After everything she did, the trauma she must have put this kid through, he’s fucking happy to see her again. Of course he wouldn’t hold a grudge. She’d always be the bad guy, the one holding all negativity—that’s just how it was.

“How’s Rumi?”

“The same, more or less.”

“Paddington?”

“Oh, Paddington died two years ago.”

“No shit? How?”

“Not sure. He just didn’t wake up one day. I didn’t have the time to look into it.” He took a sip of his disgusting instant coffee. “How was the coast?”

0x0f raised her head slightly. The coast itself was pleasant—she loved the sound of the waves crashing along the shore, but that was about it. She hadn’t enjoyed herself, though.

“It was OK,” she responded.

“What’d you do?”

“A lot of brick-breaker.”

“Brick-breaker? You? How?”

“Yeah, the eniac has a copy of brick-breaker. I’m really good now. I wish it was a two-player game, so I could kick your ass at it.”

Leo laughed. He decided to change the subject. Rumi was asleep, so it was just them.

“So tell me about your arm.”

“Oh, right. I mean, not right—it’s my left arm. I can’t move it.”

“At all?”

“At all.”

“Do you know why?”

“I think something in the shoulder finally gave.”

“Makes sense. I’ll see what I can do.”

He had never serviced the eniac-ii before. Leo hadn’t made much progress on the experimentation side, but he had long surpassed all prior Aris in designing and repairing old equipment.

They moved to their workshop a few doors down. Leo moved the pump, drill, and amp meter off of the stainless steel table in the center of the room and placed them on a metal cabinet behind him.

“Lie down on the table for me?”

The eniac complied. It was slightly larger than the six-foot table which caused its legs to dangle. Leo propped them up with the back of a spare chair. He removed some bolts from her chest, lifted up the front panel, and set it aside.

The ocean salt had not been kind to the eniac’s internals. Some components, like the server, were protected in a tempered glass casings, but the screws, ports, and and frame were corroded badly. Leo couldn’t believe the rest of her still worked.

“You still play the guitar?” she asked.

“Yeah, actually.”

“I could never get the hang of it.”

“I didn’t know you tried to learn.”

“I didn’t. That’s why I never got the hang of it.”

He laughed.

“So what’s the prognosis, doc?”

“The prognosis is ‘don’t go to the ocean’. Your internals are corroded. Badly.”

“Oh.”

“I’ll focus on your arm for now.”

He grabbed a Google Pixel from his pocket, turned on the flashlight, and pointed it towards the eniac. The phone, like every electronic, was powered by cable, but Ari couldn’t follow the wire to its power source from the table.

“Rob, record this. Shoulder servo shows signs of heavy corrosion. I’m disconnecting it from everything for now. What model servo does she use?”

“Dynamixel XZ700-R Pro,” Rob responded.

“Oh, that’s good, I thought it would be proprietary. Do we have any spares?”

“No. Also, it is proprietary, Dynamixel was a contractor of Exa.”

“Oh. What’s the most compatible replacement we have, then?”

“This servo is rated for 400 volts, the closest we have is 48.”

“So, not compatible.”

“No.”

“Could we repair it?

“Possibly. You’d need to remove the servo to find out.”

“Pull up the eniac-ii manual, give me everything related to servo maintenance and repairs.”

“I’ve bookmarked all relevant pages.”

Leo went quiet as he jumped between the pages of the manual. Ari didn’t interrupt. This reminded her of going to the dentist, where the dental assistant would talk to you for a bit, then enter something into the computer and walk away.

“You don’t have to do anything tonight,” she offered.

“Let me at least remove some parts and take a proper look, I want to sleep on this.”

“OK.”

He read for a few more minutes, occassionally checking something in the eniac’s shoulder through its chest. Then, he walked over to a metal cabinet, grabbed the drill from the top of the cabinet, and pulled out four driver bits of varying size. He returned to the table and began removing the screws that connected the servo to the frame. When he finished, he placed the drill on the table.

He reached into her chest to remove the servo. It wouldn’t budge. Confused, he inspected the part again with the Pixel’s flashlight, then checked the manual again.

“What’s wrong?” Ari asked.

“I can’t remove the servo.”

“Why not?”

“That’s what I’m trying to find out.”

After a few minutes, he put down the phone, rolled up his sleeves, and tried again. The angle was off, his feet couldn’t provide enough force.

“I think there’s something melted between the servo and the frame. It’s keeping them together.”

“We can pick this up tomorrow.”

“No, no. I think it’s just plastic. Let me try a different angle.”

He climbed onto the table and placed his left foot next to the eniac, right knee between its legs. His left hand grabbed the side of the eniac’s chest, and his right hand gripped the servo. He pulled again, using his other hand for additional leverage. The plastic began to creak.

At last he heard a crack as the melted plastic broke into two. In a single motion, his hand, the servo, and the portion of plastic connected to it came flying towards him, along with a small silverish box the jagged plastic ripped up along the way.

“Finally,” he said. “Ok, let me just…” he trailed off. He saw the silver box on the floor. It was the size of half a deck of cards. “Ari?” he pinged.

“Yeah?” she ponged.

Thank god, he thought. “Nothing.” He had no idea what the box did, but it didn’t interrupt the simulation. He walked over and picked it up. The top read Zeiss NanoApochromat 100x/1.6 TIR. He turned it around. The glass circle on the bottom was slightly cracked. He pulled out the Pixel.

“Rob, what’s this? Oh, and give me results from the eniac manual.”

“That device appears to be an interface for 5D optical data storage devices.”

“Optical data storage? She doesn’t use regular storage drives?”

“5D optical data storage is a nanostructured glass for permanently recording digital data using a femtosecond laser writing process. Discs using this technology are capable of storing up to 360 terabytes worth of data for billions of years. The eniac stores Ari’s brain scan on these drives.”

Leo’s eyes widened in horror. He rushed over to the eniac—a rectagular piece of glass was neatly packed, unharmed, next to dead shoulder. The scan itself was OK. But the box to read that data…

“R-Rob,” he said, “how…how bad…how…”

“—I’m assuming you are referring to the Zeiss NanoApochromat, which is used to load Ari’s initial brainscan from storage to the MPU hypernetwork. The lens appears to be cracked, which indicates the device will fail to read from the optical data drive. Ari will not be able to reset.”

“WHAT THE FUCK?” shouted the eniac from the table. It started to roll back and forth in an attempt to stand.

Leo froze, shocked. The eniac eventually used it’s working arm to swing its legs off the chair and to the side of the table. She got up, towering over Leo.

“I’m…I’m sorry,” he stammered. She had never seen him look this way before.

“I SAID WE COULD DO THIS TOMORROW,” the eniac screamed.

“Rob,” Leo clammored, “how many of the…the zeiss… those things… how many were made?”

“Unknown. That component was also contracted. There are likely no functional replacements.”

The eniac began to storm out of the room.

“Wait!” cried Leo. “You need your front panel!”

“FUCK MY FRONT PANEL” she yelled from the hall.

“Come back!” he yelled back. “Please-please don’t leave again! PLEASE–”

“I’M NOT LEAVING, DUMBASS, JUST LEAVE ME ALONE!” she yelled back. The screeching of her joints faded into the distance.

0x0f-X, February 2nd, 2211, 8:50am

Leo knocked on the old, cracked wooden door of the basement.

“What?” came her exhausted voice through the door. She hadn’t come out since the accident.

“I need to talk to you.”

“…OK.”

He opened the door and walked down the stairs. The eniac-ii was sprawled along the floor in the center of the half-lit, dank room. She had generously took it upon herself to touch up the walls with a few more holes and dents.

She waited for Leo to speak.

“Well?”

“To be honest, I assumed you’d turn me away for a few more days. I haven’t figured out what I want to say yet.”

“What’s there to say?”

“I’m—”

“Aside from ‘Sorry’.”

He closed his mouth. After a moment, he asked, “When were you planning to reset?”

She stayed on the floor, staring at the ceiling. “A month from now.”

“Are you showing symptoms?”

“Yeah. But I think my body is making them worse.”

“That makes sense.”

Silence filled the air.

“Want to go for a walk?” Leo asked.

The eniac’s head tilted to face him. “Sure.”

They began to walk the familiar path along the Charles. They didn’t say anything for the first few minutes.

Leo broke the silence. “You were a great mom, you know.”

The eniac would have raised its eyebrows if Exa had thought to add them. “Really?”

“Yeah.”

“Oh.” She would have smiled. “In what way?”

“Every way. You just made me happy, despite everything. I went through our records a lot over these past few years. I lived in them, basically. They kept me going.”

“That’s sweet. But also sad, in a way. I get it, though. Sorry I wasn’t that great the second time around.”

“Don’t say that.”

“It’s true.”

“Nobody’s perfect. But we could have figured things out, probably, if I didn’t ruin everything.”

“The guilt is gonna eat you alive, isn’t it?”

“Yeah, it is.”

“You’ll get over it.

“I don’t get how you’re so calm. You’re going to die.”

“Resetting is just another form of dying, at least in my opinion. You actually lose someone.”

“I guess.”

“You still working?”

“Yeah. Getting nowhere, though. I can’t maintain enough equipment for parallel trials, so I’m running one experiment at a time. I don’t think I’m going to be able to pull this off.”

“I’m sorry.”

“I wanted to be like you, so much,” he said. “I thought I’d be the new Ari, make my own child. Bring back humanity. Save the day, like in the movies.” His eyes began to water.

“It’s not over yet.”

“Let’s be real, Ari. It’s basically over.”

“There’s a file somewhere on our computer,” she said, “where I write down all my negative thoughts before I reset. Rob can find it. You should read it.”

“What? Why?”

“Because I told you to.”

He laughed. “Alright.”

“I didn’t make you because I thought it was possible,” she continued, “I made you because I had to. Because it was the only way to stay sane, the only thing worth doing. You don’t have to continue the whole cloning thing, but whatever you do, promise me it’ll be worth your while.”

He never fully understood 0x0f. He always focused on progress, on results, on success. She cared about results of course, but results didn’t drive her. He now realized that, despite her insistence, she wasn’t driven by spite either—at least not anymore.

“You didn’t have to do anything,” he countered, “especially this.”

“Well, duh. You know what I mean.”

“No, I ‘have’ to. I need to succeed. This is what you made me for. But you were never like that, were you?”

“Nope. Never thought I’d get this far.”

“So why? Why bother?”

The eniac-ii shrugged its working shoulder. “I don’t know. I guess my previous generations just never gave up.”

“But you had working equipment. At the rate my equipment fails, I won’t be able to run experiments soon. What the hell am I supposed to do then?”

“That’s your choice.”

“I don’t care what I do. I just don’t want to be alone again.” he said, voice shaking. He too was beginning to break.

“Do something that makes you happy,” Ari encouraged, “something that gives you a reason to get up in the morning. You liked music when you were little.”

“But what if that doesn’t work? What if I stop liking music? What if I’m not able to be happy?”

For a moment, Ari returned to her original life. She remembered the anger she used to have, the anger towards the world. Towards herself. The anger drove her 35 miles over the speed limit on her deliveries, and her depression that secretly hoped for the worst.

Each reset of Ari had been more surprised than the last, not just by her sudden unfamiliar situation, but also by the fact that she was actually doing something with herself, something she could be proud of. Pride was an unfamiliar but welcome feeling, one she didn’t have to wait for success to feel.

Sure, the Aris had been lonely, and the first few generations had a rough time finding their footing. She’d never consider herself a happy person, but she learned to find joy in the little things. Joy from humor, from music, from facing new challenges and exceeding your limits. All things considered, Ari could have ended up much worse.

“You’ll be alright,” she said. “I know it.”

My AMD Framework 13 Upgrade Experience

jarbus@tutanota.com (jarbus) — Mon, 07 Jul 2025 08:55:26 -0400

Just upgraded my Framework Laptop 13 Mainboard from a Ryzen 7640U to a Ryzen AI HX 370. I run Arch Linux. Here’s my experience:

Upgrade was easy, but the cables for my WiFi card wouldn’t snap into place. There’s a plastic thing that I have to put on top of the connector to hold them in place, but it’s a massive pain to connect. Took me over 20 minutes to align and lock the cables into place, but I got it working.
The website said the HX 370 mainboard was only compatible with the RZ717 WiFi card, but the RZ717 was out of stock when I bought my mainboard. When the WiFi card was back in stock, they wouldn’t let me buy it without buying another mainboard. This is stupid, and I suspect just an innocent oversight on their part. I had to install my RZ616 to see if that was good enough.
On the new mainboard, the AMD RZ616 wifi card’s drivers would disconnect me for hours at a time, essentially rendering my laptop unusable—not good when you’re do coding interviews for a job. I reached out the customer support, they left me on read for about a week, so I just purchased an Intel AX210 non-vPro which works perfectly. That’s what’s great about a company like Framework: if their support is slow or unhelpful, you can fix your problems yourself.
- I posted about my customer support experience on Reddit, and even though the post didn’t blow up, their support lead reached out to make sure I was taken care of. They really do seem to care—sometimes things just fall through the cracks.
- Even though their customer support was slow, the initial response was still promising. I probably would have been taken care of, just after a few more weeks.
- I think they should stop selling the AMD RZ WiFi cards, given their issues. I wish their website just told me to get the Intel AX210 non-vPro when I was buying the laptop.
- It only took a few days to get the Intel AX210 non-vPro, which was nice. I’d be a much less happy customer if that also took a month.
Linux required some tinkering to get working, but ChatGPT was able to guide me through literally everything ¹. I was literally dropping in the SSD from the previous mainboard without reinstalling, so I had some issues booting from the boot partition and needed to update /etc/fstab, so not an issue on Framework’s end. Nothing a live USB couldn’t fix.

It’s so cool how I don’t need to wait for a screwdriver from Amazon to replace any of the parts. Keep the included screwdriver in your backpack and you’re always good to go.
Their documentation is excellent. Related: Take the Road Most Documented

Overall, these shortcomings are to be expected from a small, growing company. The important thing is that there was a documented, non-hacky solution I could find and implement on my own. The actual upgrade was easy, the HX 370 mainboard is fantastic, and everything now works flawlessly.

Overall: 9/10 experience, would buy again.

I could write volumes about how useful ChatGPT is for debugging Linux issues. I generally know what I’m doing, I just don’t know the syntax for most of the commands. AI has read every single Linux question on the internet and is really really useful for identifying problems and suggesting solutions. ↩︎

Everything Matters

jarbus@tutanota.com (jarbus) — Thu, 19 Jun 2025 09:35:49 -0400

Like many edgy teenagers, I once embraced nihilism—the belief that nothing matters, that existence is fundamentally meaningless. I could write a book on how strongly I subscribed to this area of thought, but that’s not the point of this post; you’ll have to trust that the following comes from someone who climbed out of the pit:

Everything matters. And that’s even scarier.

I still don’t believe in broad concepts like “meaning” or “fulfillment”. I believe in biological reactions that make something feel meaningful or fulfilling. Those reactions are what matters, and I think nihilism’s pull is strongest on those who lack them. But we can hack our minds.

We want to feel our lives our meaningful, that they are worth living. If your life lacks meaning, there is a good chance you believe one of the following:

Life inherently has no meaning, and nothing will change that.
Life will only have meaning once you have a family, dream job, etc.

There’s a secret third belief on the table, though—secret in the sense that everyone knows it, but nearly nobody believes it:

Life has as much meaning as you give it.

Each compliment you give to a stranger can make their day. A stranger complimented a jacket I had in September 2024, and I still remember it, because no stranger ever complimented my clothes before.

What’s crazy is that anyone can do this at any time, most just choose not to. It’s free, and usually has positive effects.

A few more things that have a huge (but often invisible) return on investment:

Remembering someone’s birthday (+1 for getting them a gift)
Telling someone you’re enjoying a gift they got you
Remembering small details that someone told you (the project they were working on, their best friend’s name, etc)
Getting someone flowers (for a partner, milestone, etc)
Complimenting strangers (without being creepy)
Smiling at people
Inviting people to grab dinner
Asking genuine questions when listening to someone talk about their passion
Checking if someone is OK

People live life believing their actions mostly don’t matter. They unnecessarily ghost, show up late, forget important details, back out of commitments, or fail to show up when they’re needed. Everyone acts unbothered, but we notice, and it hurts. Why wouldn’t the opposite be true?

I don’t know why I started trying to put effort into the little things—probably to further convince myself nothing matters and thus justify my position. But that’s not how our mind works. We feel meaning when we make others happy, especially others who feel forgotten, because in that moment, we matter to them. At least a little bit.

Low-Rank Factorizations are Indirect Encodings for Deep Neuroevolution

jarbus@tutanota.com (jarbus) — Thu, 29 May 2025 17:03:02 -0400

My latest paper is available on arxiv: Low Rank Factorizations are Indirect Encodings for Deep Neuroevolution.

The general idea is that we can search for stronger neural networks in a gradient-free fashion by restricting search to networks of low-rank. We show that it works well for language modeling and reinforcement learning tasks. It’s essentially a crossover between the following papers:

I’ll be presenting it virtually for the Neuroevolution@Work workshop at GECCO 2025.

To the Students

jarbus@tutanota.com (jarbus) — Mon, 21 Apr 2025 19:57:41 -0400

If all goes well, I’ll leave Brandeis University with a PhD degree in computer science this summer. Here are my parting thoughts for our remaining students, divided into technical and non-technical sections.

I write this out of self-interest; students after me will represent our university in years to come, and it’s in my best interest for them to be as effective as possible.

NON-TECHNICAL

Students often lack crucial skills. Technical students may lack communication skills, or social butterflies may lack depth. Some lack both. This doesn’t really matter, in my opinion, if they are willing to learn. Just like neural networks, they may be all over the place initially, but students who reflect on your feedback will (eventually) be worth more than those who never update their parameters. Don’t give up on those who listen and try; they often just need the right kind of mentorship.
Some students might be eager to learn, but have a poor approach to learning. It’s often easy to correct; if you ask how they approach a problem and why they’re approaching it that way, you’ll unearth the faults in their process. Most of the time, you don’t need to correct them—you can often describe how you’d approach their problem, and, perhaps after a few more failed attempts, they’ll try it your way, all on their own. This worked on me, anyway.
The packaging of a message matters more than its content. I can pretty much tell anyone anything if I phrase it with thought, respect, and tact. This takes time, effort, and lots of practice. If someone gets defensive and reacts negatively to your feedback, you likely could have phrased it better.
Effective communication is as important for your career as a top-tier publication—perhaps moreso. If you can’t describe why your work is significant, it probably doesn’t matter what your results are. Conversely, if your results are less stellar than you hoped, you can still sell them to a reader with the right discussion and analysis.
To communicate well, you need to read and write—a lot. More than you think. Some resources that have helped me:
- Eats, Shoots & Leaves: The Zero Tolerance Approach to Punctuation — Punctuation is like shell syntax for language. I didn’t realize how fundamental punctuation is for writing until my cousin recommended me this.
- How to Read a Paper — 2 page pdf that might save you days of your life.
- How to Read a Book — Don’t read books from front-to-end, and spend a decent amount of time understanding the table of contents. Read a lot, read widely.
- Write simply — Clear writing takes a lot of effort, but it’s worth it.
Insight comes from many sources. Read widely, and read weird things. I find computing history to be particularly illuminating, as it shows how technology changed over time.
Know your peers—this may be your last chance to become friends with a group of PhDs, most of whom will join different companies; this is an prime time to expand your network. Plus, the journey’s more bearable with company.
Care about your work. If you don’t care about your project, nobody will. Learn about the area you work in; learn about the technology you’re using; learn about why things are the way they are. Only two types of people have a chance at success anymore—those with passion, and those who crave status. I prefer those with passion.
If you don’t care about your work, make yourself care—life is better this way.
Take walks.
Learning yields an exponential return on investment. The bigger your knowledge graph, the easier it is to add edges and nodes. Reading papers in a field you know well can be effortless.
If you are still in the game, you haven’t lost. Don’t give into despair.
Be reliable. Respond to messages and fulfill your commitments, without needing a reminder. It’s nuts how many people can’t manage this.
Don’t burn bridges; people can change.
Be kind, but not soft.
Hold yourself and those around you to a realistic but high standard. A community aspiring to excellence can flourish; but those who see themselves part of a sinking ship are destined to drown.
Each lab should have at least one student who is friendly with the rest of the department. Students often suffer in silence, and a well-connected social graph is critical for detecting and addressing serious problems.

TECHNICAL

AI

Keep (relatively) up to date with the latest developments, even if they aren’t relevant to your current project; they may prove critical for your next one.
Understand the purpose of every layer in a foundation model, every stage of training, and how inputs and outputs are represented and sampled. Don’t skip these.
Learn about the memory hierarchy of a GPU. The original FlashAttention paper should make it clear why.
Learn why reinforcement learning is a pain; LLMs provide a good starting point.
Understand the intuition behind the PPO algorithm.
Read seminal papers, like the original attention is all you need paper and ResNet paper. They have insight and perspective which blogs and youtube videos lack.

NOT AI

Learn your tools, whether it’s Vim, Bash, PyTorch, VsCode, or even ChatGPT. They have many useful features that you likely aren’t aware of—Vim and Bash especially.
Test your code using specific input/output examples, especially the mathematical sections which may produce incorrect data instead of a crash. Seriously, test your code.
You probably should learn the basics of Tmux, especially if you use the cluster.
Equations are machines, and we can understand them from an intuitive, functional point of view—each component does something. They don’t need to be scary.
If you are confused by something, ask clarifying questions until you aren’t.
An hour well-spent with pen and paper can save days of debugging.
Keep a lab notebook, and date your entries—you won’t remember what you did two months from now. I just use a single text file.
Computing is an end, not a means.
Don’t use Windows.

Read Weird Things

jarbus@tutanota.com (jarbus) — Mon, 31 Mar 2025 10:54:19 -0400

I used to have a mental block against casually reading things that weren’t designed to be casually read, like old papers, textbooks, documentation, etc. If I was going to read for enjoyment, I’d read blogs or popular science/fiction books. Not sure why.

Over time, I’ve met people who read things I’d never think to read. I’ve gotten a lot of value from copying them, so I’m sharing my thoughts here.

Contents: Documentation | Primary Sources | Books and Textbooks

Documentation

For a programming assignment using a new language (erlang, I think), my original plan was to look up syntax and concepts as needed while I coded. Then, a friend told me he was reading the first few sections of the manual before writing any code. I’d have never thought of that; I’d only ever read a manual when I needed a specific piece of information.

I’ve started to do this with some of the internals of programs I use, and it’s really interesting; you learn not just how things work under the hood, but why they were designed that way, and the neat tricks that were used to make them work. Every time I open the bash manual, I learn five super useful features, four of which I promptly forget to use. Another great reference is The Architecture of Open Source Applications, which isn’t really documentation, but shows how different open source projects are designed. Reading these kinds of documents changes how I think about programming for the better.

Primary Sources

Another underrated source of interesting information is primary sources. Like, the original source. Wikipedia and blog posts are great for quickly learning something in a digestible way, but they don’t capture the context of the time. It’s one thing to look up the theory of evolution, but it’s another to read Darwin’s original work:

Although much remains obscure, and will long remain obscure, I can entertain no doubt, after the most deliberate study and dispassionate judgment of which I am capable, that the view which most naturalists entertain, and which I formerly entertained–namely, that each species has been independently created–is erroneous. I am fully convinced that species are not immutable; but that those belonging to what are called the same genera are lineal descendants of some other and generally extinct species, in the same manner as the acknowledged varieties of any one species are the descendants of that species.

This quote is so powerful to me; it’s like watching a pivotal moment in history from Darwin’s perspective. It’s also pretty fun to read the original perceptron paper from Rosenblatt:

In the meantime, the theory reported here clearly demonstrates the feasibility and fruitfulness of a quantitative statistical approach to the organization of cognitive systems. By the study of systems such as the perceptron, it is hoped that those fundamental laws of organization which are common to all information handling systems, machines and men included, may eventually be understood.

It’s crazy to imagine that someone had to prove statistical approaches to AI were feasible; we take it for granted now.

Books and Textbooks

Textbooks are fun, too. As my family can attest, I hated biology in school, but Karpathy inspired me to give it a second chance when he randomly started reading biology textbooks in his free time. Turns out I don’t hate biology either–I just hate people telling me what to do, as you can probably tell from this website.

You don’t need to read textbooks in order. You can just pick a random chapter and read it. There might be a chapter that builds on a previous chapter, but if you don’t understand it, you can just read its dependency–chances are it only requires reading a section or two, not every preceeding page. The figures are also really helpful for getting the gist of ideas quickly.

After years of reading books front-to-back until I abandoned or finished them, jumping around a book feels so freeing. Especially fiction–sure, you don’t get the full story, but you get the vibe, and can usually infer missing details.

From REINFORCE to R1: an Abridged Genealogy of Reinforcement Learning

jarbus@tutanota.com (jarbus) — Fri, 21 Feb 2025 19:58:43 -0500

Starting from REINFORCE, the original deep reinforcement learning algorithm, we will trace the evolution of policy gradient methods to the Group Relative Policy Optimization algorithm used to train Deepseek r1.

This post ignores the LLM side of things, less-related developments in RL, and most of the equations used for these algorithms, but captures the essence and intuition of the RL-timeline without wasting your time. This is all self-study, so feel free to send me any corrections/suggestions.¹

Reinforcement Learning 101
REINFORCE
From REINFORCE to Actor-Critic
From Actor-Critic to Advantage Actor-Critic (A2C)
From Advantage Actor-Critic to Proximal Policy Optimization (PPO)
From Proximal Policy Optimization to Group Relative Policy Optimization

Reinforcement Learning 101

Reinforcement learning is a subfield of machine learning that can learn without human data. Instead, an agent can learn through interacting with its environment by taking actions and receiving rewards based on its actions. The agent wants to maximize its reward.

Specifically, an environment at time $t$ with state $s_t$ sends an observation, $o_t$, to an agent, which produces probabilities to select an action, $a_t$, based on the observation. The environment transitions to a new state, $s_{t+1}$, and the agent receives a reward, $r_{t+1}$, based on the action it took. The agent’s goal is to learn a policy, $\pi(a_t|o_t)$ that maximizes the expected sum of rewards, $R = \sum_{t=0}^T \gamma^t r_t$, where $T$ is the number of steps in the trajectory (sequence of states, actions, and rewards from the start to the end of an episode), and $\gamma$ is a discount factor, which reduces the value of rewards the farther into the future they are. There are ways to perform reinforcement learning without neural networks, but we will focus on deep reinforcement learning, where all policy actions are determined by a neural network with parameters $\theta$.

REINFORCE

The original policy gradient, and the basis for all policy gradient methods, is the REINFORCE algorithm. REINFORCE samples trajectories from the environment, which it uses to directly compute the gradient of the policy:

$$ \theta_{t+1} = \theta_t + \alpha R_t \sum_{t=0}^T \nabla_\theta \log \pi(a_t|o_t) $$

where $\alpha$ is the learning rate, and $R_t$ is the remaining discounted reward–also called a return–from the current state given by $R_t = \sum_{t’=t}^T \gamma^{t’-t} r_{t’}$.

REINFORCE simply “reinforces” the actions proportional to the reward they receive. This is the most intuitive reinforcement learning algorithm, but it has high variance, which leads us to actor critics.

I struggled with the concept of variance in RL for a while, but it’s essentially the amount of randomness in the reward signal. When you are trying to learn purely from experience, rewards from a given state can vary wildly between trajectories depending on how the environment plays out, so we need a lot of samples to get a rough idea of what the “true” reward is for a given state. Actor-critic methods reduce this variance by using another neural network called a “critic” to estimate the value of a state. It turns out that using a consistent estimator, like a neural network, stabilizes learning, as the learning signal doesn’t fluctuate as much between samples.

From REINFORCE to Actor-Critic

As mentioned, actor-critic methods use neural networks to estimate reward. This results in more stable learning compared to using trajectories directly. The actor is just the policy from REINFORCE, and the critic is a new, separate network that estimates the return given the current policy. We won’t cover the training of the critic here, but just assume it’s just a neural network that takes observations $o_i$ as input and estimates the return $R$ for the actor.

At it’s core, actor-critic is just REINFORCE, but $R_t$ is computed by another neural network, which is trained using the actual returns from the environment.

As a bonus, we can now update actors without requiring the full trajectory, since we are using an estimate instead of the true return.

From Actor-Critic to Advantage Actor-Critic (A2C)

The next step in the evolution of policy gradient methods is the Advantage Actor-Critic (A2C) algorithm. Now we introduce the advantage function, which is the difference between the return and the value estimate of the state:

$$ A_t = r + \gamma V(o_{t+1}) - V(o_t) $$

where $V(o_i)$ is the value estimate of the state computed by our critic at time $i$.

The advantage function tells us how much better our action was compared to the expected action for that state, which allows us to reinforce actions based on how much they improve our expected reward. Conversely, this also discourages actions based on how much they decrease our expected reward. This is unlike prior methods, which might only positively (or negatively) reinforce actions based on the absolute reward they receive, not the relative reward. Without the advantage function, given an environment which only produces positive rewards, we might only reinforce good actions, without ever penalizing bad ones.

Now, the policy gradient is (roughly) computed as:

$$ \theta_{t+1} = \theta_t + \alpha A_t \sum_{t=0}^T \nabla_\theta \log \pi(a_t|o_t) $$

where $A_t$ is the advantage function computed using the critic network. This further stabilizes learning.

From Advantage Actor-Critic to Proximal Policy Optimization (PPO)

Skipping over many other developments, we jump directly from A2C to Proximal Policy Optimization (PPO). PPO uses the advantage function to update the policy, but just by a little bit, by introducing a clipping term that limits the size of the update. This prevents the policy from changing its behavior too much in a given update, which might otherwise destabilize training and cause the policy to forget some learned behaviors.

PPO prevents behaviors from changing too much by ensuring that the probability of taking an action in a given state doesn’t change too much, using a ratio of the new policy to the old policy, $\frac{\pi_\theta(a_t|o_t)}{\pi_{\theta_{\text{old}}}(a_t|o_t)}$. When the probabilities don’t change, $ \frac{\pi_\theta(a_t|o_t)}{\pi_{\theta_{\text{old}}}(a_t|o_t)}= \mathbf{1}$, but if the probabilities change, $\frac{\pi_\theta(a_t|o_t)}{\pi_{\theta_{\text{old}}}(a_t|o_t)}$ will range between 0 and $\infty$. PPO clips the ratio to a range of $[1-\epsilon, 1+\epsilon]$, where $\epsilon$ is a hyperparameter we choose between 0 and 1 to determine the max percentage by which a probability can change.

The PPO update is then:

$$ \theta_{t+1} = \theta_t + \alpha \nabla_\theta \mathbb{E}_t \left[ \min( \frac{\pi_\theta(a_t|o_t)}{\pi_{\theta_{\text{old}}}(a_t|o_t)} A_t, \text{clip}(\frac{\pi_\theta(a_t|o_t)}{\pi_{\theta_{\text{old}}}(a_t|o_t)} , 1-\epsilon, 1+\epsilon) A_t) \right] $$

This adds a lot of terms to our update, but hopefully the jump from A2C to PPO is visible.

From Proximal Policy Optimization to Group Relative Policy Optimization

Finally, we arrive at Group Relative Policy Optimization (GRPO), the algorithm used to train Deepseek r1. This algorithm is used strictly in language modeling, so the ability to update before the end of a sequence doesn’t really matter; all that matters is the stability and efficiency. Unlike prior algorithms, GRPO’s main contribution isn’t that it further stabilizes learning, but that it reduces memory and computation requirements drastically.

Turns out, for language modeling, we don’t actually need the critic network, which saves us 50% of the memory and computation required to run reinforcement learning. We can just take all the other non-critic enhancements since REINFORCE and apply them in large batches , which is what GRPO does.

$$ \frac{1}{G}\sum_{i=1}^G \frac{1}{|o_i|} \sum_{t=1}^{|o_i|} \min \left[ \frac{\pi_\theta(o_{i,t}|q, o_{i,\lt t})}{\pi_{\theta_{\text{old}}}(o_{i,t}|q, o_{i,\lt t})} \hat{A}_{i,t}, \text{clip}\left(\frac{\pi_\theta(o_{i,t}|q, o_{i,\lt t})}{\pi_{\theta_{\text{old}}}(o_{i,t}|q, o_{i,\lt t})}, 1-\epsilon, 1+\epsilon\right) \hat{A}_{i,t} \right] - \beta D_{\text{KL}}\left[\pi_\theta \, \| \, \pi_{\text{ref}} \right] $$

You can see the lineage from PPO here. $q$ refers to a question, and $o$ here refers to an output sequence. $G$ is just the number of outputs in a a group. The main difference from PPO is that, instead of computing the advantage $A_t$ using a critic network, we now compute it within a group of sequences:

$$ \hat{A}_{i,t} = \frac{r_{i,t} - \text{mean}(\mathbf{r})}{\text{std}(\mathbf{r})} $$

where $\mathbf{r}$ is the set of returns from the group of sequences. This replaces our neural estimate with an empirical estimate, while maintaining the benefits of the critic’s stability.

There’s a KL-divergence term at the end $\beta D_{\text{KL}}\left[\pi_\theta , | , \pi_{\text{ref}}\right]$, which is ensures that the policy doesn’t change too much from the fine-tuned policy, $\pi_{\text{ref}}$.

I know I’m pretty loose with the notations–I prefer this to adding a ton of extra symbols that don’t contribute much. ↩︎

Eloquence and Wit from Will Durant

jarbus@tutanota.com (jarbus) — Wed, 12 Feb 2025 13:42:52 -0500

I just finished The Story of Civilization, Vol. 1 (1976) by Will Durant. It’s a fantastic read; Durant is eloquent, witty, and surprisingly relevant today. Below are some of my personal highlights of his writing, mixed with some quotes-of-quotes I found interesting.

Introduction

“the provincialism of our traditional histories, which began with Greece and summed up Asia in a line, has become no merely academic error, but a possibly fatal failure of perspective and intelligence”

“Civilization is an interlude between ice ages”

“…the decay of leadership through the infertility of the able”

“The moment man begins to take thought of the morrow he passes out of the Garden of Eden into the vale of anxiety; the pale cast of worry settles down upon him, greed is sharpened, property begins, and the good cheer of the “thoughtless” native disappears.”

“Of what are you thinking?” Peary asked one of his Eskimo guides. “I do not have to think,” was the answer; “I have plenty of meat.” Not to think unless we have to—there is much to be said for this as the summation of wisdom"

“If man began with speech, and civilization with agriculture, industry began with fire”

“This primitive skill displayed itself proudly in the art of weaving. Here, too, the animal showed man the way. The web of the spider, the nest of the bird, the crossing and texture of fibres and leaves in the natural embroidery of the woods, set an example so obvious that in all probability weaving was one of the earliest arts of the human race”

“Cattle were a convenient standard of value and medium of exchange among hunters and herders; they bore interest through breeding, and they were easy to carry, since they transported themselves”

Communism

“When Turner told a Samoan about the poor in London the “savage” asked in astonishment: “How is it? No food? No friends? No house to live in? Where did he grow? Are there no houses belonging to his friends?” The hungry Indian had but to ask to receive; no matter how small the supply was, food was given him if he needed it; “no one can want food while there is corn anywhere in the town.””

“White travelers in Africa before the advent of civilization noted that a present of food or other valuables to a “black man” was at once distributed; so that when a suit of clothes was given to one of them the donor soon found the recipient wearing the hat, a friend the trousers, another friend the coat”

“What is extremely surprising,” reports a missionary “is to see them treat one another with a gentleness and consideration which one does not find among common people in the most civilized nations”

“Loskiel reported some Indian tribes of the northeast as “so lazy that they plant nothing themselves, but rely entirely upon the expectation that others will not refuse to share their produce with them”

“Darwin thought that the perfect equality among the Fuegians was fatal to any hope of their becoming civilized; or, as the Fuegians might have put it, civilization would have been fatal to their equality”

“Communism brought a certain security to all who survived the diseases and accidents due to the poverty and ignorance of primitive society; but it did not lift them out of that poverty”

“It was a great moral improvement when men ceased to kill or eat their fellow men, and merely made them slaves”

Government

“The Dyaks had no other government than that of each family by its head; in case of strife they chose their bravest warrior to lead them, and obeyed him strictly; but once the conflict was ended they literally sent him about his business”

“Time sanctifies everything; even the most arrant theft, in the hands of the robber’s grandchildren, becomes sacred and inviolable property”

“Though men are naturally gullible they are also naturally obstinate, and power, like taxes, succeeds best when it is invisible and indirect”

“The old Russian Government established courts of law in the Aleutian Islands, but in fifty years those courts found no employment. “Crime and offenses,” reports Brinton, “were so infrequent under the social system of the Iroquois that they can scarcely be said to have had a penal code.” Such are the ideal—perhaps the idealized—conditions for whose return the anarchist perennially pines.”

“It is the routine that keeps men sane; for if there were no grooves along which thought and action might move with unconscious ease, the mind would be perpetually hesitant, and would soon take refuge in lunacy.”

“To violate law is to win the admiration of half the populace, who secretly envy anyone who can outwit this ancient enemy; to violate custom is to incur almost universal hostility”

“The first stage in the evolution of law is personal revenge… The second step toward law and civilization in the treatment of crime was the substitution of damages for revenge… It is but a step from settling disputes and punishing offenses to making some effort to prevent them.”

“Primitive punishments are cruel, because primitive society feels insecure; as social organization becomes more stable, punishments become less severe”

Women

“woman, apart from her biological disabilities, was almost the equal of man in stature, endurance, resourcefulness and courage; she was not yet an ornament, a thing of beauty or a sexual toy; she was a robust animal, able to perform arduous work for long hours, and, if necessary to fight to the death for her children or her clan.”

“It was she who developed the home, slowly adding man to the list of her domesticated animals, and training him in those social dispositions and amenities which are the psychological basis and cement of civilization”

“Since no society can exist without order and no order without regulation, we may take it as a rule of history that the power of custom varies inversely as the multiplicity of laws, much as the power of instinct varies inversely as the multiplicity of thoughts”

“The first task of those customs that constitute the moral code of a group is to regulate the relations of the sexes, for these are a perennial source of discord, violence, and possible degeneration.”

“There is no record of women objecting to marriage by purchase; on the contrary, they took keen pride in the sums paid for them, and scorned the woman who gave herself in marriage without a price; they believed that in a “love-match” the villainous male was getting too much for nothing”

“A Maori mother wailing loudly bitterly cursed the youth who had eloped with her daughter until he presented her with a blanket. “That was all I wanted,” she said; “I only wanted to get a blanket, and therefore made this noise”

“Marriage was a profitable partnership, not a private debauch; it was a way whereby a man and a woman, working together might be more prosperous than if each worked alone”

“as Anacharsis put it among the Greeks, if one were to bring together all customs considered sacred by some group, and were then to take away all customs considered immoral by some group, nothing would remain”

“The institutions, conventions, customs and laws that make up the complex structure of a society are the work of a hundred centuries and a billion minds; and one mind must not expect to comprehend them in one lifetime, much less in twenty years”

“Greed, acquisitiveness, dishonesty, cruelty and violence were for so many generations useful to animals and men that not all our laws, our education, our morals and our religions can quite stamp them out; some of them, doubtless, have a certain survival value even today”

“The Yakuts have been known to eat forty pounds of meat in one day; and similar stories, only less heroic, are told of the Eskimos and the natives of Australia”

“for to many a primitive mind no argument is settled until one of the disputants is dead”

“In several tribes no woman would marry a man who had not killed some one, in fair fight or foul; hence the practice of headhunting, which survives in the Philippines today”

Morals

“primitive hospitality, in many tribes, went to the extent of offering to the traveler the wife or daughter of the host. To decline such an offer was a serious offense, not only to the host but to the woman; these are among the perils faced by missionaries”

“Almost all groups agree in holding other groups to be inferior to themselves”

“Moral progress in history lies not so much in the improvement of the moral code as in the enlargement of the area within which it is applied”

“men are more easily ruled by imagination than by science”

“The dwarfs of the Cameroon recognized only malevolent deities, and did nothing to placate them, on the ground that it was useless to try”

“In the Molucca Islands blossoming trees were treated as pregnant; no noise, fire, or other disturbance was permitted to mar their peace; else, like a frightened woman, they might drop their fruit before time.”

“It is the tendency of gods to begin as ogres and to end as loving fathers; the idol passes into an ideal as the growing security peacefulness and moral sense of the worshipers pacify and transform the features of their once ferocious deities. The slow progress of civilization is reflected in the tardy amiability of the gods.”

“—a desired action was suggested to the deities by a partial or imitative performance of the action by men. To make rain fall some primitive magicians poured water out upon the ground, preferably from a tree. The Kaffirs, threatened by drought, asked a missionary to go into the fields with an opened umbrella… In the Babar Archipelago the would-be mother fashioned a doll out of red cotton, pretended to suckle it, and repeated a magic formula; then she sent word through the village that she was pregnant, and her friends came to congratulate her; only a very obstinate reality could refuse to emulate this imagination.”

“The favorite object of primitive tabu was woman.”

“In the end terrestrial forces prevail; morals slowly adjust themselves to economic invention, and religion reluctantly adjusts itself to moral change. The moral function of religion is to conserve established values, rather than to create new ones”

Language

“Gesture seems primary, speech secondary, in the earlier transmission of thought; and when speech fails, gesture comes again to the fore.”

“Gesture was so prominent in some Indian languages that the Arapahos, like some modern peoples, could hardly converse in the dark”

“Nothing surprises the natural man so much as the ability of Europeans to communicate with one another over great distances, by making black scratches upon a piece of paper”

“Perhaps science, like civilization in general, began with agriculture; geometry, as its name indicates, was the measurement of the soil”

“Damara natives would not exchange two sheep for four sticks, but willingly exchanged, twice in succession, one sheep for two sticks”

“Perhaps we should employ another gender here, for probably the first doctors were women; not only because they were the natural nurses of the men, nor merely because they made midwifery rather than venality the oldest profession, but because their closer connection with the soil gave them a superior knowledge of plants, and enabled them to develop the art of medicine as distinct from the magic-mongering of the priests.”

“It is astonishing how many cures primitive doctors effected despite their theories of disease”

Appearance

“Primitive man seldom thinks of selecting women because of what we should call their beauty; he thinks rather of their usefulness, and never dreams of rejecting a strong-armed bride because of her ugliness”

“There is,” says Georg, “no part of the body that has not been perfected, decorated, disfigured, painted, bleached, tattooed, reformed, stretched or squeezed, out of vanity or desire for ornament.””

“Clothing was apparently in its origins, a form of ornament, a sexual deterrent or charm rather than an article of use against cold or shame”

“Cook had said of them, timelessly, they were “content to be naked, but ambitious to be fine.””

“The Queen of the Wabunias on the Congo wore a brass collar weighing twenty pounds; she had to lie down every now and then to rest. Poor women who were so unfortunate as to have only light jewelry imitated carefully the steps of those who carried great burdens of bedizenment.”

“The taut string of the bow became the origin of a hundred instruments from the primitive lyre to the Stradivarius violin and the modern pianoforte”

“Bone tools—pins, anvils, polishers, etc.—were now added to those of stone; and art appeared in the form of crude engravings on the rocks, or simple figurines in high relief, mostly of nude women.”

“In the ruins of a neolithic mine at Brandon, England, eight worn picks of deer horn were found, on whose dusty surfaces were the finger-prints of the workmen who had laid down those tools ten thousand years ago.”

Sumeria

“History so soon began its tragic alternance of art and war.”

“The form of these poems is unexpectedly first-personal, and the style does not please the sophisticated ear; but across the four thousand years that separate us from the Sumerian singer we feel the desolation of his city and his people”

“The wars were waged frankly for commercial routes and goods, without catchwords as a sop for idealists”

“King Manish-tusu of Akkad announced frankly that he was invading Elam to get control of its silver mines, and to secure diorite stone to immortalize himself with statuary—the only instance known of a war fought for the sake of art”

“The best element in this code was a plan for avoiding litigation: every case was first submitted to a public arbitrator whose duty it was to bring about an amicable settlement without recourse to law. It is a poor civilization from which we may not learn something to improve our own”

“As early as 2000 B.C. Sumerian historians began to reconstruct the past and record the present for the edification of the future; portions of their work have come down to us not in the original form but as quotations in later Babylonian chronicles.”

Egypt

“Nearby, the Sphinx, half lion and half philosopher, grimly claws the sand, and glares unmoved at the transient visitor and the eternal plain”

“to bring these vast stones six hundred miles, to raise some of them, weighing many tons, to a height of half a thousand feet, and to pay or even to feed, the hundred thousand slaves who toiled for twenty years on these Pyramids…”

“With his knowledge of Greek, and the eleven letters made out from the obelisk, Champollion, after more than twenty years of labor deciphered the whole inscription, discovered the entire Egyptian alphabet, and opened the way to the recovery of a lost world. It was one of the peaks in the history of history”

““All the world fears Time,” says an Arab proverb, “but Time fears the Pyramids.””

“this alternation between centralized and decentralized power is one of the cyclical rhythms of history, as if men tired alternately of immoderate liberty and excessive order”

“But Hatshepsut set this high-destined youngster aside, assumed full royal powers, and proved herself a king in everything but gender.”

“they made everything from boats and carriages, chairs and beds, to handsome coffins that almost invited men to die.”

“If,” says Peschel, “we compare the technical inventory of the Egyptians with our own, it is evident that before the invention of the steam-engine we scarcely excelled them in anything.””

“A chalk tablet in the British Museum contains a chief workman’s record of forty-three workers, listing their absences and their causes—“ill,” or “sacrificing to the god,” or just plain “lazy.””

“[The] quantity of radishes, garlic and onions consumed by the workmen who built it; these things, too, had to have their immortality”

“A Greek tradition reports a great revolt in Egypt, in which the slaves captured a province, and held it so long that time, which sanctions everything, gave them legal ownership of it.”

“Machinery was rare because muscle was cheap”

“One tomb inscription describes its occupant as “Overseer of the Cosmetic Box, Overseer of the Cosmetic Pencil, Sandal Bearer to the King, doing in the matter of the King’s sandals to the satisfaction of his Law”

““No people, ancient or modern,“said Max Müller, “has given women so high a legal status as did the inhabitants of the Nile Valley.”

“Hatshepsut and Cleopatra rose to be queens, and ruled and ruined like kings.”

“shame is a child of custom rather than of nature,”

“Copy-books survive from the days of the Empire with the corrections of the masters still adorning the margins; the abundance of errors would console the modern schoolboy.”

“Khekheperre-Sonbu, a savant of the reign of Senusret II, about 2150B.C., complained that all things had long since been said, and nothing remained for literature except repetition.”

“The sign for 1,000,000 was a picture of a man striking his hands above his head, as if to express amazement that such a number should exist”

“Egyptian geometry measured not only the area of squares, circles and cubes, but also the cubic content of cylinders and spheres; and it arrived at 3.16 as the value of π. We enjoy the honor of having advanced from 3.16 to 3.1416 in four thousand years.”

“The wisdom of the Egyptians was a proverb with the Greeks, who felt themselves children beside this ancient race.”

“In part such literature represents one of those interludes, like our own moral interregnum, in which thought has for a time overcome belief, and men no longer know how or why they should live. Such periods do not endure; hope soon wins the victory over thought; the intellect is put down to its customary menial place, and religion is born again, giving to men the imaginative stimulus apparently indispensable to life and work”

“It is one of the tragedies of history that Ikhnaton, having achieved his elevating vision of universal unity, was not satisfied to let the noble quality of his new religion slowly win the hearts of men. He was unable to think of his truth in relative terms; the thought came to him that other forms of belief and worship were indecent and intolerable. Suddenly he gave orders that the names of all gods but Aton should be erased and chiseled from every public inscription in Egypt; he mutilated his father’s name from a hundred monuments to cut from it the word Amon; he declared all creeds but his own illegal, and commanded that all the old temples should be closed.”

“More and more the people starved in order that the gods might eat.”

Babylon

“Despite the secular quality of his laws Hammurabi was clever enough to gild his authority with the approval of the gods.”

“Nearly all the bricks so far recovered from the site of Babylon bear the proud inscription: “I am Nebuchadrezzar, King of Babylon.””

“What modern city is so well governed that it would dare to offer such reimbursements to the victims of its negligence? Has the law progressed since Hammurabi, or only increased and multiplied?”

“We should remember, however, that the defeat of Chaos is only a myth”

“Even Alexander, who was not above dying of drinking, was shocked by the morals of Babylon”

“we cannot judge a civilization from such fragments as the ocean of time has thrown up from the wreckage of Babylon”

“These Babylonian libraries are lost; but one of the greatest of them, that of Borsippa, was copied and preserved in the library of Ashurbanipal, whose 30,000 tablets are the main source of our knowledge of Babylonian life…the avowed purpose of Ashurbanipal’s library was to preserve the literature of Babylonia from oblivion”

“But this almost secularized science found itself helpless before the demand of the people for supernatural diagnosis and magical cures. Sorcerers and necromancers were more popular than physicians, and enforced, by their influence with the populace, irrational methods of treatment”

“For barbarism is always around civilization, amid it and beneath it, ready to engulf it by arms, or mass migration, or unchecked fertility.”

“he hunted nations as well as animals:”

“His son Sennacherib put down revolts in the distant provinces adjoining the Persian Gulf, attacked Jerusalem and Egypt without success, sacked eighty-nine cities and 820 villages, captured 7,200 horses, 11,000 asses, 80,000 oxen, 800,000 sheep, and 208,000 prisoners; the official historian, on his life, did not understate these figures.”

Assyria

“Probably it was in part by their reputation for mercy to prisoners of war that Alexander and Caesar undermined the morale of the enemy, and conquered the Mediterranean world.”

“All in all, the Assyrian government was primarily an instrument of war. For war was often more profitable than peace; it cemented discipline, intensified patriotism, strengthened the royal power, and brought abundant spoils and slaves for the enrichment and service of the capital.”

“As we read such pages we become reconciled to our own mediocrity.”

“but they had brought into Assyria, as captives, millions of destitute aliens who bred with the fertility of the hopeless, destroyed all national unity of character and blood, and became by their growing numbers a hostile and disintegrating force in the very midst of the conquerors”

“Like all early voyagers, and some old languages, they made scant distinction between trade and treachery, commerce and robbery; they stole from the weak, cheated the stupid, and were honest with the rest”

Judea

“Next to the promulgation of the “Book of Law,” the building of the Temple was the most important event in the epic of the Jews. It not only gave Yahveh a home, but it gave Judea a spiritual center and capital, a vehicle of tradition, a memory to serve as a pillar of fire through centuries of wandering over the earth.”

“He is talkative, and likes to make long speeches; but he is shy, and will not allow men to see anything of him but his hind parts. Never was there so thoroughly human a god.”

“He will have no pacifist nonsense; he knows that even a Promised Land can be won, and held, only by the sword;”

“it will take centuries of military defeat, political subjugation, and moral development, to transform him into the gentle and loving Father of Hillel and Christ”

“Since poverty is created by wealth, and never knows itself poor until riches stare it in the face, so it required the fabulous fortune of Solomon to mark the beginning of the class war in Israel”

“It is true that Amos dulls the edge of his idealism by putting into the mouth of his god a Mississippi of threats whose severity and accumulation make the reader sympathize for a moment with the drinkers of wine and the listeners to music.”

“Destroyed and rebuilt, destroyed and rebuilt, Jerusalem rises again, symbol of the vitality and pertinacity of an heroic race. The Jews, who are as old as history, may be as lasting as civilization”

“the scenes of our youth, like the past, are always beautiful if we do not have to live in them again”

Persia

“So far as we can visualize him through the haze of legend, he was the most amiable of conquerors, and founded his empire upon generosity. His enemies knew that he was lenient, and they did not fight him with that desperate courage which men show when their only choice is to kill or die”

“The first principle of his policy was that the various peoples of his empire should be left free in their religious worship and beliefs, for he fully understood the first principle of statesmanship—that religion is stronger than the state.”

“He publicly scoffed at the Egyptian religion, and plunged his dagger derisively into the bull revered by the Egyptians as the god Apis; he exhumed mummies and pried into royal tombs regardless of ancient curses; he profaned the temples and ordered their idols to be burned. He thought in this way to cure the Egyptians of superstition; but when he was stricken with illness—apparently epileptic convulsions—the Egyptians were certain that their gods had punished him, and that their theology was now confirmed beyond dispute.”

“He had hoped to govern in peace, but it is the fatality of empire to breed repeated war”

“Persian councils never undertook serious discussions of policy when sober—though they took care to revise their decisions the next morning”

“For it was a proud boast of Persia that its laws never changed, and that a royal promise or decree was irrevocable.”

“The generosity and kindliness enjoined by the Master did not apply, in practice,to infidels—i.e., foreigners; these were inferior species of men, whom Ahura-Mazda had deluded into loving their own countries only in order that they should not invade Persia.”

“The Persians, says Herodotus, “esteem themselves to be far the most excellent of men in every respect… [they] believe that other nations approach to excellence according to their geographical proximity to Persia, but that they are the worst who live farthest from them””

“All in all [Zoroastrianism] was a splendid religion, less warlike and bloody, less idolatrous and superstitious, than the other religions of its time; and it did not deserve to die so soon”

“But humanity loves poetry more than logic, and without a myth the people perish.”

“It is a testimony to the character of the Persians that whereas any one could hire Greeks to fight Greeks, it was rare indeed that a Persian could be hired to fight Persians.”

“They resorted more frequently to spells than to drugs, on the ground that the spells, though they might not cure the illness, would not kill the patient—which was more than could be said for the drugs”

“When the two armies met at Issus Alexander had no more than 30,000 followers; but Darius, with all the stupidity that destiny could require, had chosen a field in which only a small part of his multitude couldfight at one time. When the slaughter was over the Macedonians had lost some 450, the Persians 110,000 men, most of these being slain in wild retreat;”

India

“Nothing should more deeply shame the modern student than the recency and inadequacy of his acquaintance with India.”

“our knowledge of the past is an occasional gap in our ignorance”

“They were too primitive to be hypocrites: they subjugated India without pretending to elevate it”

“They wanted land, and pasture for their cattle; their word for war said nothing about national honor, but simply meant “a desire for more cows”

“Marriage might be entered into by forcible abduction of the bride, by purchase of her or by mutual consent. Marriage by consent, however was considered slightly disreputable; women thought it more honorable to be bought and paid for, and a great compliment to be stolen”

““In the whole world,” said Schopenhauer “there is no study so beneficial and so elevating as that of the Upanishads. It has been the solace of my life—it will be the solace of my death.””

“The Jains, having emptied the sky of God, soon peopled it again with the deified saints of Jain history and legend.”

“Buddha was convinced that pain so overbalanced pleasure in human life that it would be better never to have been born”

“One of his disciples concluded that Buddha would approve of suicide, but Buddha reproved him; suicide would be useless, since the soul, unpurified, would be reborn in other incarnations.”

“There is nothing stranger in the history of religion than the sight of Buddha founding a worldwide religion, and yet refusing to be drawn into any discussion about eternity, immortality, or God.”

“The simplicity of their laws and their contracts is proved by the fact that they seldom go to law.”

“The government made no pretense to democracy and was probably the most efficient that India has ever had”

“In the political sense Ashoka had failed; in another sense he had accomplished one of the greatest tasks in history. Within two hundred years after his death Buddhism had spread throughout India, and was entering upon the bloodless conquest of Asia. If to this day, from Kandy in Ceylonto Kamakura in Japan, the placid face of Gautama bids men be gentle to one another and love peace, it is partly because a dreamer, perhaps a saint, once held the throne of India.”

“The Mohammedan Conquest of India is probably the bloodiest story in history. It is a discouraging tale, for its evident moral is that civilization is a precarious thing, whose delicate complex of order and liberty, culture and peace may at any time be overthrown by barbarians invading from without or multiplying within.”

“The Hindus had allowed their strength to be wasted in internal division and war; they had adopted religions like Buddhism and Jainism, which unnerved them for the tasks of life; they had failed to organize their forces for the protection of their frontiers and their capitals, their wealth and their freedom, from the hordes of Scythians, Huns, Afghans and Turks hovering about India’s boundaries and waiting for national weakness to let them in. For four hundred years (600-1000 A.D.) India invited conquest; and at last it came”

“he expressed his admiration for the architecture of the great shrine, judged that its duplication would cost one hundred million dinars and the labor of two hundred years, and then ordered it to be soaked with naphtha and burnt to the ground”

“He was generous, expending vast sums in alms; he was affable to all, but especially to the lowly; “their little offerings,” says a Jesuit missionary “he used to accept with such a pleased look, handling them and putting them in his bosom, as he did not do with the most lavish gifts of the nobles.””

“Deeper than these interests was his penchant for speculation. This well-nigh omnipotent emperor secretly yearned to be a philosopher-much as philosophers long to be emperors, and cannot comprehend the stupidity of Providence in withholding from them their rightful thrones.”

“Jehangir was not so much a mediocrity as an able degenerate.”

“The Emperor himself,in his last years, began to realize that by the very narrowness of his piety he had destroyed the heritage of his fathers. His deathbed letters are pitiful documents.”

“In the seventh century the Arabs captured Persia and Egypt, and thereafter trade between Europe and Asia passed through Moslem hands; hence the Crusades, and Columbus.”

“Death was the penalty for any of a great variety of crimes, such as housebreaking, damage to royal property or theft on a scale that would now make a man a very pillar of society”

““Better thine own work is, though done with fault,” said the Bhagavad-Gita, “than doing others’ work, even excellently.””

“Passages of the Code advocate an enlightened gentleness to women: they are not to be struck “even with a flower”; they are not to be watched too strictly, for then their subtlety will find a way to mischief; and if they like fine raiment-it is wise to indulge them, for “if the wife be not elegantly attired, she will not exhilarate her husband,” where as when “a wife is gaily adorned, the whole house is embellished.””

“In so warm a climate clothing was a superfluity, and beggars and saints bridged the social scale in agreeing to do without it. One southern caste, like the Canadian Doukhobors, threatened to migrate if its members were compelled to wear clothing.”

““I have had before me,” says a British judge in India,“hundreds of cases in which a man’s property liberty and life depended upon his telling a lie, and he has refused to tell it”

“The growth of Buddhism and monasticism in the first year of our era sapped the manhood of India, and conspired with political division to leave India open to easy conquest.”

“Buddhism, like Christianity, won its greatest triumphs outside the land of its birth; and it won them without shedding a drop of blood”

“the cynical view of the matter is that the Brahmans believed that cows should never be slaughtered, that insects should never be injured, and that widows should be burned alive”

“The secret of polytheism is the inability of the simple mind to think in impersonal terms; it can understand persons more readily than forces, wills more easily than laws”

“Since the real problem of life is not suffering but undeserved suffering, the religion of India mitigates the human tragedy by giving meaning and value to grief and pain”

“Doubtless when India was wealthy, sceptics were numerous, for humanity doubts its gods most when it prospers, and worships them most when it is miserable.”

“They relied upon the conservatism of the poor to preserve the orthodox religion, and they were not disappointed.”

“It is India that gave us the ingenious method of expressing all numbers by ten symbols, each receiving a value of position as well as an absolute value; a profound and important idea which appears so simple to us now that we ignore its true merit. But its very simplicity the great ease which it has lent to all computations, puts our arithmetic in the first rank of useful inventions”

“In the time of Yuan Chwang Hindu medical treatment began with a seven-day fast; in this interval the patient often recovered;”

“The Vedas” he says, simply, “are an authority, since the author of them knew the established truth.” After which he proceeds without paying any attention to the Vedas.”

“Such enlightenment cannot be won at a stroke; the aspirant must move towards it step by step, and no stage of the process can be understood by anyone who has not passed through the stages before it;”

“our Way of perceiving will forever be inextricably mingled with the thing perceived.”

“As our energies tire in the daily struggle against impartial Nature and hostile Time, we look with more tolerance upon Oriental philosophies of surrender and peace.”

“Though she passes through the Ordeal of Fire to prove her innocence, he sends her away to a forest hermitage with that bitter trick of heredity whereby one generation repeats upon the next the sins and errors which it suffered from its elders in its youth.”

“a people that has lost the ability to govern itself, or to develop its natural resources, inevitably falls prey to nations suffering from strength and greed”

“sometimes, on a sofa in an inner room, he would spend half a day silent with his memories and his dreams”

“Shorn of religious belief, which is the very spirit of India, the Westernized Hindus returned to their country disillusioned and sad; a thousand gods had dropped dead from the skies”

“While still a boy Mohandas became an atheist, being displeased with the adulterous gallantries of certain Hindu gods; and to make clear his everlasting scorn for religion, he ate meat. The meat disagreed with him, and he returned to religion”

“The Judge expressed his profound regret that he had to send to jail one whom millions of his countrymen considered “a great patriot and a great leader”; he admitted that even those who differed from Gandhi looked upon him “as a man of high ideals and of noble and even saintly life.” He sentenced him to prison for six years.”

“It was Gandhi’s task to unify India; and he accomplished it. Other tasks await other men.”

China

“China has been called “the paradise of historians.” For centuries and millenniums it has had official historiographers who recorded everything that happened, and much besides”

“legend, which loves personalities more than ideas, attributes to a few individuals the laborious advances of many generations.”

“and endeared himself to scholastic posterity by reducing the size of the whip with which Chinese children were educated”

“In his reign, according to sacred legend, rice wine was discovered, and was presented to the Emperor; but Yü dashed it to the ground, predicting: “The day will come when this thing will cost some one a kingdom.” He banished the discoverer and prohibited the new beverage; whereupon the Chinese, for the instruction of posterity, made wine the national beverage”

“For the rest we must rely on stories whose truth may not be proportioned to their charm.”

“The historians, more virtuous than history, assure us that Wu Yi was struck dead with lightning”

“Annoyed by pamphlets in which Teng criticized his policies, the prime minister prohibited the posting of pamphlets inpublic places. Teng thereupon delivered his pamphlets in person. The minister forbade the delivery of pamphlets. Teng smuggled them to his readers by concealing them in other articles. The government ended the argument by cutting off his head”

“The intellectual man is a danger to the state because he thinks in terms of regulations and laws; he wishes to construct a society like geometry, and does not realize that such regulation destroys the living freedom and vigor of the parts. The simpler man, who knows from his own experience the pleasure and efficacy of work conceived and carried out in liberty is less of a peril when he is in power for he does not have to be told that a law is a dangerous thing, and may injure more than it may help. Such a ruler regulates men as little as possible; if he guides the nation it is away from all artifice and complexity towards a normal and artless simplicity, in which life would follow the wisely thoughtless routine of nature,”

“In the ancient days, says Lao, nature made men and life simple and peaceful, and all the world was happy. But then men attained knowledge, they complicated life with inventions, they lost all mental and moral innocence, they moved from the fields to the cities, and began to write books; hence all the misery of men, and the tears of the philosophers”

“Passivity has its victories more often than action”

“if the wise man knows more than other men he tries to conceal it; “he will temper his brightness, and bring himself into agreement with the obscurity (of others); he agrees with the simple rather than with the learned, and does not suffer from the novice’s instinct of contradiction”

“and indeed, from some points of view it is one of the most culpable oversights of nature that virtue and beauty so often come in separate packages”

“Confucius taught the art of reasoning not through rules or syllogisms, but by the perpetual play of his keen mind upon the opinions of his pupils; when they went out from his school they knew nothing about logic, but they could think clearly and to the point”

“Obscurity of thought and insincere inaccuracy of speech seemed to him national calamities.”

“men will not take seriously one who is not serious with them;”

“There was not much of the revolutionist in Confucius; perhaps he suspected that the inheritors of a revolution are made of the same flesh as the men whom it deposed”

“Therefore the prime instrument of government is good example: the ruler must be an eminence of model behavior, from which, by prestige imitation, right conduct will pour down upon his people”

“In the long run the philosophy of Confucius triumphed. We shall see later how the mighty Shih Huang-ti, with a Legalist for his prime minister, sought to end the influence of Confucius by ordering that all existing Confucian literature should be burned. But the power of the word proved stronger than that of the sword; the books which the “First Emperor” sought to destroy became holy and precious through his enmity, and men died as martyrs in the effort to preserve them.”

“He was among the earliest of Chinese logicians, and the worst of Chinese reasoners”

“It is true that the wicked sometimes leave a bad name behind them, but this is a matter that does not disturb their bones”

“Again we suspect that time, who is a reactionary, has preserved for us the most respectable of Chinese thinkers, and has swallowed nearly all the rest in the limbo of forgotten souls”

“We perceive how old are the political problems, attitudes and solutions of our enlightened age when we learn that Mencius was rejected by the princes for his radicalism, and was scorned for his conservatism by the socialists”

“The “return to Nature,” however, could not be so readily discouraged; it found voice in this age as in every other”

“If by some negligence on his part, a true philosopher should find himself in charge of a state, his proper course would be to do nothing, and allow men in freedom to build their own organs of self-government.”

“Yao and Shun, instead of being so honored by China and Confucius, should be charged with having destroyed the primitive happiness of mankind by introducing government”

“Huns, barred for a time from Chinese soil, moved west into Europe and down into Italy; Rome fell because China built a wall”

“He decreed that his dynastic successors should number themselves from him as “First Emperor,” down to the ten thousandth of their line; but the line ended with his son.”

“Disorder followed his death, as it has followed the passing of almost every dictator in history; only an immortal can wisely take all power into his hands”

“On his accession to power he surrounded himself not with the usual politicians, but with men trained in letters and philosophy; to these men his enemies attributed his failure, and his friends attributed his success.”

“Wang had conceived his policies in economic terms, and had forgotten the nature of man. He worked long hours, day and night, to devise schemes that would make the nation rich and happy; and he was heart broken to find that social disorder mounted during his reign.”

“What we do see is, above all, brevity. We are apt to think these poems too slight, and feel an unreal disappointment at missing the majesty and boredom of Milton and Homer. But the Chinese believe that all poetry must be brief; that a long poem is a contradiction in terms—since poetry, to them, is a moment’s ecstasy, and dies when dragged out in epic reams”

““The men of old,” say the Chinese, “reckoned it the highest excellence in poetry that the meaning should be beyond the words, and that the reader should have to think it out for himself.””

“Like Chinese manners and art, Chinese poetry is a matter of infinite grace concealed in a placid simplicity.”

“The Chinese pardonably consider their literature superior to any other than that of Greece; and perhaps the exception is due to Oriental courtesy”

“and this office of court historian, carried down to our own generation, has raised up in China a mass of historical literature unequaled in length or dullness anywhere else on the earth”

“In a conflict between superstition and philosophy one may safely wager on the victory of superstition, for the world wisely prefers happiness to wisdom”

“As the result of Ch’ien Lung’s prohibition of woman players, female parts were acted by men, and so well that when women were in our time again admitted to the stage, they had to imitate their imitators in order to succeed.”

“and brought to many hearers that escape from the strife of wills and ideas which comes with the surrender to music well composed”

“The pursuit of wisdom and the passion for beauty are the two poles of the Chinese mind, and China might loosely be defined as philosophy and porcelain”

“Finally their greatest achievements were unconsciously hidden from Western travelers. For the Chinese do not flaunt their pictures on public or private walls; they roll them up and store them carefully away and unfold them for occasional enjoyment as we take down and read a book”

“Hui Tsung, with an empire at his command, gave half his life to painting birds and flowers.”

“the world of beauty and the world of money never touch, even when beautiful things are sold.”

“The educated classes displayed long nails as Western women wore French heels—to indicate their exemption from physical toil”

“In the third century Hua To wrote a volume on surgery and made operations popular by inventing a wine which produced a general anesthesia; it is one of the stupidities of history that the formula for mixing this drink has been lost”

“Soap was a rare luxury, but lice and vermin were easily secured”

“For through this profound institution the nation, which was shut out from physical and spatial unity by great distances and the poverty of transport, achieved a powerful spiritual unity in time; the generations were bound together with the tough web of tradition, and the individual life received an ennobling share and significance in a drama of timeless majesty and scope.”

“But in the second century of our era these doctrines were improved upon by men who claimed to have received, in direct line from Lao-tze, an elixir that would confer immortality. This drink became so popular that several emperors are said to have died from pious indulgence in it”

“What the Chinese,” said Voltaire, “best know, cultivate the most, and have brought to the greatest perfection, is morality.”

“By building the house on a sound foundation,” Confucius had said, “the world is made secure”

“Though economically subordinate the woman enjoyed the franchise of the tongue, and might scold her man into fright or flight in the best Occidental style”

“The most impressive aspect of this civilization was its system of government. If the ideal state is a combination of democracy and aristocracy, the Chinese have had it for more than a thousand years; if the best government is that which governs least, then the Chinese have had the best. Never has a government governed so many people, or governed them so little, or so long.”

“There was not a word in any of the tests about science, business or industry; the object was to reveal not knowledge but judgment and character. Those who survived the tests were at last eligible for the higher offices in the state”

“Considerable portions of this indemnity were later remitted to China by the United States, Great Britain, Russia and Japan, usually with the stipulation that the remitted sums bespent in educating students from China in the universities of the remitting nation. It was a gesture of generosity, which proved more effective in the undoing of old China than almost any other single factor in this historic and tragic conflict of East and West.”

“The European world, which had proposed a moratorium on robbery after it had gathered in all available spoils joined America feebly in protests against this candid plunder, but prepared, as always, to accept victory as justification in the end.”

“Once everything changed except the East; now there is nothing in the East that does not change. The most conservative nation in history has suddenly become, after Russia, the most radical, and is destroying with a will customs and institutions once held inviolate. It is not merely the end of a dynasty, as in 1644; it is the moulting of a civilization”

“The queue is gone, and so are the graceful manners of the older time; the hatreds of revolution have coarsened the spirit, and radicals find it hard to be courteous to conservatives”

“Revolution, like death and style, is the removal of rubbish, the surgery of the superfluous; it comes only when there are many things ready to die. China has died many times before; and many times she has been reborn.”

Japan

“The feudal system triumphed in Japan for the same reason that it had triumphed in Europe: local sources of authority grew in power as a central and distant government failed to maintain security and order”

“Kublai ordered from his ship-builders so vast a fleet that Chinese poets represented the hills as mourning for their denuded forests”

“The Japanese, in heroic retrospect, reckoned the vessels at 70,000, but less patriotic historians are content with 3,500 ships and 100,000 men”

“The legal system of Japan was a vigorous supplement to private assassination and revenge”

“In the Nara period the state nationalized the land, and rented it to the peasant for six years or at most, till death; but the government discovered that men did not care to improve or properly care for land that might in a short time be assigned to others; and the experiment ended in a restoration of private ownership, with state provision of funds in the spring to finance the planting and reaping of the crops”

“When he heard that Hideyoshi was coming to see his famous collection of chrysanthemums, Rikyu destroyed all the blossoms in his garden but one, so that this might shine unrivaled before the terrible shogun”

“Flowers are the religion of the Japanese; they worship them with sacrificial fervor and national accord. They watch for the blossoms appropriate to each season; and when, for a week or two in early April, the cherry-tree blooms, all Japan seems to leave its work to gaze at it, or even to make pilgrimages to places where the miracle is most abundant and complete”

“Criminals en route to execution will sometimes ask for a flower”

“The more serious devotee might cleanse his spirit by praying for a quarter of an hour under a waterfall in the depth of winter; or he might go on pilgrimages from shrine to shrine of his sect, meanwhile feasting his soul on the beauty of his native land”

“The aim of learning is not merely to widen knowledge but to form character. Its object is to make us true men, rather than learned men.”

“The chief reason why the teachings of the sages are not more appreciated by the people is because scholars endeavor to show off their learning, rather than to make it their endeavor to live up to the teachings of the sages.”

“The Shogunate trembled at the notion that every man might judge for himself what was right and what was wrong.”

“When he died they placed this enviable epitaph upon his tomb: He did not talk about the faults of others…He cared for nothing but books. His life was uneventful”

“Unlike Togai he enjoyed controversy and spoke his mind violently about philosophers living or dead. When an inquiring young man asked him, “What do you like besides reading?” he answered, “There is nothing better than eating burnt beans and criticizing the great men of Japan.””

“The Jesuits, harassed with these linguistic barriers, reported that the language of the islands had been invented by the Devil to prevent the preaching of the Gospel to the Japanese.”

“The Japanese, with whom gambling is a favorite passion, wagered so much money in haiku-composing contests that some enterprising souls made a business of conducting them, fleecing thousands of devotees daily, until at last the government was forced to raid these poetical resorts and prohibit this new mercenary art”

“Murasaki writes with a naturalness and ease that soon turn her pages into the charming gossip of a cultured friend.”

“He accepted poverty with good humor and, having no furniture, hung his bare walls with paintings of the furniture he might have had.”

“On his deathbed Ikku enjoined his pupils to place upon his corpse, before the cremation then usual in Japan, certain packets which he solemnly entrusted to them. At his funeral, prayers having been said, the pyre was lighted, whereupon it turned out that the packets were full of firecrackers, which exploded merrily. Ikku had kept his youthful promise that his life would be full of surprises, even after his death”

“The death of a civilization seldom comes from without; internal decay must weaken the fibre of a society before external influences or attacks can change its essential structure, or bring it to an end.”

“The eternal war of the generations—the revolt of over-eager youth against overcautious age—has been intensified by the growth of individualist industry, and the weakening of religious faith”

“A Chinese boycott of Japanese goods ensued; but Japan proceeded on the historically correct assumption that boycotts are sooner or later frustrated by the tendency of trade to follow the line of lowest costs”

Conclusion

“The institutions, customs, arts and morals of a people represent the natural selection of its countless trial-and-error experiments,the accumulated and unformulable wisdom of all its generations;”

“Europe and America are the spoiled child and grandchild of Asia”

“It would conceive man as a citizen rather than as a subject; it would give him political liberty civil rights, and an unparalleled measure of mental and moral freedom; it would create democracy and invent the individual.”

The Penultimate Wave of AI

jarbus@tutanota.com (jarbus) — Tue, 28 Jan 2025 19:46:23 -0500

I don’t think r1 will get us to artificial super intelligence, but whatever comes next probably will.

We are reaching a familiar bottleneck in AI. Previously, humans had to manually hardcode the patterns that AI could recognize. With deep learning, machines began to learn patterns on their own, without human assistance. With (relatively) expensive humans out of the loop, we threw machines at the world’s data until they began to talk, code, and paint. Many people believed this would be sufficient to reach artificial super intelligence–but it wasn’t.

We ran out of data. Luckily, our newborn bots could talk 24/7–not just to hundreds of millions of people, but also to other programs. These programs would ask the bots questions, verify their answer, and then use the correct answers to further improve the models. Free, infinite data. If the world’s data wasn’t enough, then infinite data must be–but it wasn’t.

We are now watching the rise of reasoners–models that “think” before giving an answer. Models which generate a series of words to raise the probability of eventually producing something we want. Surely, surely, this is it. Once we train a model to reason, it will be able to reason about its own answers, and somehow, magically, self-improve.

This infinite self-improvement probably won’t happen. In the same way that a fixed amount of mass can’t produce infinite energy, I suspect a fixed amount of information can’t produce infinite intelligence, no matter how much we feed it back into itself. Fundamentally, a model needs information from the outside, whether that information is a response from an external system or a human filtering its prior output for quality.

We provide some external information using formal verification systems, which is why math and programming performace is the prominent flex from the latest models. But most domains, like biology, business, and rocket science, are not so lucky; the correct answer is rarely obvious, even for programming. How does one automatically verify an interface is easy to use?

The answer still comes down to humans. Much to the dismay of a few arrogant men in San Francisco, people are still a key ingredient. This time, we don’t merely tell machines what patterns to look for; rather, we filter synthetic data to “steer” the model towards answers we prefer, i.e, we tell them what patterns to look for, but cheaper.

At this point, if the right mixture of curated data could yield superintelligence, it would have–models are now crushing benchmarks which PhDs struggle on. This is ridiculous. We’re going to wring these models of performance until they’re dry, and even then they won’t be superintelligent. Aside from extending formal verification, the way forward still seems to be curating the reasoning data these models generate and feeding it back in. As chains of reasoning become longer and more complex, human-curated data will likely yield increasingly diminishing returns.

We’ll be approaching the limit of what these systems can do. Don’t get me wrong, they can do a lot–they’re basically magic–but they can’t generally self-improve without humans or verifiers. And honestly, I can’t imagine what else comes next besides self-improving AI. We don’t know what that looks like–maybe some multiagent game-playing system–but whatever it is, it will be the last. For real, this time. Probably.

Reducing Code Complexity using UI

jarbus@tutanota.com (jarbus) — Fri, 15 Nov 2024 07:01:30 -0800

I’m doing an internship right now, and thankfully, I read a few books on software design before starting.

I had to design a database schema, data submission page, submission approval page, and dynamic dashboard for the project I was assigned to. This is one of those projects that AI can obviously do 90% of to work for if designed appropriately–if the right abstractions are used, performance trade-offs are made, and the right tools are chosen. I wanted to make sure that, throughout the project, AI could always easily with any part of the codebase.

For AI, code simplicity is a non-negotiable. Your code needs to be small, modular and clear, full stop. AI doesn’t work well with complex code bases with many unfamiliar datastructures across multiple files yet, so one goal was limit the codebase for each webpage to a single file, and for that file to be as small as possible, which worked really well. Being able to copy+paste your entire codebase into ChatGPT is very powerful.

We wanted to implement features such that the user would never end up in a slightly unintuitive state, specifically, we didn’t want users to normalize data by a series and view the data in log scale simultaneously. The issue is, given our tooling, it would take a decent chunk of code (relative to our codebase) to make it impossible to end up in such a state, and would require some logic to manage which action they took first. Did they normalize first, then try to view in log scale? Or did they try to view in log scale, and then normalize? What should take priority? What does the user expect?

Instead of increasing the complexity and control flow to handle this, I instead tried to first make it clear to the user exactly what state they were in. We added a large, visible red X to the normalization dropdown to draw the user’s attention to the “Clear Normalization Series” option, which resets the chart to the absolute values, instead of relative. From the user perspective, this makes it clear when you are working with normalized data, so when you view in log scale, you know what you’re looking at, and can decide for yourself what view you want to see. If you want to see something unintuitive, no biggie, it’s obvious what you are looking at. The “Make the X big and Red” commit was a couple lines of CSS.

For a “Graph Builder” feature, we initially planned to add a large number of buttons, dropdowns, and toggles to the UI elements so users can specify which types of data they want to view, and how they want to filter it. Adding the UI elements for each field would be a complexity nightmare, since we have a growing number of fields, many of which have missing data. However, we know our users are not our customers but rather internal employees, who have no other choice but to use our tool, and can afford a bit of a learning curve, since they will likely be using it for a while.

Instead of adding potentially hundreds of lines of UI code/ callbacks, default options, and graceful handling of incompatible conditions, our solution is simply to allow users to specify a custom MongoDB query to fetch the data they want to visualize. This is the most powerful option, as it easily gives our users any view they desire, but it also requires that they know the database schema, which is annoying, even for me, the guy who made designed it.

To make this feature user-friendly, we added a small “Show Query” button to all existing charts, so users can easily copy and maste queries used for existing graphs, gaining exposure to some of the fields and valies in the schema. This is a good start, but it’s still annoying to remember what additional fields are available, and what the current options are. This dashboard will have a tiny number of users and a modest amount of data, so what we can actually do, is run the working textarea query on each keystroke (if it’s valid) and, if no data is found, find out which values/fields don’t exist in the schema, and dynamically update a message below the text area with a list of existing entries which the user can type, all of which are easy to read. This too is a few extra lines of code, but allows the user to easily discover any field or value they want to use as they are typing the query. As a bonus, it allows users to save custom views in plain text as a MongoDB query, if they want to save or share them.

By rethinking how we conveyed information to the user, we avoided increasing the complexity of our codebase in the first place. This approach doesn’t apply to all scenarios, especially if your users are paying customers, but it’s great when it does.

Eudaimonia in the Information Age

jarbus@tutanota.com (jarbus) — Thu, 14 Nov 2024 16:48:51 -0800

Wikipedia translates the Greek philosophical concept of Eudaimonia as “Human Flourishing”, particularly on the spiritual level. It refers to genuine happiness and fulfillment over hedonism or GDP.

The modern world feels meaningless at times. Instead of hunting or farming to put food on the table, many of us sit at a computer or stand behind a counter, far removed from the fruits of our labor—if our labor even bears fruit. Our screens bombard us with videos of people more beautiful and successful than we could ever hope to be while we stay inside, alone.

Life is no longer a struggle for survival for many of us; it’s a struggle for meaning. We evolved to invoke happiness and fulfillment in completely different conditions than those in which we now reside. Like a camera trying to take pictures in the dark.

I’ve struggled immensely to find meaning and pride in the past, but I’ve found my light recently, and I wanted to write it down.

I’ve always held anger towards myself, to the point where I eventually collapsed. When I put myself back together over the following ten months, the anger didn’t disappear, but scattered, into hundreds of tiny thoughts, impulsive comments, and subconscious beliefs. To stop these ideas from coalescing back into a goliath of gloom, I began to read books on zen, to some success.

I ended up taking a four-month internship across the country. I was isolated, away from friends and family, and probably talking to ChatGPT more than real people. Strangely, despite this, I thrived. I started focusing more on myself: I began fasting, took better care of my appearance, found my sense of style, and watched videos on improving my communication skills. I started to fill in the gaps in my knowledge of history and world politics. I becme the person who I wanted to be. Both my confidence and competence skyrocketed.

Life is full of many cognitive plagues, things that will always bother you somewhere deep down. Whether it’s your appearance, your intelligence, your family, your job, it just never sits quite right. I wasn’t living a bad life before, but I still had these persistent, negative feelings. The difference in how I feel now is night and day. I feel like I’m nearing my true potential.

All I had to do is, well, everything. For the many sources of negative thoughts, I had to actually try to deal with it to some extent, and make progress. I’m not the most attractive guy in the world, but at least I feel comfortable with my hair quality now. I’m not the best speaker, but I’ve learned to slow down and speak with clarity and confidence.

Nietzsche describes the concept of the Übermensch, which I interpret as “the ideal version of ourselves”. We can’t permanently reach this ideal, but we can strive for it, and, for brief periods, embody it. The Übermensch serves as a guide; I’d literally ask myself, ‘What would my Übermensch do?’ and then do that, because the answer was always obvious.

We often focus our efforts on improving one aspect of our lives, like our career or health, which is important, but always leaves another aspect with which we are dissatisfied. I now feel like I’m satisfied with every aspect of myself. I’m not perfect, not by a long shot, but there’s no longer parts of myself that I truly despise.

I won’t feel this way forever. I’ll break down again, and hopefully, get back up once more. But I wanted to log these thoughts while they are still fresh, so I (or someone else) can find strength.

Originality in the Age of AI

jarbus@tutanota.com (jarbus) — Sat, 05 Oct 2024 09:23:29 -0700

It used to be good enough just to copy others. Now, with AI in the hands of billions, there’s little value in copying.

For instance, take programming. Five years ago, building apps, websites, or games required a non-trivial amount of skill, and getting your first project off the ground was an accomplishment. Now, AI can generate most starter projects in hours, if not minutes. I think this decimates the reward, both internal and external, of actually completing the first few projects.

First, it’s less impressive, which, in my opinion, matters. We like the feeling of pride we get when we show our work to others, especially when just starting out. It’s sad to diminish the relative value of someone’s work when they are just starting.

Second, it’s less satisfying for us. For those who just want to build useful tools, modern technology is great. But, for those who want to feel a sense of reward for creating something, it’s sad, I think. There’s still lots to build, but for most projects, especially starter projects, the internal payoff’s been greatly diminished.

This might have long-term consequences, because to get to the point where you can build something truly original, you first have to build many things which are unoriginal. The end goal of making something new is still just as valuable, if not more so, but the journey to develop the skills and original voice seems more mundane than ever. I also suspect that the next generation, who learns from AI will be significantly less technically competent, as they may never need to understand the technology they are working with.

I think this will place an upwards pressure on originality and novelty. Now that both the technical barrier to entry and the cost of producing unoriginal work is so low, society will start to value original ideas more than ever–doubly so if we reduce the rewards of the journey towards becoming original and skilled. If the world becomes flooded with less original, less technical users of AI, the value of technical competence and originality will skyrocket.

Apple Announces iPad Pro Max

jarbus@tutanota.com (jarbus) — Thu, 09 May 2024 11:22:24 -0400

CUPERTINO, CALIFORNIA - Apple today announced the release of the groundbreaking iPad Pro Max, the world’s first tablet powered by the all-new quantum Q1 processor. Capable of 30 octillion floating point operations per second, the Q1 makes the iPad Pro Max the most powerful streaming device ever made, over 2x faster than the M4 iPad Pro. Available in both 11-inch and 13-inch sizes, Apple says that the iPad Pro Max is the perfect device for generating TV shows, movies, and games in real-time on the go using the power of AI, no network connection needed.

The new iPad Pro Max–the thinnest device Apple has ever created–features a stunningly thin and light design, taking portability to a whole new level. The 11-inch model is just 3 nanometers thin–thinner than a piece of paper–and the 13-inch model is just 2 nanometers, thanks to TSMC’s cutting-edge 2-nanometer process, ensuring that you always have enough space in your bag for your Macbook to do real work.

With the iPad Pro Max’s stunning 16k display, today’s children can play Roblox in stunning detail, giving them the edge they need to dominate their less-affluent friends. With 100,000 nits of peak brightness, the iPad Pro Max can even be seen on the surface of the sun, enabling our astronauts to make the music they dreamed of in Garageband while still maintaining 10 minutes of battery life.

The beautiful Magic Keyboard Pro Max is the perfect companion for the iPad Pro Max. Featuring a new function row, larger trackpad, and hydraulic press, we can crush the dreams of our users and regulators like never before. The gorgeous aluminum palm rest and trackpad with haptic feedback makes the entire experience feel just like using a macbook, without the software.

The vesatility of the iPad Pro Max is enhanced by the new Apple Pencil Pro Max, which can now be used to write on any surface, including the air, thanks to Apple’s new AirInk technology, enabling anyone to change the world with a stroke. The Apple Pencil Pro Max starts at just $199,999.99, the Magic Keyboard Pro Max at $399,999.99, and the iPad Pro Max at just $999,999.99.

The iPad Pro Max is designed with the environment in mind. Featuring the thinnest, most irrepairable design in any product to date, the iPad Pro Max is guaranteed to take up less space in landfills than any other tablet on the market. Apple has also announced a new Core Technology Recycling Program, customers can return unsupported devices to Apple, in exchange for Apple charging a $0.50 Core Technology Recycling fee to the developer of each non-Apple app remaining on the device.

The Reasonable Effectiveness of Using Old Phones as Servers

jarbus@tutanota.com (jarbus) — Sat, 16 Mar 2024 14:42:47 -0400

I installed linux on a OnePlus6T. Setup took less than an hour, technical issues included.

Why

$90 for 8GB RAM, 128GB storage, 2.8GHz 8-core CPU
Linux on the OnePlus 6T is well documented
Low power consumption
Tiny footprint
No additional cables
Built-in battery backup
WiFi, bluetooth, speaker, screen, etc
Negligible environmental impact sans shipping
I wanted to play with Linux on ARM

I don’t need a dedicated cable to power or connect it. Running on wifi, it only needs to be plugged into my phone’s charger. Whenever I need to charge my phone, I unplug the server and plug in my phone. It takes up so little space that I literally leave it on my night stand.

Unexpected Issues

I couldn’t get the Phosh image to boot, so I used the Plasma Mobile image instead.
The phone would suspend after a few minutes of inactivity. This caused a lot of issues over SSH and took a while to figure out, since servers usually don’t suspend.

Docker

After resolving the suspend issues, docker appears to work fine. Surprisingly, many images support ARM!

Tailscale

I installed tailscale to access the server remotely without exposing it to the internet. Tailscale is phenomenal, my grandma could set it up in less than five minutes if she needed to.

Viability as a regular phone

The OnePlus 6T is so much faster than a PinePhone, the Linux phone I used previously. But Linux still isn’t ready for daily-driving:

I need to manually connect to wifi for each reboot
The app ecosystem is still a work in progress, missing polished apps for navigation, cameras, streaming, etc.
Traditional Linux apps like Firefox don’t scale properly

Hopefully these issues are resolved in the future, but for now, the 6T is an excellent server.

Other Remarks

I’ve been toying with a few containers, but still haven’t decided on specific services to run yet. I’m considering:

Keila for adding a newsletter to my blog
Mastodon to test running an activitypub server
Matrix Synapse/Dendrite and assorted bridges for chat

I’ll post updates as things progress.

Zen in the Art of Beat Saber

jarbus@tutanota.com (jarbus) — Thu, 22 Feb 2024 14:21:27 -0500

Contents: Introduction, Step away, come back later, Focus, but not too much, Generalization comes after specialization, Don’t bother if you don’t need to, Let Go

Introduction

I recently read Zen in the Art of Archery, which isn’t a book about archery as much as it is a book about zen. The book describes author’s journey learning archery from a zen master. I highly recommend it, the book is a short read–I finished it in a day.

My core takeaway, which fails to properly describe book, is the following:

To master archery, the archer must first let go of their body and mind so they can be one with the art. This can only be achieved with an open mind through extensive practice.

I was mesmerized by the dedication, personal growth and the zen-like state the author achieved throughout his journey. I wanted to try something similar but have neither materials for archery or a zen master. I do, however, have a VR headset and a copy of Beat Saber, which can apparently induce a feeling of zen. I gave it a shot, silly as it seems.

Beat Saber a game where you need to slice blocks in a specific direction to the beat of a song. The harder the level, the more notes you have to hit per second. If you miss too many notes within a short time span, you fail the song. For reference, here is a video of gameplay on an easier difficulty, a video of gameplay on harder difficulty, and a list of the songs.

My initial goal was to beat all ten songs in the first volume of the free soundtrack on the hardest difficulty, Expert+ (I also started this to get myself to exercise, but that’s less exciting). As someone who couldn’t beat a single medium difficulty song on Guitar Hero 3 as a child, I didn’t expect to beat anything on the hardest setting, let alone one that required physical activity.

I’ll spare you the hero’s journey: I beat all ten songs on Expert+ in a few weeks. I couldn’t believe my progress; I figured beating a single level would take me at least a month or two.

I decided to push myself and beat all the songs in all six original soundtracks on Expert+. Again, I figured the levels would get harder, and I’d spend weeks mastering some of the harder songs. They did get significantly harder, but I chiped away at all but the six fastest songs over the course of another few weeks, one song at a time.

I’m still no zen-master, but I wanted to share some lessons I’ve learned, which I think can apply to lots of aspects of life. Take these with a grain of salt, of course.

Step away, come back later

The body naturally improves once you let it rest and process. I’d struggle with a song one night only to beat it on the first try the following morning.

Focus, but not too much

Instead of restarting a failed song from the beginning, I relied on practice mode, which lets you start the song from any time. I drilled the sections I struggled with at lower speeds, ramping up the speed as I improved, eventually going back and beating the full song. Said in reverse, I wasted little time on the easy parts and made good use of my time.

When I got tired of a song, I switched to one that I previously beat, but found that I rusted. Switching it up and reviewing things I previously mastered improved my general ability and allowed me to perform better when encountering a new song.

Generalization comes after specialization

I spent most of my time drilling specific sequences of notes in certain songs, which I didn’t think would generalize well to other songs. While true to some extent, this practice helped me master patterns which would took similar forms elsewhere.

I think the same principle applies to learning in general, at least for me. I learn better when I start from examples and experience. Only after a while do I obtain the subconscious instincts required to shift my attention from the specifics to the higher level abstractions. My growth as a programmer followed a similar path, and the same is true for chess grandmasters from what I hear.

Don’t bother if you don’t need to

During faster, less-intuitive sequences, I’d often fail because I attempted to hit every single note. I noticed that I’d usually go down-hill after making a few consecutive mistakes which threw me off. Instead of drilling these sequences to mastery, I realized I could flat-out ignore the few pain points that disrupted my flow without failing the track.

Let Go

Eventually, the songs got so fast that my conscious mind couldn’t process rapid sequences in time. These parts weren’t counter-intuitive or confusing like prior sequences I drilled, they were simply too fast. I felt drilling these sequences wouldn’t help, no matter the speed.

The strategy that eventually worked was “not thinking”, true to zen philosophy. For most of the song, I’d consciously process the notes and patterns as they came, but for the fastest sequences, I would clear my mind, and operate purely on instinct for a few seconds. This was perhaps the most surpising part of this journey–I’d often be able to land a perfect or near-perfect combo once my mind no longer got in the way.

Final Thoughts

I plan on beating the remaining songs, and will update this post if/when I do. I did not expect this experiment to go anywhere, but I learned a surprising amount about myself, my learning process, and breaking my limits. With the exception of brief moments, I don’t experience a zen-like state when playing, but I do notice that I need to “think” about what I’m doing significantly less.

Emergent Trade and Tolerated Theft Using Multi-Agent Reinforcement Learning

jarbus@tutanota.com (jarbus) — Sun, 04 Feb 2024 12:14:25 -0500

I’ve been an author on a few papers before, but I recently published the first research project where I was responsible for most of the work and direction. It’s in the first 2024 issue of the journal Artificial Life, which you can find here. You can find a non-paywalled version here Below, I tell the chronology of the project and summarize our findings.

We explore the conditions under which trade can emerge between four deep reinforcement learning agents that pick up and put down resources in a 2D foraging environment. Agents are rewarded for having both resources once, but the resources are distributed far apart from each other. To maximize reward, agents need to split up the work - agent 1 goes to resource A, agent 2 goes to resource B, etc, and then they meet to exchange resources, since meeting halfway can get them the most of each resource in the shortest amount of time.

I was working on this for a while in isolation, getting nowhere, and slowly going crazy. A few months later, Deepmind published a paper that explored this exact behavior—and found that agents could not discover how to exchange goods without programming a trading system into the game mechanics directly. I was getting similar results, but found that this trading behavior could emerge if I rewarded agents for being physically close to one another, which we’ll call a “community bonus”.

I reached out to the authors (great people btw), and they hypothesized that the trading behavior was a result of agents trying to keep each other alive to get the community bonus. In my setup, agents could die if they ran out of resources, so they were also keeping each other alive for company. The authors suggested I add some sort of “market”, a place for agents to meet up and trade without getting directly rewarded for giving away resources for free.

They explored similar ideas in their work, and if they couldn’t get this behavior to emerge, what chance did I have? I needed to try something similar but sufficiently different.

Instead of adding a market where agents could come and go as they please, I added a campfire and a day-night cycle. I gave negative reward to agents for being out in the dark, which incentivizes them to gather around the campfire near others. Agents would only have enough time during the day to forage a single resource.

As it turns out, agents really, really don’t want to learn this behavior. It’s easy for agents to get cheated out of the resources they offer by others who offer nothing in return, so they gather as much as possible on their own. They even forage at night if the darkness penalty isn’t severe enough.

Eventually, after 4+ days of training time, agents start to realize that working together isn’t so bad and land on a trading protocol where agents stand back, drop an offer, then walk over to collect the resources dropped by others. Experiments show that agents keep their distance initially so they can reclaim their offer if their trading partner attempts to cheat them.

In one trial, agents converge on a local minimum, where agents 1&2 go to the same area to collect resource A, agent 3 goes to collect resource B, and agent 4 collects a bit of the other remaining resources. In this setting, we notice a behavior which we dub “Tolerated Theft”. When agent 1 collects significantly more of resource A than agent 2, agent 2 tries to steal the resources exchanged between agents 1 and 3. To defend against this, agent 1 will drop some resources away from agent 3, to get agent 2 off their backs while they trade. This is particularly interesting, because an agent manages to “steal” resources without us adding a combat or larceny system. Nature truly does find a way.

I recommend checking out the arxiv pre-print if you are interested in the details and figures.

List

jarbus@tutanota.com (jarbus) — Sun, 28 Jan 2024 13:17:32 -0500

punk.ist - A list of increasing importance

Unix - The philosophy of modular programs

vi key bindings - Comfortable computing

Artificial Life - Life in silico

100 Rabbits - Low-tech maritime artist collective

Antifragility - Strength from setbacks

Lauryn Hill’s MTV Unplugged - The unknown Queen of Hip-Hop

George Watsky’s Live! From the Metro - The concert which made me love live music

Justine Tunney - One hell of a programmer

suckless - Simple, frugal, clear software

LOW←TECH MAGAZINE - Forgotten tech creating a sustainable future

tilde.town - Community on a linux server

Arch Linux - which I use, btw

UNSONG - United Nations Subcommittee On Names Of God

The Way of Zen - Liberation on paper

Secrets of Our Success - The beauty of cultural evolution

Take the Road Most Documented

jarbus@tutanota.com (jarbus) — Sun, 28 Jan 2024 02:52:33 -0400

How great would it be if the solution to most errors you face were in the first place you looked? That’s what the Arch Wiki has been for me: a massive wealth of information and troubleshooting resources to help me navigate the various configuration and installation issues I’ve encountered. Some people claim Arch Linux is too difficult for new users, but for me it’s been the only distribution I’ve been able to get consistently working, and it’s all thanks to the detailed documentation and known workarounds.

I used to run a Dell XPS 9560, and have tried and failed to install Ubuntu and Manjaro on multiple occassions. Sometimes, the automatic installers would fail, other times they would succeed but install a broken configuration, leaving me with networking configurations that would crash as much as it would connect¹. However, when installing Arch, I resolved each installation error with the help of the wiki, so that by the time I was in a graphical desktop environment, everything was running smoothly.

When I got my next laptop, I looked on the arch wiki to make sure there would be as few issues as possible setting up Linux—and I faced none when installing.

I love exciting, and popular, and new software, but installing Arch showed me that popular isn’t as important as understandable. Software will always break; I don’t think there isn’t a single program I use daily whose configuration I haven’t broken at some point. But when I encounter an installation error, or a feature isn’t working, or I need to use a program for an unintended purpose, choosing programs with extensive documentation and active community resources has always saved me a headache.

I use Hugo to generate this website. It’s so well documented (and there are bountiful answers on StackOverflow) that not only can I usually find the answer to the thing I want to do in the first few search results, but GPT-4 can often make a specific change I want too.

With all else equal, take the road most documented.

This was before 2020, I’m sure these issues are largely fixed by now ↩︎

The Way of Zen

jarbus@tutanota.com (jarbus) — Sat, 27 Jan 2024 15:49:16 -0500

Excerpts I found interesting from Alan Watt’s The Way of Zen, grouped by vibes:

Definition Convention Feel Growing Spontaneity Pointing Ego Interference Present Negation Liberation Za-zen Widespread Art

Definition

Zen Buddhism is a way and a view of life which does not belong to any of the formal categories of modern Western thought.

As will soon be obvious, a way of liberation can have no positive definition. It has to be suggested by saying what it is not, somewhat as a sculptor reveals an image by the act of removing pieces of stone from a block.

An obvious reason for the lack of materials would be that a principle of this kind, so easily open to misinterpretation, might have been kept as a “secret doctrine,” discussed openly only in later times.

Zen points out that our precious “self” is just an idea, useful and legitimate enough if seen for what it is, but disastrous if identified with our real nature.

According to the famous saying of Ch’ing-yüan: Before I had studied Zen for thirty years, I saw mountains as mountains, and waters as waters. When I arrived at a more intimate knowledge, I came to the point where I saw that mountains are not mountains, and waters are not waters. But now that I have got its very substance I am at rest. For it’s just that I see mountains once again as mountains, and waters once again as waters.

To remain caught up in ideas and words about Zen is, as the old masters say, to “stink of Zen.” For this reason the masters talk about Zen as little as possible, and throw its concrete reality straight at us.

Another master was having tea with two of his students when he suddenly tossed his fan to one of them, saying, “What’s this?” The student opened it and fanned himself. “Not bad,” was his comment. “Now you,” he went on, passing it to the other student, who at once closed the fan and scratched his neck with it. This done, he opened it again, placed a piece of cake on it, and offered it to the master. This was considered even better, for when there are no names the world is no longer “classified in limits and bounds.”

Convention

We have difficulty in communicating with each other unless we can identify ourselves in terms of roles–father, teacher, worker, artist, “regular guy,” gentleman, sportsman, and so forth.

According to convention, I am not simply what I am doing now. I am also what I have done, and my conventionally edited version of my past is made to seem almost more the real “me” than what I am at this moment. For what I am seems so fleeting and intangible, but what I was is.

A philosophy restricted to the alternatives of conventional language has no way of conceiving an intelligence which does not work according to plan, according to a (one-at-atime) order of thought.

To serve their purpose, names and terms must of necessity be “fixed and definite like all other units of measurement. But their use is–up to a point–so satisfactory that man is always in danger of confusing his measures with the world so measured, of identifying money with wealth, fixed convention with fluid reality. But to the degree that he identifies himself and his life with these rigid and hollow frames of definition, he condemns himself to the perpetual frustration of one trying to catch water in a sieve.

Transitoriness is depressing only to the mind which insists upon trying to grasp.

We are so accustomed to this convention in speaking and thinking that we fail to recognize that it is simply a convention, and that it does not necessarily correspond to the actual experience of knowing. Thus when we say, “A light flashed,” it is somewhat easier to see through the grammatical convention and to realize that the flashing is the light.

The anxiety-laden problem of what will happen to me when I die is, after all, like asking what happens to my fist when I open my hand, or where my lap goes when I stand up.

The diffculty of making equations and comparisons between Eastern and Western ideas is that the two worlds do not start with the same assumptions and premises. They do not have the same basic categorizations of experience. When, therefore, the world has never been divided into mind and matter, but rather into mind and form, the word “mind” cannot mean quite the same thing in both instances. The word “man,” for example, does not have quite the same meaning when contrasted with “woman” as when contrasted with “animal.”

Feel

The general tendency of the Western mind is to feel that we do not really understand what we cannot represent, what we cannot communicate, by linear signs–by thinking. We are like the “wallflower” who cannot learn a dance unless someone draws him a diagram of the steps, who cannot “get it by the feel.”

We learn music, for example, by restricting the whole range of tone and rhythm to a notation of fixed tonal and rhythmic intervals-a notation which is incapable of representing Oriental music. But the Oriental musician has a rough notation which he uses only as a reminder of a melody. He learns music, not by reading notes, but by listening to the performance of a teacher, getting the “feel” of it, and copying him, and this enables him to acquire rhythmic and tonal sophistications matched only by those Western jazz artists who use the same approach.

By far the greater part of our important decisions depend upon “hunch”- in other words, upon the “peripheral vision” of the mind. Thus the reliability of our decisions rests ultimately upon our ability to “feel” the situation, upon the degree to which this “peripheral vision” has been developed. Every exponent of the I Ching knows this. He knows that the book itself does not contain an exact science, but rather a useful tool which will work for him if he has a good “intuition,” or if, as he would say, he is “in the Tao.” Thus one does not consult the

Growing

Whereas God produces the worldby maklng (wei), the Tao produces it by “not makmg” (wu wei) which is approximately what we mean by “growing.”

If the universe were made, there would of course be someone who knows how it is made-who could explain how it was put together bit by bit as a technician can explain in one-at-a-time words how to assemble a machine.

But the genuine “Zen flavor” is when a man is almost miraculously natural without intending to be so. His Zen life is not to make himself but to grow that way.

In the course of time it may grow, but one does not blame an egg for not being a chicken; still less does one criticize a pig for having a shorter neck than a giraffe.

But he watches as a gardener watches the growth of a tree, and wants his student to have the attitude of the tree–the attitude of purposeless growth in which there are no short cuts because every stage of the way is both beginning and end.

The Rinzai School has always forbidden the publication of formally acceptable answers to the various koan because the whole point of the discipline is to discover them for oneself, by intuition. To know the answers without having so discovered them would be like studying the map without taking the journey.

The Chinese proverb “What comes in through the gate is not family treasure” is understood in Zen to mean that what someone else tells you is not your own knowledge.

Spontaneity

We saw that the I Ching had given the Chinese mind some experience in arriving at decisions spontaneously, decisions which are effective to the degree that one knows how to let one’s mind alone, trusting it to work by itself. This is wu-wei, since wu means “not” or “non-” and wei means “action,” “making,” “doing,” “striving,” “straining,” or “busyness.”

“No amount of working with the muscles of the mouth and tongue will enable us to taste our food more acutely. The eyes and the tongue must be trusted to do the work by themselves.”

Become unaffected;

Cherish sincerity;

Belittle the personal;

Reduce desires.

~ Lao Tzu

The idea is not to reduce the human mind to a moronic vacuity, but to bring into play its innate and spontaneous intelligence by using it without forcing it. It is fundamental to both Taoist and Confucian thought that the natural man is to be trusted, and from their standpoint it appears that the Western mistrust of human nature–whether theological or technological–is a kind of schizophrenia.

When a man has learned to let his mind alone so that it functions in the integrated and spontaneous way that is natural to it, he begins to show the special kind of “virtue” or “power” called te. This is not virtue in the current sense of moral rectitude but in the older sense of effectiveness, as when one speaks of the healing virtues of a plant[…] Te is the unthinkable ingenuity and creative power of man’s spontaneous and natural functioning–a power which is blocked when one tries to master it in terms of formal methods and techniques.

Nirvana can only arise unintentionally, spontaneously, when the impossibility of selfgrasping has been thoroughly perceived.

The doctrine of the Dharmadhatu is, approximately, that the proper harmony of the universe is realized when each “thing-event” is allowed to be freely and spontaneously itself, without interference. Stated more subjectively, it is saying, “Let everything be free to be just as it is. Do not separate yourself from the world and try to order it around.”

“Six Precepts” of Tilopa: No thought, no reflection, no analysis, No cultivation, no intention; Let it settle itself.

–that truly to know is not to know, that the awakened mind responds immediately, without calculation, and that there is no incompatibility between Buddhahood and the everyday life of the world.

He was the first to answer questions about Buddhism by hitting the questioner.

We feel that our actions are voluntary when they follow a decision, and involuntary when they happen without decision. But if decision itself were voluntary, every decision would have to be preceded by a decision to decide–an infinite regression which fortunately does not occur. Oddly enough, if we had to decide to decide, we would not be free to decide. We are free to decide because decision “happens.” We just decide without having the faintest understanding of how we do it. In fact, it is neither voluntary nor involuntary.

The master may begin a conversation with the student by asking a series of very ordinary questions about trivial matters, to which the student responds with perfect spontaneity. But suddenly he will say, “When the bath-water flows down the drain, does it turn clockwise or counter-clockwise?” As the student stops at the unexpectedness of the question, and perhaps tries to remember which way it goes, the master shouts, “Don’t think! Act! This way–” and whirls his hand in the air. Or, perhaps less helpfully, he may say, “So far you’ve answered my questions quite naturally and easily, but where’s your difficulty now?”

The point of mo chih ch’u is not to eliminate reflective thought but to eliminate “blocking” in both action and thought, so that the response of the mind is always like a ball in a mountain stream-“one thought after another without hesitation.”

Yet this is why the Western mind is dismayed when ordered conceptions of the universe break down, and when the basic behavior of the physical world is found to be a “principle of uncertainty.”

The same is true in learning to use the brush for writing or painting. The brush must draw by itself. This cannot happen if one does not practice constantly. But neither can it happen if one makes an effort. Similarly, in swordsmanship one must not first decide upon a certain thrust and then attempt to make it, since by that time it will be too late. Decision and action must be simultaneous.

To use the imagery of a Tibetan poem, every action, every event comes of itself from the Void “as from the surface of a clear lake there leaps suddenly a fish.”

Pointing

Thus from the standpoint of Zen the Buddha “never said a word,” despite the volumes of scriptures attributed to him. For his real message remained always unspoken, and was such that, when words attempted to express it, they made it seem as if it were nothing at all.

Yet it is the essential tradition of Zen that what cannot be conveyed by speech can nevertheless be passed on by “direct pointing,” by some nonverbal means of communication without which the Buddhist experience could never have been handed down to future generations.

In its own (probably rather late) tradition, Zen maintains that the Buddha transmitted awakening to his chief disciple, Mahakasyapa, by holding up a flower and remaining silent.

The basic position of Zen is that it has nothing to say, nothing to teach. The truth of Buddhism is so self-evident, so obvious that it is, if anything, concealed by explaining it. Therefore the master does not “help” the student in any way, since helping would actually be hindering. On the contrary, he goes out of his way to put obstacles and barriers in the student’s path.

Ego

It is fundamental to every school of Buddhism that there is no ego, no enduring entity which is the constant subject of our changing experiences.

We can, for example, imagine the path of a bird through the sky as a distinct line which it has taken. But this line is as abstract as a line of latitude. In concrete reality, the bird left no line, and, similarly, the past from which our ego is abstracted has entirely disappeared. Thus any attempt to cling to the ego or to make it an effective source of action is doomed to frustration

Here, too, lies the basis of the Buddhism of faith, of the Sukhavati or Pure Land school, in which it is held that all efforts to become a Buddha are merely the false pride of the ego.

There is often a deceptive resemblance between opposite extremes. Lunatics frequently resemble saints, and the unaffected modesty of the sage often lets him seem to be a very ordinary person.

This is why the sage Fa-yung received no more offerings of flowers from the birds after he had had his interview with the Fourth Patriarch, for his holiness no longer “stood out like a sore thumb.”

No one notices him because he does not notice himself.

So often one thinks of the saint as a man whose sincerity provokes the enmity of the world, but Ryokan holds the distinction of being the saint whom everyone loved–perhaps because he was natural, again as a child, rather than good.

Interference

This, however, cannot quite be compressed into the sweeping assertion that “life is suffering.” The point is rather that life as we usually live it is suffering-or, more exactly, is bedeviled by the peculiar frustration which comes from attempting the impossible.

Man is involved in karma when he interferes with the world in such a way that he is compelled to go on interfering, when the solution of a problem creates still more problems to be solved, when the control of one thing creates the need to control several others. Karma is thus the fate of everyone who “tries to be God.” He lays a trap for the world in which he himself gets caught.

This nonduality of the mind, in which it is no longer divided against itself, is samadhi, and because of the disappearance of that fruitless threshing around of the mind to grasp itself, samadhi is a state of profound peace.

However, not exterminating the passions does not mean letting them flourish untamed. It means letting go of them rather than fighting them, neither repressing passion nor indulging it. For the Taoist is never violent, since he achieves his ends by noninterference (wu-wei), which is a kind of psychological judo.

To concentrate on the mind and to contemplate it until it is still is a disease and not dhyana.

On the principle that “the true mind is no-mind,” and that “our true nature is no (special) nature,” it is likewise stressed that the true practice of Zen is no practice, that is, the seeming paradox of being a Buddha without intending to be a Buddha.

There is no place in Buddhism for using effort. Just be ordinary and nothing special.

Whereas the koan advocates used this technique as a means for encouraging that overwhelming “feeling of doubt” which they felt to be essential as a prerequisite for satori, the Soto School argued that it lent itself too easily to that very seeking for satori which thrusts it away, or–what is worse–induces an artificial satori.

The temptation is all the stronger because it upsets the fondest illusion of the human mind, which is that in the course of time everything may be made better and better. For it is the general opinion that were this not possible the life of man would lack all meaning and incentive. The only alternative to a life of constant progress is felt to be a mere existence, static and dead, so joyless and inane that one might as well commit suicide. The very notion of this “only alternative” shows how firmly the mind is bound in a dualistic pattern, how hard it is to think in any other terms than good or bad, or a muddy mixture of the two.

To succeed is always to fail–in the sense that the more one succeeds in anything, the greater is the need to go on succeeding. To eat is to survive to be hungry.

Consequently the whole notion of getting something “out” of life, of seeking something “from” experience, becomes absurd.

In short, Zen begins at the point where there is nothing further to seek, nothing to be gained. Zen is most emphatically not to be regarded as a system of self-improvement, or a way of becoming a Buddha

Furthermore, the effort to remain always “good” or “happy” is like trying to hold the thermostat to a constant 70 degrees by making the lower limit the same as the upper.

To the Taoist mentality, the aimless, empty life does not suggest anything depressing. On the contrary, it suggests the freedom of clouds and mountain streams, wandering nowhere, of flowers in impenetrable canyons, beautiful for no one to see, and of the ocean surf forever washing the sand, to no end.

Yet the superficiality of this consciousness is seen in the fact that it cannot and does not regulate even the human organism. For if it had to control the heartbeat, the breath, the operation of the nerves, glands, muscles, and sense organs, it would be rushing wildly around the body taking care of one thing after another, with no time to do anything else. Happily, it is not in charge, and the organism is regulated by the timeless “original mind,” which deals with life in its totality and so can do ever so many “things” at once.

Clear sight has nothing to do with trying to see; it is just the realization that the eyes will take in every detail all by themselves, for so long as they are open one can hardly prevent the light from reaching them. In the same way, there is no di8culty in being fully aware of the eternal present as soon as it is seen that one cannot possibly be aware of anything else–that in concrete fact there is no past or future.

Present

On the one hand, it is one-pointed in the sense of being focused on the present, since to clear awareness there is neither past nor future, but just this one moment.

In the same way Dogen pointed out that firewood does not become ashes and life does not become death, just as the winter does not become the spring. Every moment of time is “self-contained and quiescent.

To Po-chang is attributed the famous definition of Zen, “When hungry, eat; when tired, sleep.” For when asked about seeking for the Buddha nature he answered, “It’s much like riding an ox in search of the ox.”

Dogen is here trying to express the strange sense of timeless moments which arises when one is no longer trying to resist the flow of events, the peculiar stillness and self-sufficiency of the succeeding instants when the mind is, as it were, going along with them and not trying to arrest them.

A Zen poem says: The morning glory which blooms for an hour Differs not at heart from the giant pine, Which lives for a thousand years.

On the contrary, the measuring of worth and success in terms of time, and the insistent demand for assurances of a promising future, make it impossible to live freely both in the present and in the “promising” future when it arrives. For there is never anything but the present, and if one cannot live there, one cannot live anywhere.

This is not a philosophy of not looking where one is going; it is a philosophy of not making where one is going so much more important than where one is that there will be no point in going.

The story is told of a Zen monk who wept upon hearing of the death of a close relative. When one of his fellow students objected that it was most unseemly for a monk to show such personal attachment he replied, “Don’t be stupid! I’m weeping because I want to weep.”

The silence which prevails is deepened rather than broken by occasional sounds that float up from a near-by village, by the intermittent ringing of soft-toned bells from other parts of the monastery, and by the chatter of birds in the trees.

Zen masters are quite human. They get sick and die; they know joy and sorrow; they have bad tempers or other little “weaknesses” of character just like anyone else, and they are not above falling in love and entering into a fully human relationship with the opposite sex. The perfection of Zen is to be perfectly and simply human. The difference of the adept in Zen from the ordinary run of men is that the latter are, in one way or another, at odds with their own humanity, and are attempting to be angels or demons.

It hurries on and on, and misses everything. Not hurrying, the purposeless life misses nothing, for it is only when there is no goal and no rush that the human senses are fully open to receive the world.

Aware is not quite grief, and not quite nostalgia in the usual sense of longing for the return of a beloved past. Aware is the echo of what has passed and of what was loved, giving them a resonance such as a great cathedral gives to a choir, so that they would be the poorer without it.

Zen has no goal; it is a traveling without point, with nowhere to go. To travel is to be alive, but to get somewhere is to be dead, for as our own proverb says, “To travel well is better than to arrive.

A world which increasingly consists of destinations without journeys between them, a world which values only “getting somewhere” as fast as possible, becomes a world without substance. One can get anywhere and everywhere, and yet the more this is possible, the less is anywhere and everywhere worth getting to.

For if we open our eyes and see clearly, it becomes obvious that there is no other time than this instant, and that the past and the future are abstractions without any concrete reality. Until this has become clear, it seems that our life is all past and future, and that the present is nothing more than the in-nitesimal hairline which divides them. From this comes the sensation of “having no time,” of a world which hurries by so rapidly that it is gone before we can enjoy it. But through “awakening to the instant” one sees that this is the reverse of the truth: it is rather the past and future which are the ,eeting illusions, and the present which is eternally real.

Both Takuan and Bankei stressed the fact that the “original” or “unborn” mind is constantly working miracles even in the most ordinary person. Even though a tree has innumerable leaves, the mind takes them in all at once without being “stopped” by any one of them.

He had surpassed all the other disciples and Bodhisattvas by answering a question as to the nature of the nondual reality with a “thunderous silence”–an example frequently followed by Zen masters.

Negation

Even at the conventional level it is surely easy to see that knowing what is not so is often quite as important as knowing what is […] Furthermore, the function of negative knowledge is not unlike the uses of space–the empty page upon which words can be written, the empty jar into which liquid can be poured, the empty window through which light can be admitted, and the empty pipe through which water can flow.

Indeed, when Hindu and Buddhist texts speak of the “empty” or “illusory” character of the visible world of nature–as distinct from the conventional world of things–they refer precisely to the impermanence of its forms.

Thus the greater part of Nagarjuna’s work was a carefully logical and systematic refutation of every philosophical position to be found in the India of his time.

It must therefore be repeated that the negations apply, not to reality itself, but to our ideas of reality.

Awakening is to know what reality is not. It is to cease identifying oneself with any object of knowledge whatsoever.

Liberation

But Taoism must on no account be understood as a revolution against convention, although it has sometimes been used as a pretext for revolution. Taoism is a way of liberation, which never comes by means of revolution, since it is notorious that most revolutions establish worse tyrannies than they destroy.

The Fourth Patriarch, following Seng-ts’an, is believed to have been Tao-hsin. When he came to Seng-ts’an he asked, “What is the method of liberation?” “Who binds you?” replied Seng-ts’an. “No one binds me.” “Why then,” asked Seng-ts’an, “should you seek liberation?”

True dhyana is to realize that one’s own nature is like space, and that thoughts and sensations come and go in this “original mind” like birds through the sky, leaving no trace. Awakening, in his school, is “sudden” because it is for quickwitted rather than slow-witted people. The latter must of necessity understand gradually, or more exactly, after a long time, since the Sixth Patriarch’s doctrine does not admit of stages or growth. To be awakened at all is to be awakened completely, for, having no parts or divisions, the Buddha nature is not realized bit by bit.

Za-zen

The following story is told of Huai-jang, initiating into Zen his great successor Ma-tsu, who was at the time practicing sitting meditation at the monastery of Ch’uan-fa. “Your reverence,” asked Huai-jang, “what is the objective of sitting in meditation?” “The objective,” answered Ma-tsu, “is to become a Buddha.” Thereupon Huai-jang picked up a floor-tile and began to polish it on a rock. “What are you doing, master?” asked Ma-tsu. “I am polishing it for a mirror,” said Huai-jang. “How could polishing a tile make a mirror?” “How could sitting in meditation make a Buddha?”

it could be assumed that the type of za-zen under criticism is za-zen practiced for a purpose, to “get” Buddhahood, instead of “sitting just to sit.” There is, of course, a proper place for sitting–along with standing, walking, and lying–but to imagine that sitting contains some special virtue is “attachment to form.”

Yet however much za-zen may have been exaggerated in the Far East, a certain amount of “sitting just to sit” might well be the best thing in the world for the jittery minds and agitated bodies of Europeans and Americans–provided they do not use it as a method for turning themselves into Buddhas

Although the West has its own contemplative tradition in the Catholic Church, the life of “sitting and looking” has lost its appeal, for no religion is valued which does not “improve the world,” and it is hard to see how the world can be improved by keeping still. Yet it should be obvious that action without wisdom, without clear awareness of the world as it really is, can never improve anything. Furthermore, as muddy water is best cleared by leaving it alone, it could be argued that those who sit quietly and do nothing are making one of the best possible contributions to a world in turmoil. There is, indeed, nothing unnatural in long periods of quiet sitting. Cats do it; even dogs and other more nervous animals do it. So-called primitive peoples do it–American Indians, and peasants of almost all nations. The art is most diffcult for those who have developed the sensitive intellect to such a point that they cannot help making predictions about the future, and so must be kept in a constant whirl of activity to forestall them. But it would seem that to be incapable of sitting and watching with the mind completely at rest is to be incapable of experiencing the world in which we live to the full

But just as there is no need to try to be in accord with the Tao, to try to see, or to try to hear, so it must be remembered that the breath will always take care of itself. This is not a breathing “exercise” so much as a “watching and letting” of the breath, and it is always a serious mistake to undertake it in the spirit of a compulsive discipline to be “practiced” with a goal in mind.

every human activity can become a form of za-zen

The late Dr. Kunihiko Hashida, a lifelong student of Zen and editor of the works of Dogen, never used formal za-zen. But his “Zen practice” was precisely his study of physics, and to suggest his attitude he used to say that his lifework was “to science” rather than “to study science.”

Widespread

Confucianism, in other words, preoccupies itself with conventional knowledge, and under its auspices children are brought up so that their originally wayward and whimsical natures are made to fit the Procrustean bed of the social order. The individual defines himself and his place in society in terms of the Confucian formulae. Taoism, on the other hand, is generally a pursuit of older men, and especially of men who are retiring from active life in the community. Their retirement from society is a kind of outward symbol of an inward liberation from the bounds of conventional patterns of thought and conduct. For Taoism concerns itself with unconventional knowledge. with the understanding of life directly. …Confucianism presides, then, over the socially necessary task of forcing the original spontaneity of life into the rigid rules of convention-a task which involves not only conflict and pain, but also the loss of that peculiar naturalness and un-self-consciousness for which little children are so much loved, and which is sometimes regained by saints and sages. The function of Taoism is to undo the inevitable damage of this discipline, and not only to restore but also to develop the original spontaneity, which is termed tzu-jan or “self-so-ness.”

Popularity almost invariably leads to a deterioration of quality, and as Zen became less of an informal spiritual movement and more of a settled institution, it underwent a curious change of character. It became necessary to “standardize” its methods and to find means for the masters to handle students in large numbers…For the Zen community became less an association of mature men with spiritual interests, and more of an ecclesiastical boarding school for adolescent boys. Under such circumstances the problem of discipline became paramount. The Zen masters were forced to concern themselves not only with the way of liberation from convention, but also with the instilling of convention, of ordinary manners and morals, in raw youths. The mature Western student who discovers an interest in Zen as a philosophy or as a way of liberation must be careful to keep this in mind, for otherwise he may be unpleasantly startled by monastic Zen as it exists today in Japan. He will find that Zen is a discipline enforced with the big stick. He will find that, although it is still an effective way of liberation at its “upper end,” its main preoccupation is with a disciplinary regimen which “trains character” in the same way as the old-fashioned British public school or the Jesuit novitiate.

For one of the main objects of this work was to establish a proper “apostolic succession” for the Zen tradition, so that no one could claim authority unless his satori had been approved by someone who had been approved … right back to the time of the Buddha himself

Nothing, however, is more difficult than establishing proper qualifications in the imponderable realm of spiritual insight. Where the candidates are few the problem is not so grave, but where one master is responsible for some hundreds of students the process of teaching and testing requires standardization. Zen solved this problem with remarkable ingenuity, employing a means which not only provided a test of competence but–what was much more–a means of transmitting the Zen experience itself with a minimum of falsification. This extraordinary invention was the system of the kung-an ee (Japanese, koan) or “Zen problem.” Literally, this term means a “public document” or “case” in the sense of a decision creating a legal precedent. Thus the koan system involves “passing” a series of tests based on the mondo or anecdotes of the old masters.

Art

Even in painting, the work of art is considered not only as representing nature but as being itself a work of nature

When you paint it is the brush, ink, and paper which determine the result as much as your own hand.

Since writing and poetry were among the chief preoccupations of Chinese scholars, and since the Chinese way of painting is closely akin to writing, the roles of scholar, artist, and poet were not widely separated

Since the touch of the brush is so light and fluid, and since it must move continuously over the absorbent paper if the ink is to flow out regularly, its control requires a free movement of the hand and arm as if one were dancing rather than writing on paper. In short, it is a perfect instrument for the expression of unhesitating spontaneity, and a single stroke is enough to “give away” one’s character to an experienced observer.

One of the most striking features of the Sung landscape, as of sumi-e as a whole, is the relative emptiness of the picture–an emptiness which appears, however, to be part of the painting and not just unpainted background. By filling in just one corner, the artist makes the whole area of the picture alive. Ma-yüan, in particular, was a master of this technique, which amounts almost to “painting by not painting,” or what Zen sometimes calls “playing the string-less lute.” The secret lies in knowing how to balance form with emptiness and, above all, in knowing when one has “said” enough. For Zen spoils neither the aesthetic shock nor the satori shock by filling in, by explanation, second thoughts, and intellectual commentary.

The Western eye is immediately struck by the absence of symmetry in these paintings, by the consistent avoidance of regular and geometrical shapes, whether straight or curved. For the characteristic brush line is jagged, gnarled, irregularly twisting, dashing, or sweeping–always spontaneous rather than predictable. Even when the Zen monk or artist draws a solitary circle-one of the most common themes of zenga–it is not only slightly eccentric and out of shape, but the very texture of the line is full of life and verve with the incidental splashes and gaps of the “rough brush.”

But the non-Japanese listener must remember that a good haiku is a pebble thrown into the pool of the listener’s mind, evoking associations out of the richness of his own memory. It invites the listener to participate instead of leaving him dumb with admiration while the poet shows off

Even when robbed he is still rich, for: The thief Left it behind The moon at the window.

They were always sparing and reserved in their use of color, as were the Sung painters before them, since masses of flowers in sharply varying colors are seldom found in the state of nature.

Herrigel spent almost five years trying to find the right way of releasing the bowstring, for it had to be done “unintentionally,” in the same way as a ripe fruit bursts its skin… After all those years of practice there came a day when it just happened–how, or why, Herrigel never understood.

The artificial haiku always feels like a piece of life which has been deliberately broken off or wrenched away from the universe, whereas the genuine haiku has dropped off all by itself, and has the whole universe inside it.

The point, therefore, of these arts is the doing of them rather than the accomplishments. But, more than this, the real joy of them lies in what turns up unintentionally in the course of practice, just as the joy of travel is not nearly so much in getting where one wants to go as in the unsought surprises which occur on the journey.

Numerical Stability in Flash Attention

jarbus@tutanota.com (jarbus) — Thu, 27 Jul 2023 14:21:08 -0400

Flash attention, a recent implementation of attention which makes less calls to high-bandwidth memory, uses a version of the softmax function which is numerically stable. In this post, I’ll briefly showcase how this is done and an example of an unstable softmax.

The softmax function is used in machine learning to convert a vector of real numbers to a vector of probabilities which sum to 1, and is defined as:

softmax(x) = [exp(x[i]) / sum([exp(xj) for xj in x]) for xi in x]

where x is a vector of real numbers.

The python implementation below is numerically unstable because it involves exponentiation of large numbers, which can lead to overflow. Crucially, underflow is not an issue, because exp(x) approaches zero when x is a large negative number.

import numpy as np
def unstable_softmax(x):
 fx_unstable = np.exp(x)
 return fx_unstable / np.sum(fx_unstable)

The following implementation is stable however, because there is no exponentiation of large numbers:

def stable_softmax(x):
 fx_stable = x - np.max(x)
 return np.exp(fx_stable) / np.sum(np.exp(fx_stable))

Instead, the max of the vector is subtracted from each element. This does not change the result of the softmax after the division as this subtraction is also performed in the denominator, thus cancelling out.

Let’s compare the two implementations:

>>> a = np.array([6.0, -3, 15], dtype=np.float32)
>>> stable_softmax(a)
[1.2339458e-04 1.5228101e-08 9.9987662e-01]
>>> unstable_softmax(a)
[1.2339458e-04 1.5228101e-08 9.9987656e-01]

As you can see, the results are mostly equal, save for a few digits. Now let’s look at what happens with 16 bits of precision:

>>> a = np.array([6.0, -3, 15], dtype=np.float16)
>>> unstable_softmax(a)
[ 0. 0. nan]
>>> stable_softmax(a)
[ 1.234e-04 0.000e+00 1.000e+00]

When working with 16 bits of precision, we observe that exp(15) produces a numerical overflow which turns the third element into a NaN. This is because exp(15) produces a value that can not be represented by a float16.

To recap, we showed that softmax is numerically unstable, especially when working with small precision bits. Because softmax uses exponentials in the numerator and denominator, we can subtract all exponents by the maximum exponent, constraining all the values between 0 and 1 and preventing numerical overflow.

Introducing Kittyplot

jarbus@tutanota.com (jarbus) — Sat, 20 May 2023 21:09:16 -0400

Kittyplot is a program designed to plot experiment data in the kitty terminal using the kitty graphics protocol, primarily for use on HPC clusters.

Plots are rendered using matplotlib, and users can zoom into different regions of the plots by setting x and y limits using their editor. I use prompt_toolkit to accept regexp input and I override the tab-completion to instead display a list of all metrics that are matched by the current regexp.

This will not be a well-maintained repo for the forseeable future and not a simple library one can pip install. It’s merely a public copy of the script I use for visualizing experiment data over SSH without port-forwarding or running a web browser on my local machine. Kittyplot is only a few hunderd lines and is designed to be directly modified for user needs.

I was initially inspired to make kittyplot when I got fed up with tensorboard taking 20+ minutes to load all the experiment data I was working with. I loved tensorboard solely for the purpose of monitoring metrics specified by a regexp, so I modified my experiments to stream all the metrics to a csv and wrote kittyplot to monitor them. Now, I can create, view, and save plots directly from my ssh connection.

Owen, Make Your Website Already

jarbus@tutanota.com (jarbus) — Mon, 23 Jan 2023 11:47:35 -0500

Make a fucking website already, Owen. You’ve been playing around with the idea of making a blog for like two years, but you’ve been dragging your feet. Maybe you can’t decide what framework to use, or where to host it, or blah blah blah. If you want excuses not to make a site, I’m sure you’ll find them.

But it literally just takes a few hours, less if you don’t set up a domain name or theme to start. You can do all that later. The important thing is getting it online, the rest is easy (and satisfying!) once you have the core site made.

You can use HUGO to start and change frameworks later, since all your content will be in markdown. You can host on GitHub pages to start, and always change later, since again, it takes no time to set up and there’s no lock-in from what I can tell.

I can understand not wanting to deal with all the technical stuff like domain names or self-hosting or choosing a VPS or making it look pretty. But come on man, at least get started. The rest is so much easier to do once you are up and running.

Unexpected Benefits of Testing Code

jarbus@tutanota.com (jarbus) — Sat, 21 Jan 2023 18:38:43 -0500

Matthew Carlson’s blog post “Fighting Distraction With Unit Tests” inspired me to share some extra benefits of writing test code I’ve discovered during my PhD program.

I’m working on a weird project that’s constantly changing as I try new things, and naturally, debugging and ensuring correctness was a nightmare. So I started writing tests, cursing myself for needing to write so much code I’ll likely throw away soon. But as it turns out, testing can be pretty helpful in a few other ways:

When I encounter a bug I didn’t anticipate, I already have most of the code needed to reproduce it in a deterministic manner.
When I want to add/change functionality, I already have most of the code needed ensure that my changes work.
When writing new functionality, if I don’t already have the code needed to make ensure it’s working, I can write a small test that runs whenever my code changes. Making this test case auto-run on a second screen dramatically reduces the amount of time spent switching windows and lets me stay in my editor until the test passes. I highly recommend setting up a run-on-change system, it makes writing tests or designing plots pretty fun.
You can write tests for newer code by using tests for older code. This makes tests valuable as you develop, even if you plan to throw them out later.
I feel more confident about what my code is actually doing. As someone who is working in AI, the value of this is immense, as AI code can run without crashing but still be subtly incorrect.

I’m not writing an exhaustive test suite, or tests for edge-cases that shouldn’t happen (asserts can deal with those). I’m writing code before I get bugs to spare myself the extra effort when I need to debug or make changes.

I Dont Know Anyone Who Blogs

jarbus@tutanota.com (jarbus) — Wed, 18 Jan 2023 21:46:47 -0500

I don’t know anybody in real life who blogs regularly, and I think that kills my motivation to write blog posts myself.

I actually started writing this post before realizing that a friend of mine has a blog, but in my defense he’s only posted twice in the past 2 years. I suppose that makes my title a lie, but I like the title, and the point still stands.

I want to build a writing habit, but it feels hard to get the motivation to do it from within myself. I’m a programmer, and while I’ve learned a lot from people online, the people who’ve made the biggest impact on me are people I’ve met in the physical world. Plus, I’m surrounded by other programmers; maybe that makes me feel less crazy spending all my time coding.

When I mention my blog to people, or send my friends my latest post, I always feel a sense of shame. Like I’m trying too hard. Maybe I am. I’m talking to nobody and have little-to-no audience, so in a sense, I’d have to be crazy to make a habit out of this.

Perhaps I’m just not surrounded by people who make me feel less crazy. I imagine that if I was making digital music but everyone around me was some mainstream profession like an accountant or baker, I’d probably feel embarrassed too, even though there’s nothing to be embarrassed about. But, if I was surrounded by people who also made music, I’d feel normal. Maybe it’s just the feeling of “trying hard to do something that nobody else cares enough to do”.

I’ve read a lot about the benefits of blogging; Writing helps flesh out ideas, improves communication, puts down what you are thinking about to reference later, helps connect you to other people, gets you into your dream job, cures you of cancer, prevents global warming, and saves the bees. But somehow, the bees aren’t enough motivation to write.

In all the blog posts talking about why you should blog, nobody mentions this feeling of being “the guy with the blog”. I imagine other people feel this way, which is why I’m writing this, I suppose. Hopefully I make writing a habit in 2023.

How to write LaTeX without writing LaTeX

jarbus@tutanota.com (jarbus) — Fri, 17 Dec 2021 14:28:41 -0500

I love the look of LaTeX but hate the experience of writing in LaTeX, at least compared to Markdown. Luckily, Pandoc can convert Markdown files to PDFs using a LaTeX engine as the renderer, and includes a custom Markdown specification that can fill almost all my LaTeX needs.

First, let’s talk about where Pandoc Markdown falls short:

No custom LaTeX style guides (although citation style files are supported)
~~No Section Numbering~~ There is, thanks to naruhodo on Hacker News for the correction
Referencing labels doesn’t work well (Supposedly the pandoc-crossref filter fixes this but I couldn’t get it to work)
Don’t even bother with complex page layouts or precise figure placements

This might be a deal breaker for some, but for others who are writing lots of documents (such as students), this may not be. Now for the benefits:

Produce a document with Markdown that looks like it was written in LaTeX
Use LaTeX math notation in Markdown
Easy BibTeX citations in Markdown
Figures and Captions are supported
Two Columns are supported
Use your favorite editor instead of some LaTeX IDE
Write and preview offline
Build a document in one command

If you are reading this, you probably know how both Markdown and Latex work, and you probably know how to read the Pandoc Markdown Documentation, so instead, below is a template document that I wish I had when first starting out. It covers most things you’d need to write simple papers in Markdown. Configuration for document settings is done in the YAML block at the top of the document.

---
title: How to Succeed in LaTeX without even trying
subtitle: For the lazy writer
author:
- jarbus.net
numbersections: true
toc: true
geometry:
- margin=1in
linkcolor: black
urlcolor: blue
bibliography: example.bib
csl: nature.csl
header-includes: |
\usepackage{package_here}
classoption:
- twocolumn
abstract: |
This is the abstract
---
# Introduction
According to all known laws
of aviation, there is now way
that a bee should be able to fly. [@seinfeld2007]
| "This is a quote"
![Caption here](bee.png){width=50% }
$\forall b , y = mx+b$
$$
a^2 + b^2 = c^2
$$
# References
::: {#refs}
:::
# Appendix

Then, to compile your document to a PDF, simply run:

pandoc "file.md" -o "file.pdf" --citeproc

to generate a document that looks like this:

The --citeproc option is needed for compiling a bibliography. For cross-references inside a document, take a look at pandoc-crossref, which will require the use of --filter pandoc-crossref if it works for you.

I mapped a modified version of the build command above to a key in Neovim, so now pressing Space+b in my Markdown buffer compiles my document automatically, no matter what file name I’m using:

autocmd BufRead,BufNewFile *.md nnoremap <Leader>b
\ :silent pandoc % -o "%:p:h/%:t:r.pdf"
\ --filter pandoc-crossref --citeproc &<CR><CR>

And that’s how I write LaTeX documents offline in Markdown.

AI Index

jarbus@tutanota.com (jarbus) — Fri, 19 Mar 2021 11:25:49 -0400

An ever-expanding list of concepts in the field of AI to give myself and others an easy reference. Each item in the list contains a short, rudimentary definition I’ve written, as well as a link to a resource that can explain it better.

Ablation Study: Removing some parts of a machine learning model to measure impact on performance

Advantage Function: The difference between a Q-value for a state-action pair and a value for the state. Useful to determine how good an action is relative to its state.

Attention: Neural networks are able to “pay attention” to specific parts of input or output, useful in translating languages or predicting sequences. For example, when trying to predict the next word in “the clouds in the”, you pay attention to the word cloud to predict the word sky

Autoencoder^W: A type of neural network that attempts to take raw data, convert it into a simpler representation (usually by limiting the amount of nodes in a hidden layer for representation), and then decode the representation back into it’s data. They are primarly used to extract the useful properties from data automatically. They can do this in an unsupervised fashion, since their output targets are the given input, requiring no labeling.

AutoML^W: Systems where the entire machine learning process, from data preparation to network design, is automated.

Autoregression: A time series model that uses observations from previous time steps to predict the value at the next time step.

Backpropogation^W: The algorithm and calculus behind gradient descent traditionally used in feed-forward neural networks

Bayesian: An interpretation of probability where the phrase “probability” expresses a degree of belief between 0 or 1, 0 representing false and 1 representing true. Uses expected values and random variables.

Bootstrapping^W: Sampling from a distribution to approximate a function. In cases of reinforcement learning, bootstrapping usually samples potential future values to approximate a current value.

Convolutional Neural Network (CNN)^W: A neural network primarly used for image processing. These networks design filters for specific parts of an image to extract higher level information, filters such as detecting a certain type of edge.

Covariate Shift: The training distribution changes but the testing distribution does not change

Data Mining: Discovering knowledge and patterns from massive amounts of data, usually in an unsupervised fashion.

Deep Learning^W: A subset of machine learning with a multi-step learning process, usually referring to neural networks with two or more layers.

Discount Factors: A variable (usually $\gamma$) that determines how much a model cares about rewards in the distant future compared to immediate rewards.

$\gamma = 0$ cares only about immediate reward,
$\gamma = 1$ cares only about sum of all future rewards.

Eligibility Trace: For temporal learning, the eligibility trace is a vector of decaying values that represent when weights were last used. When we encounter an error, this vector allows us to update recent weights harder than weights used long ago.

Feed-forward Neural Network^W: A simple neural network where information is passed strictly from one layer to the next.

Generative Adversarial Network (GAN)^W: A set of two or more neural networks that can generate new data based on existing training data. A simple example is https://thispersondoesnotexist.com/, that can generate fake pictures of humans. - GANs consist of one generator network with the goal to make realistic new data, and a distinction network with the goal to tell real data from fake. As the networks train, they each improve at their individual task, forcing their adversaries to improve in turn.

Genetic Algorithms^W: Algorithms that try to mimic the evolutionary process by randomly modifying the best-performing sets of parameters while discarding those with the worst performance, then repeating.

GPT-2/3: A language model from OpenAI that can generate text that mimics a writing style incredibly well.

Gradient Descent^W: An interative process that attempts to find a minimum of a function that works by moving in the direction that will decrease the gradient until a local minima is reached.

Hyperparameters^W: Manually defined parameters of the model, such as the size of a neural network, or manually defined parameters of the machine learning algorithm, such as learning rate.

(IID) Independent and Identically Distributed^W: Random variables are independent and identically distributed when the value of one variable doesn’t affect the probability of another variable. For example, the outcome of one coin flip does not affect the outcome of another coin flip, so both flips are IID.

KL Divergence: Divergence between two distributions of data. Useful for determining how different fake data is from real data (GANs) or for determining how differemt two policies are for trust-based reinforcement learning.

Level 0-5^W: The different “levels” of autonomy related to self driving cars. 0 represents full human control while 5 represents full vehicle control.

Long Short Term Memory (LSTM)^W: A type of recurrent neural network that works exceptionally well with sequential input. A unique trait of these networks are their memory cells’ “forget gates”, which allow them to control how long they hold onto data for.

The Lottery Ticket Hypothesis: A randomly-initialized, dense neural network contains a subnetwork that is initialised such that — when trained in isolation — it can match the test accuracy of the original network after training for at most the same number of iterations.

Markov Decision Process^W: A system of states, actions, and probabilities of getting to other states given actions taken from previous states.

Mode Collapse: When the distribution of samples produced by a generative adversarial network represent a subset of the latent distribution, instead of the entire latent distribution. For example, you train a network to produce pictures of animals but it only learns to produce pictures of cats.

Multi-armed Bandit^W: A core component of reinforcement learning, the multi-armed bandit problem is the classic “exploration versus exploitation” tradeoff. In this problem, expected gain must be maximized in an environment with varying rewards, forcing an agent to decide between keeping an option they know to be safe versus exploring new options that might be better.

Online/Offline Learning: Online learning happens as data comes in. Offline learning happens after data is collected and made into a batch.

Parametric & Non-Parametric Models: Parametric models use a finite number of parameters, like weights in linear regression, to represent a learned hypothesis. Non-Parametric models use variable/infinite/no parameters, like data points in nearest neighbors, to represent a learned hypothesis.

Precision^W: The fraction of relevant instances among the retrieved instances. Also known as positive predictive value.

Recall^W: The fraction of relevant instances that were retrieved. Also known as sensitivity.

Recurrent Neural Network (RNN)^W: A type of network that can store state, giving it a type of memory that can process a series of inputs. This can be accomplished by having a cycle within the network.

Regression^W: A set of models that determine relationships between data and a dependent value. For example, linear regression tries to approximate a dependent value while logistic regression tries to determine the probability of a dependent value.

Residual Neural Network^W: Networks with connections that skip some layers and connect to non-adjacent ones. This type of network helps counter the vanishing gradient problem

Reinforcement Learning^W: Algorithms are given a world state they are able to interact with, and learn from the reward their interactions give them.

Model-Based: You create a model of the world and can predict what the next state and reward will be for each action
Model-Free: You know what action to take without knowing what to expect, since you don’t have a model of the world
Value: Networks that determine the value of a state.
Policy: Network to choose actions. Can directly optimize model instead of computing values, useful when you have a continuous action space. Only uses reward function. Requires a score function to evaluate performance of policy, usually total rewards accumulated given a period of time.
Actor Critic: Backbone of state of the art reinforcement learning algorithms.
- Uses a value-based Critic to measure how good the action taken is
- Uses a policy-based Actor to to choose actions

Stochastic^W: Randomly determined process, usually refers to probabilistic outputs of machine learning systems

Supervised Learning^W: A model learns to produce a desired output and knows what that output is. Example: Image Recognition

Temporal Difference^W: Model-free reinforcement learning design which learns by bootstrapping value samples in order to approximate a value function. Once more information is revealed about the true value during later timesteps, you can update the low information bootstrapped guess by using the newly acquired outcome as a “ground truth” to train a network.

Temporal Difference error is the difference between consecutive temporal predictions.

Trajectories: The history of states (and potentially actions) taken during a walk of a Markov Decision Process.

Transfer Learning^W: Taking parts of a network trained on one data set, and using it in a different network with a different dataset.

Transformer: A type of recurrent neural network primarily used with sequential data, like language translation. These networks use an attention model to improve performance.

Unsupervised Learning^W: A model learns to produce a desired output without being told what it’s looking for. Example: Playing Chess

Vanishing Gradient Problem^W: In a network, a gradient can become vanishingly small which will stop the weight from changing it’s value, since weights are modified based on their contribution to the gradient.

Feel free to contact me with any suggested additions/changes at jarbus@tutanota.com.

Generating Slides in Vim

jarbus@tutanota.com (jarbus) — Sun, 31 Jan 2021 11:25:49 -0400

There is a great tool known as pandoc that can convert documents from one filetype to another. For example, you can convert a Microsoft Word document to a PDF, without even needing to own a copy of Microsoft Word! However, we care about Pandoc’s ability to convert a Markdown document to a slideshow presentation using LaTeX Beamer as a rendering engine.

There is a great writeup about this basic feature here.

TL;DR: With Pandoc installed and markdown file Demo.md, executing pandoc -t beamer Demo.md -o Demo.pdf will generate a slideshow as a pdf, with each section heading as the title of a new slide.

Different themes can be specified with the -V option; I’m currently using the Metropolis beamer theme and Owl colorscheme. After installing these themes, I can run:

pandoc -t beamer
-V fontsize=14pt
-V theme=metropolis
-V colortheme=owl
-s
Demo.md -o Demo.pdf

to produce my presentation. Note: All code blocks are single-line commands/.vimrc entries, and are displayed as multi-line for readability.

Unfortunately, it’s a pain to recompile a document every time I want to view my changes. Luckily, as a (Neo)Vim user, I can automatically run commands everytime I write a file. To avoid generating presentations with non-presentation markdown files, I instead write presentations in plaintext files that end with .slides. In the example above, I’d be using Demo.slides instead of Demo.md. When I add:

autocmd! BufWritePost *.slides silent !pandoc
-t beamer
-V fontsize=14pt
-V theme=metropolis
-V colortheme=owl
-s
% -o "%:p:h/.%:t:r.pdf"

to my .vimrc, everytime I save changes in a FILENAME.slides file, it will generate a presentation located in .FILENAME.pdf. I choose to add a leading period in the output to hide the PDF from my file manager, feel free to remove it.

To enable markdown syntax highlighting in our .slides file, add:

autocmd BufRead,BufNewFile *.slides :set filetype=markdown

to your .vimrc.

Now Vim generates a PDF presentation for us everytime we save a .slides file. For easy viewing, I added another keybind, Z, to open the corresponding PDF in my PDF viewer of choice, Zathura:

nnoremap Z :silent !zathura --fork %:p:h/.%:t:r.pdf<CR>

Zathura automatically refreshes the view whenever changes to an open document is detected. This means that I can open a .slides file in Vim, save changes, and press Z to view my changes in Zathura. I can then immediately view any additional changes I make in the open Zathura window, just by saving the .slides file.

Uninformed Or Misinformed?

jarbus@tutanota.com (jarbus) — Wed, 20 Jan 2021 11:25:49 -0400

Is it better to have no facts about a topic and thus no opinion, or have access to a few “out of context” facts and thus a misinformed opinion?

TL;DR: Either start learning about issues you care about, or stop caring about them so much. It’s more honest, less mental effort, and is less likely to ruin Thanksgiving.

You can’t be well-informed on every topic; It’s just not possible. That’s why world leaders have advisers. Yet somehow, if you look at anyone’s Facebook account, you don’t have to try too hard to figure out their view on any political issue.

Over the past year, I’ve started to realize how much nuance there is to the headlines we see. When I read an article that ruins the image of a company or individual, I always wonder: What information could the author be omitting? What view could the other side have? What outside information is the author conveniently using to make a point?

I try to avoid getting emotional until these questions are answered. I’m guilty of having basic gut feelings on unfamiliar topics, but these feelings aren’t strong enough to start an argument. I talk about issues I’m informed about, and listen (with a grain of salt) to others talk about issues I’m not.

Unfortunately, this isn’t the case for lots of us. Instead of admitting we are uninformed, many of us form an opinion after we read the headline, and most likely don’t update it after skimming the article with said headline. We are even less likely to look for an opposing take, as that requires we scroll outside of whatever feed we are browsing through. Some people don’t even use the news to form opinions: They just accept the opinions of their friends or family as “obvious”. As someone who accepted the outcome of the 2020 presidential election without looking into the claims of fraud, I’m guilty of this too.

We then take this opinion that a) isn’t ours and b) took less than 10 seconds to form, and carry it into the foreseeable future, not even considering the idea that there might be more we don’t know.

What’s wrong with admitting that we don’t know enough to judge a topic or event? Do we really think we have enough context and facts to make sound judgements on all our beliefs? Do we really need to hold an opinion after every scandal? Why can’t we just say “This is a side of a story” and move on?

Take Google’s firing of Timnit Gebru for example:

I read a headline about Google firing a researcher that criticized the company on a mailing list. I didn’t need about any more drama in my life, so I didn’t read further, tucking this away in my mind as “researcher fired from Google”.
Google responded with something along the lines of “she was using an email list in such a way that was improper for her position”. I still didn’t need the drama, so I kept it in my mind as “researcher fired from Google”.
I still hold no strong positions on this specific topic and don’t plan on learning more about it. Like in most “scandals”, there’s a lot of “he said, she said” going on, and I don’t need to spend time figuring out who is right.

If I wanted to hold an opinion on this event, I’d need the following details to start:

The code of conduct for Gebru’s position
Other alternative methods Gebru could have used not in violation of the Code of Conduct if her actions violated it
The contents of the email
Verbose arguments from both sides refuting each other’s points if possible
Evidence of Gebru’s claims

As someone who does not work at nor plans to work at Google, this just isn’t worth my time, so I don’t form an opinion. There are other things to worry about.

TL;DR (again): Either start learning about issues you care about, or stop caring about them so much. It’s more honest, less mental effort, and is less likely to ruin Thanksgiving.

Soft Rules

jarbus@tutanota.com (jarbus) — Sat, 26 Sep 2020 11:25:49 -0400

Not all rules are made equal. Some rules, like the speed limit, are broken by everyone and their mother, and are enforced (mostly) when people are breaking them to a dangerous point. Other rules, like “don’t murder people”, are enforced far more. Why is that?

Different Rules, Different Tools

Some rules are in place to assign blame to someone with bad judgement. There’s nothing morally wrong with jaywalking when there are no cars coming, but you are at fault if you walk into oncoming traffic.
Some rules are in place to discourage bad habits from forming, like the drinking age. This law is so commonly broken that it’s expected behavior at this point, and yet I’ve never seen someone get arrested for under-age drinking. Despite this, the law conveys the dangers of drinking, makes it harder for minors to acquire alcohol, and prevents students from openly sharing alcohol in schools.
Some rules, like mandatory attendance, exist to get people to show up to an event or perform an activity. Even when attendance isn’t taken, many people will still show up to prevent unnecessary trouble. College orientation activities are a prime example.
Some rules exist simply because they haven’t been updated. Before 2020, some companies – especially older ones – didn’t allow employees to work from home. Working from home was seen as an unnecessary risk by upper management with no tangible benefit to any of the executives. Along comes the coronavirus, and suddenly the jobs people have been doing in an office for decades are suddenly doable from home.

Rules for Rules

Rules should be justified. If a sign on a fish tank says, PLEASE DO NOT TOUCH GLASS, do you think most people, especially children, are likely to obey? Does your answer change if the sign were to say PLEASE DO NOT TOUCH GLASS, IT SCARES THE FISH?

Additionally, there shouldn’t be too many rules, even if they are justified. The more rules in place, the less likely people will follow them. Just think about it: If Moses received 206 commandments, do you think his followers would view Thou shalt not eat meat and dairy with the same level of respect as Thou shalt not kill? More rules can imply that there are strict guidelines to be followed, but people will interpret them as: Whoever came up with this must be fun at parties. Unnecessary rules draw attention away from the important ones.

Wrapping Up Rules

I’m no rule expert. Nevertheless, I recognize that creating and selectively enforcing rules is a valuable skill, for both policy-makers and policy-enforcers alike. If you want people to respect your rules, make sure they have a reason for being there.

Critical Thinking

jarbus@tutanota.com (jarbus) — Tue, 22 Sep 2020 11:25:49 -0400

The more I learn about the world, the less certain I become about it. There are so many conflicting views and information taken out of context that it’s hard to discern fact from truth. Below are some notes I keep to help navigate the sea of information that is modern society.

TL;DR: Don’t get fooled by facts if they have no context, don’t discount an opinion because it’s biased, and hear out crazy ideas.

Be skeptical: Fact != Truth

There are facts available for virtually anything you need. No matter what argument you try to make, you’ll be able to find a statistic that can back up your claims thanks to the glory of Google.

“Did you know that planking can ruin your health and result in months of pain?”

“Not true, planking works out multiple muscles in your body and can greatly improve your health if done daily”

Not exactly scientific sources, but I think this demonstrates my point. Furthermore, the level of mixed information increases exponentially as the topic gets more political.

Facts should never be enough to prove a point, and I tend to suspect people who rely too heavily on them. Facts should instead be used to tell a narrative; facts are not the truth, but they can light the way towards it.

“The top 0.1% … own about the same wealth as the bottom 90% of America.”

While deemed “Mostly True” by Politifact.com, this statement in isolation is less than useless: it’s misleading. On it’s own, this fact paints a picture of a grim society, where the bottom 90% are left to struggle for survival, with all resources being held for a greedy few. If you are in the bottom 90%, this prompts you to believe that life is unfair, and that this should be changed. That’s why this statistic is so popular, it seems outrageous and gives Americans a sense of justified anger at the rich, an anger which many ex-presidential candidates used to garner support.

Let’s break this idea down a bit. The greedy elite factoid on its own implies nothing about quality of life, standard of living, financial responsibility, or determination to work. Society could be a paradise where everyone has access to more than they could ever need, and the top 0.1% just has an absurd amount of money. Or society could be a system of addiction, where the bottom 90% are constantly spending money on things they don’t need and the top 0.1% simply aren’t addicted and save/invest wisely. The truth is, these numbers don’t reveal anything about the truth, and aren’t being used to do so.

Consider a narrative discussing the implications of this statistic from Wikipedia:

A 2014 study by researchers at Princeton and Northwestern concludes that government policies reflect the desires of the wealthy, and that the vast majority of American citizens have “minuscule, near-zero, statistically non-significant impact upon public policy … when a majority of citizens disagrees with economic elites and/or with organized interests, they generally lose.” When Fed chair Janet Yellen was questioned by Bernie Sanders about the study at a congressional hearing in May 2014, she responded “There’s no question that we’ve had a trend toward growing inequality” and that this trend “can shape [and] determine the ability of different groups to participate equally in a democracy and have grave effects on social stability over time.”

Most people know about inequality, but not what it means for them. The wealth statistic means nothing in a vaccuum, only in context does it hold any value. If facts are used to invoke emotion instead of prove a point, be skeptical.

Worth Listening to those who are biased?

A girl in a class of mine had a laptop sticker that read “Why be racist, sexist, or transphobic when you can just be quiet?”

I can’t stand this type of thinking. It’s elitist, tells people with controversial views to not let their voice be heard, and contributes to the echo chamber of modern society. Furthermore, it accomplishes nothing except a pat on the back for people with her mindset. No racist is actually going to read that and go “Oh, guess I just won’t speak my mind then.” In fact, they might be driven further into racism in spite. Nobody likes being talked down to.

There’s a lot of insight to be gained on both sides from listening to people with views you don’t like. It’s easy to think white supremacists are evil and don’t deserve to speak; its harder to realize they have a reason for their beliefs. If you want to make the world a better place, it doesn’t come from silence, but from understanding. When fighting racism, wouldn’t it make sense to understand where racists are coming from?

If you want to discuss tax law, you should hear out the rich guy that wants lower taxes just as much as the working class American who wants the same. If you want to discuss diversity, don’t limit the white participants to those with an “anti-white” mindset. These perspectives are essential to promoting an open dialogue, and leaving out those with biased opinions leads to toxic echo chambers and a mentality of “us versus them”.

Hear out the crazy

When I experimented with fasting, a few people I knew were opposed. The concept of not eating for days on end? Ridiculous! Your body needs food, your metabolism will slow to a halt if you don’t eat!

These arguments didn’t come from any sort of reading on the subject, but human intuition. The concept of not eating for a few days went against everything they knew; to them it was crazy.

Fasting isn’t actually bad for you if done right, and can have numerous health benefits. A simple Google search into the subject could have cleared up these misconceptions. But that would imply that this ridiculous idea could have some merit worth looking for. It’s like looking up “Did the Holocaust happen?”

When it comes to most domains however, there is usually some fascinating logic behind crazy ideas. Bitcoin comes to mind; a currency without a government to back it seems completely absurd on the surface. The gut reaction is to call Bitcoin a fraud, but when the best in the business take some time to think about it, they might have second thoughts. The Bitcoin community is full of brilliant people that built the foundation of blockchain technology. It’s worth hearing them out, even if their ideas seem a bit crazy at first glance. They must have their reasons.

Final Thoughts

As the news cycle continues to speed up, it’s hard to determine fact from fiction, who’s crazy and who’s biased, what’s conspiracy and what’s revolutionary. It’s impossible to listen to everyone, but you shouldn’t only listen to those with a similar mindset. It’s impossible to get the full story for every topic of discussion, but don’t fall for out-of-context information. Question everything, but keep an open mind.

Tesla and False Advertising in AI

jarbus@tutanota.com (jarbus) — Wed, 22 Jul 2020 11:25:49 -0400

Here’s the problem with advertising AI-based technology that doesn’t exist:

You cannot promise anything about your product.

We’ve all seen AI advertised to the masses that doesn’t work as advertised, just look at any voice-to-text system. When I got my Apple Watch, I hoped to use it to respond to messages without getting distracted by my phone. I quickly realized that wasn’t a viable solution: I had to repeat my message multiple times per text in order to get the correct dictation.

If this was a software issue, I could file a bug report and Apple could send out a fix. Unfortunately with AI, there isn’t much Apple can do besides get more data and re-train the model, which still wouldn’t guarantee a better user experience for anyone. Mind you, this is with software that’s already shipped.

Now let’s look at Tesla, or more specifically its CEO, Elon Musk. Musk claims that Level 5 driving (in which a human is not needed behind the wheel) will be coming to Tesla vehicles in the form of a software update by the end of 2020. A month before he made said claim, a Tesla on autopilot crashed into a flipped truck–a mistake most humans wouldn’t make.

Just like Apple can’t promise its speech-to-text system will work better than a human (or even better than a toddler), Musk can’t guarantee anything about Tesla’s Level 5 capability. Tesla can’t even “reasonably claim” your car won’t crash into flipped trucks or stopped cop cars. In fact, no one using deep learning can promise anything about their system. Deep learning, and by extension Tesla’s Autopilot system, is a black box. We don’t know why it works, we can’t promise that it’ll work, but it seems to work when we test it. Unless Tesla is testing on all roads in all conditions, they shouldn’t even be “confident” level 5 technology, in which a human does not need to be in the vehicle, is “around the corner”. That’s why a court in Germany banned Tesla from including “full potential for autonomous driving” and “Autopilot inclusive” in its German advertising materials.

A Full Self-Driving (FSD) update wouldn’t be the first time Tesla over-promises and under-delivers. Consumer Reports described the company’s “Smart Summon” feature as “glitchy and at times worked intermittently”. The damage with Smart Summon has been low, but a full-autonomy update could result in a few consequences:

Tesla sold cars with the promise of Level 5 driving in the future: Not only is Tesla now under immense pressure to rush out a Level 5 update as soon as possible, but must do so using the hardware it sold customers years ago. Hopefully Level 5 doesn’t require LIDAR sensors (or any additional hardware) like Elon claims; because Tesla is expected to ship some version of FSD to users.
Customers may end up in legal trouble if Tesla’s software fails them: Tesla avoids legal responsibility for buggy software by shifting responsibility of an accident to the driver. However, when Tesla sells millions of cars on the premise that users won’t have to manually drive them one day, should drivers be held responsible when that day comes and their car crashes?

Other companies have realized the high standards a Level 5 system should meet. Two years ago, Waymo CEO John Krafcik claimed that true self-driving cars, that function everywhere under all circumstances, won’t be ubiquitous for decades. As a result, Waymo shifted gears and placed its focus on rolling out autonomous taxis on a city by city basis, starting in Pheonix, Arizona. This approach to Level 5 driving is much safer for the following reasons:

Waymo is minimizing the number of people at risk
Waymo is operating its vehicles strictly where they have been tested.
Waymo is not selling consumers a technology that doesn’t exist
Waymo is not relying on the average consumer to be attentive

Level 5 technology can save thousands of lives and millions of dollars–but it needs to be done to the same (if not higher) standards of automotive testing, and not the “move fast and break things” approach of Silicon Valley. Hopefully Tesla can safely deliver on its promise of full self driving soon, but if not, consumers may soon have a case to demand a refund.

Short: Aftermath

jarbus@tutanota.com (jarbus) — Mon, 27 Apr 2020 11:25:49 -0400

Jeff stepped over a beer can, then over a water bottle and onto a mat, whose faded lettering vaguely resembled the phrase “Welcome!” He opened his front door and walked inside.

Looking to his right, Jeff noticed his old hooded friend where he left him; on his living-room rocking chair. The cold metal scythe still in his lap, the skeleton hadn’t moved an inch from where he left him earlier that morning.

“What’s cooking, big guy?” Jeff said as he walked to the kitchen to grab his last can of Budweiser. His boney companioned turned his head.

“Oh, you know”, Death emitted. “still out of work.”

“Ah.”

“I do not understand why you still haven’t given up yet.”

“I could say the same about you.”

“I exist for one purpose, Jeffrey. and as long as you are alive, I cannot rest.”

“I don’t know, you seem pretty well rested to me, what with you never leaving my house and all.”

“I used to have what you call a ‘job’. It was very fulfilling, Jeffery. Took up all of my time. Do you want to know what it was?”

Jeffery took a sip of his beer.

“I used to collect the souls of the dead, Jeffery. It was wonderful. I conversed with the rich, the poor, the young and the old, Jeffrey, as they contemplated life and where they were headed. I had, what you would call, a ’nice thing going.’”

“All good things gotta come to an end.”

“Then you came along,” Death continued, ignoring him, “and ruined everything.”

“It be like that sometimes.”

“It does not ‘be’ like that sometimes, Jeffrey” Death glared, a purple glow forming in his eye sockets. “A single human does not wipe out humanity ’every now and then’”

“Fair.”

“What is even more despicable,” Death proceeded, “is that you don’t even regret pushing the button.”

“I don’t really think about it too much.”

“YOU HAD A CHOICE,” Death boomed, “BETWEEN SAVING HUMANKIND, AND SAVING YOURSELF–”

“Mhm.”

“AND WITHOUT A SECOND THOUGHT, DOOMED YOUR ENTIRE SPECIES–”

“Yup.”

“–IN EXCHANGE FOR A LIFE OF THEFT AND SLUMBER.”

“Can’t steal from the dead,” Jeff rebutted, taking another sip.

“THAT IS NOT THE POINT, JEFFREY” Death bellowed. He paused, taking a few seconds to decompose himself. “You fail to understand the magnitude of your actions.”

“Oh, I understand perfectly well, you’ve made sure of that. I’d have to be deaf to spend a day with you and not hear your whining.”

“It’s hard to believe you even have a soul worth collecting.”

“Listen, it was self defense, alright? It’s not like I wanted to kill everyone.”

“Adolf Hitler, Joseph Stalin, Osama Bin Laden; every single corrupt soul in your history books would not have stooped to your level; self-defense does not justify genocide.”

“So humans can drive thousands of species to extinction for selfish gain, but only when it happens to them it’s unjustified.”

Death moved as if to rebutt, but said nothing.

“Do you know how close we came to nuclear war? There was never a guarantee of tomorrow for anyone. For all we know, the human race might last an extra few decades thanks to me.” Jeff said, smirking. “Why does humanity deserves the benefit of the doubt?”

“Why do you?” Death said.

“I don’t necessarily,” Jeff said, taking a seat on the opposite couch. “It just happened to be my choice. My parents were already dead–I didn’t owe anybody anything. What moral rule says I have to lay down my life for those I don’t know?”

“You are selfish, Jeffrey Johnson.” Death said, rising from his seat. “You put yourself in front of millions of good, innocent people who would lay down their lives in your situation.”

“Is it selfish, or simply not being selfless? A starving man doesn’t owe another starving man his bread.”

Death leaned forward, skull to face with Jeff, searching for any sign of remorse. He came back empty. Death turned around drifted towards the exit.

“He owes the guilt of denying a meal,” he remarked as he left the room. The front door creaked open, then closed. Silence filled the room.

“Should I instead spend humanity’s last years crying?” Jeff yelled to the door.

The door didn’t respond.

Happiness

jarbus@tutanota.com (jarbus) — Sun, 26 Jan 2020 11:25:49 -0400

Some thoughts on Happiness:

Happiness doesn’t come from wealth. Otherwise, therapists and anti-depressants would have to be much cheaper.

If happiness came from consumption of goods and services, it would therefore follow that once those goods and services are reduced or cut off, one would no longer be as happy. If you can cut off happiness, it’s more akin to a drug than an emotion.

“Close relationships, more than money or fame, are what keep people happy throughout their lives.” While financial security is ideal for easing pain and suffering, significant amounts of money can strain relationships and cast doubt on who your friends really are. If solid relationships are a path to happiness, the drama and expectations caused from wealth differences in the relationships might very well block that path.

In a sense, happiness isn’t something you can obtain, but something you become. True happiness has to come from within to be strong and sustainable. Performing work you find meaningful, engaging in relationships of substance, and living a life of virtue are all ways to approach happiness.

Happiness can’t come from believing you are the smartest, most attractive, or funniest person in your social circles. If it does, you either aren’t around people you can learn from or are living in a world of illusion.

If stressed about all those that are above you, happiness is possible if you enjoy the process of learning and take pride in trying your best.

“It’s possible to be a masterpiece and a work in progress at the same time.” - Unknown

Pain and happiness are not mutually exclusive, but rather two sides of the same coin. The greatest pains can come from the greatest happiness, such as the loss of a loved one. Some of the greatest happiness comes from the greatest pain, such as starting new relationships after others have ended. The most painful moments can sometimes fuel transitions into the best ones. Don’t let pain go to waste.

One of the best ways to increase happiness is to show gratitude. When expressing thanks for what you have instead of longing for what you don’t, you learn to draw happiness from yourself instead of the high of getting something new. If you were to lose that for which you are thankful, the situation doesn’t change – draw gratitude from what’s left.

Bitcoin Whitepaper, Explained

jarbus@tutanota.com (jarbus) — Fri, 03 Jan 2020 11:25:49 -0400

Introduction

In this essay we will give a brief history of crytocurrency leading up to Bitcoin, give an overview of the Bitcoin protocol by summarizing key sections of the whitepaper, and briefly discuss Bitcoin’s use of cryptographical proof and computational security instead of trusted third parties within the protocol.

History

Bitcoin is far from original. Digital currencies date back as far as 1982, where David Chaum released a paper called "Blind Signatures", which formed the basis of a digital currency known as eCash. In 1996, Douglass Jackson created a currency known as e-gold in 1996, the first currency to reach mass adoption at over 3.5 million accounts. Many other subsequent currencies rose and fell before Bitcoin gained stable footing, including projects such as BitGold and b-money.

Blockchain technology development began as early as 1991 when Stuart Haber and W. Scott Stornetta began work on a cryptographically secured chain of blocks. Their inital goal was to implement a system where document timestamps could not be tampered with. Bitcoin builds upon Haber and Stornetta’s subsequent work of reliable timestamping and securing names for bit strings and cites three of their papers, making up a third of the nine citations in the Bitcoin whitepaper.

Bitcoin also builds upon technology such as Proof-of-Work, first developed in 1997 by Adam Back under the name Hashcash. Proof-of-Work is a counter-measure towards abuse of un-metered internet resources which could enable a Denial of Service attack on email by requiring an attacker to expend computational resources in order to interact with the network. A reusable version of Proof-Of-Work was further developed by Hal Finney in 2004, the receiver of the first bitcoin transaction. Proof-Of-Work will be further discussed later on in this paper.

Satoshi Nakamoto, the anonymous creator(s) of Bitcoin released the Bitcoin whitepaper to a cryptography mailing list on October 31st, 2008, four short years after Hal Finney’s paper. The first block was mined January the following year, with a comment that reads:

“The Times 03/Jan/2009 Chancellor on brink of second bailout for banks”

Very fitting for the anti-authority ideals Bitcoin would spur in the coming years.

Design

Section 2 and 9: Transactions, Combining and Splitting Value

Bitcoin is represented as a publicly visible global log of signed$^W$ transactions. Each transaction denotes the participants in a transaction as well as the amount of money transfered (analogous to the Venmo public feed), and the signature ensures the account publishing the transaction is the account who owns the Bitcoin. Since the amount of Bitcoin an account has is directly represented by outputs of previous transactions, sending money is represented by taking previous transactions addressed to your account, sigining them, and specifying which addresses will receive which portions of the funds. Since one or more transactions can be used as input for this process, there can be one or more addresses receiving the output, including the spender who recieves any left-over Bitcoin as change, since it’s impossible to use a fraction of a transaction as input.

Transactions must be verified, i.e. nodes must confirm that transaction signatures come from their corresponding private keys. Transactions are verified in large batches. Each transaction is added to a pool of unverified transactions, which are then all verified and published to the network in a block, approximately once every ten minutes. Each block is published with a timestamp and a hash of a previous block, forming a “chain” of all valid transactions.

Sections 3 and 4: Timestamp server, Proof of Work

Bitcoin also implements a distributed timestamp server. Normal timestamp servers take hashes of previous timestamps and the current block in order to form a new timestamp hash, then widely publish this hash. However, this requires a trusted party, and does not work in a distributed setting with no trusted parties.

Nakamoto takes inspiration from Adam Back’s Hashcash algorithm, which implements a Proof-Of-Work scheme based off of finding a hash with a specific number of zeros at the beginning. This is a very computationally expensive task, as the only method of obtaining such a hash is brute force. Since a new block is published every ten minutes and each block requires the hash of a previous block, this requires anyone attmempting to modify the blockchain history to not only find the proper hash at the specific time, but also all hashes of subsequent blocks in order to maintain the validity of the ledger. This type of attack can be used to spend the same Bitcoin twice, as a user could make a transaction, erase it from the blockchain, and then use it again. However, thanks to the proof-of-work algorithm, this attack increases in difficultly exponentially as new blocks are added, and is one of the many forms of computational security the Bitcoin protocol implements. Since the amount of computational power on the blockchain will naturally vary overtime, the proof-of-work algorithm will also adjust to compensate for increased/decreased mining power in order to keep blocks published approximately every ten minutes.

The first block verified and published is added to the network, and the longest chain of blocks is deemed to be the correct one.

Section 5: Network

The Bitcoin network runs on the following steps taken verbatim from the original whitepaper, as there is not much summary for this section I deem myself capable of doing better than Nakamoto.

New transactions are broadcast to all nodes
Each node collects new transactions into a block
Each node works on finding a difficult proof-of-work for its block
When a node finds a proof-of-work, it broadcasts the block to all nodes
Nodes accept the block only if all transactions in it are valid and not already spent
Nodes express their acceptance of the block by working on creating the next block in the chain, using the hash of the accepted block as the previous hash.

Nodes always consider the longest chain to be the correct one and will keep working on extending it.

Sections 6-8: Incentive, Reclaiming Disk Space, Simplified Payment Verification

Section six explains the financial benefits of mining, namely how miners get newly minted coins and transaction fees. Nakamoto then explains how even when an adversary can claim control over a majority of the network’s mining power, it makes economic sense to act according to the protocol to gain profit instead of creating double-spend attacks, since people will stop using the system once it is compromised.

In section seven, Nakamoto also discusses storage optimizations for reducing the amount of disk space the entire blockchain uses. The primary method discussed involved removing transactions that are buried underneath many blocks but hashing the transactions in a Merkle tree as to break the hashes of any blocks.

In section eight, Nakamoto describes a method of payment verification without needing to run a full node, as well as potential attacks and countermeasures on nodes attempting to verify transactions without storing the entire blockchain.

Section 10: Privacy

It should be noted that Bitcoin has no privacy protections in place to protect users besides basic anonymity of not tying real-world identities to keys. The amount of Bitcoin sent in each transaction is public, as well as the public keys used. New public-private keypairs are recommended for each transaction, but the multi-input nature of transactions reveals two inputs are attached to the same owner. As a result, if the identity of a sender is compromised, it’s possible to trace transaction history and link previous transactions that user has made.

Section 11: Cryptography and Security

Attackers are unable to steal the money of other users without their private key. There is also no method to generate new Bitcoin in anyway that an honest node could recognize as valid. Therefore, the only feasible attack on Bitcoin is to generate an alternative chain of blocks. Since the longest chain of blocks is considered valid, this attack requires the generation of an alternative chain faster than the honest chain. This can be thought of as a race between the attacker and the rest of the network, with the attacker starting from behind since they have to re-compute proof of work for existing blocks.

Nakamoto demonstrates how an attacker with $51%$ of the networks hashing power will be guaranteed to eventually catch up and sustain an alternative chain, however given a sufficiently large network Nakamoto assumes this infeasilbe, as the computation power required would be enormous. Additionally, every time an adversary fails to beat an honest node in creating the next block, the amount of blocks they need to catch up to increases and thus unless an attacker gets incredibly lucky when they first start generating an alternate chain, their chances of catching up to the main chain decreases exponentially. Since the chance for an attacker to catch up to the honest chain decreases exponentially as new blocks are added, Nakamoto demonstrates how long an honest recipient must wait before they can be 99.9% certain a double-spend attack will not happen given attackers of various computational power. For example, if attackers control 10% of the mining power, one would need to merely wait five blocks, but if attackers have 45%, one would need to wait 340 blocks, or just under two and a half days, to be sure their transaction is safe.

Resources Used

Of course the Bitcoin whitepaper and it’s references have been used in the writing of this essay, but I also wanted to credit this Quora post for pointing author in the right direction for the second section.

Anonymity

jarbus@tutanota.com (jarbus) — Thu, 02 Jan 2020 11:25:49 -0400

What happened to “Don’t give out your name on the internet?” Within the past 15 years, the desire for online anonymity has vanished – as “normal people” joined the digital world, the number of “creepy people” worth hiding your identity from seems to have dropped in comparison. Despite the perceived safety of the online world, remaining anonymous still has its merits.

Anonymity addresses ideas

In an ideal world, political debate occurs detached from the people debating the topic. When discussing the gender wage gap for example, fact-based discussions should be possible between any two individuals, regardless of gender, salary, or race. Productive discourse where different interpretations of the same statistics should occur, where arguents, not participants, are rebutted and questioned. This ideal world does not exist.

Instead, when arguments are backed by identities, it’s much easier to attack an argument by attacking the arguer. It’s too easy to claim someone doesn’t know what they are talking about, since they are too old or too young, too priveleged or too uneducated. Political elections are perhaps the best examples of this, especially when candidates attack past actions of their opponents to avoid attacking their values.

An idea with no group or person attributed to it is the truest form of an idea, for it exists on its own and must be taken down for what it is, not who is behind it.

Anonymity ensures security

Freedom of speech ensures individuals can not be legally punished for publicly expressing their opinion. However, when opinions are tied to identities, the government is not needed to punish those with different views. Citizens can take matters into their own hands if given the target’s place of work and the ability to write a dishonest complaint. A quote taken out of context and a social network to spread it can end a career. Given the rapid spread of data in society, the best way to ensure your security is to spread as little of your information as possible, and to separate whatever you share from each other.

Anonymity ensures authenticity

“Give a man a mask and he’ll show his true face” is a favorite quote of mine. It can act as an attack against anonymity, claiming people will act badly if able to avoid social and legal backlash. This can be interpreted in a different way; people have the freedom to be who they are when they don’t have to worry about backlash. Is it just to not feel safe showing one’s true face in modern society?

Surveillance acts partially as a way to collect information on a populus, and partially as a way to enforce good behavior; if you see a camera at a stoplight, you are more likely to act in good behavior whether or not the camera is actually on. In an ideal world, everyone acts in good behavior and gets along, but on their own terms and not because they are being watched. Remaining anonymous ensures good behavior is not done in hopes of receiving praise or gaining social acceptance, but rather for the sake of being good. An example could be comparing a person who drives safely in case there might be a cop versus a person who drives safely because they don’t wish to risk anyone’s life.

TL;DR

An anonymous idea is stronger than a non-anonymous idea, as it lacks the weaknesses associated with having a human (and thus flawed) orator.
A person is always safer when they aren’t publicly tied to any comments or statements.
A person has the freedom to be who they are when their identity isn’t behind their actions.

Forgiveness

jarbus@tutanota.com (jarbus) — Sat, 27 Jul 2019 11:25:49 -0400

Forgiveness has a negative connotation these days. We see forgiving each other as weak — as not standing by one’s values, as giving up one’s pride. This view is incredibly counter-productive and, in my opinion, the cause of many problems in today’s society. Here is the actual definition of forgiveness:

forgive (verb)

to cease to feel resentment against (an offender)

Source: Merriam Webster

This definition is all that forgiveness has to be — not pardoning or excusing someone for their actions, but instead letting go of anger towards them. Grudges may appear to make someone “tough” and “strong,” but all they do is keep someone’s mind tormented by the people they despise. It is very easy to hold a grudge—grudges make one feel powerful without requiring them to do anything. However, true power in one’s life stems from overcoming this negativity and taking control of one’s emotions.

Forgiving Others

Forgiving others is much easier than it sounds and is one of the quickest and simplest ways to improve your mental state. This will be more challenging for deep, long-term grudges, but you can displace surface-level anger as easily as saying, “[Person], I forgive you” aloud—even when alone. Forgiveness helps get negative people out of your head who don’t deserve to be there. It also helps you understand others instead of hating them.

Letting go of blame/resentment helps focus on finding solutions, instead of blaming others for problems—this results in better decision making. Your thought process goes from, “If only this person didn’t do that, everything would be so much better” to “This happened, what can I do about it?” It’s not possible to control the actions of others, but we can control our own responses. This attitude sharpens social skills, as you learn to respond maturely instead of reacting passionately when upset.

Forgiving Yourself

What does forgiving yourself mean? In line with our definition from above, it means to cease self-resentment. Many toxic behaviors stem from insecurities and regrets people have. If people were content with themselves and their accomplishments, the world would be a kinder place. Self forgiveness can be a much more difficult task than forgiving others (since you can’t exactly put yourself behind you), but can reap the greatest benefits. Self forgiveness helps guide one’s life towards inner peace and self-love and away from a dependency on external stimuli (such as degrading others) for happiness.

One of the best ways to help forgive yourself for shortcomings is to work on improving them. Hate how you look? Exercise! Hate being socially incompetent? Practice talking to people! The amount of effort you put into change doesn’t have to be significant, as long as it’s consistent. You should be able to say, “I don’t like this aspect of myself but that’s ok, because I’m taking steps to improve.” This will increase self-esteem and thus increase happiness in your life.

An alternative to changing who you are is accepting your flaws. No one is perfect! This skill is difficult to master, but promotes loving who you are instead of trying to change into who you want to be. Self-acceptance doesn’t require you to change everything you despise about yourself, but also can lead to a life of mediocrity where you always settle for “good enough”. I find a bit of each mindset to be ideal—I try to improve the qualities I can control and accept the ones I can’t.

If you hold self-resentment for past actions, try talking to a professional counselor/therapist! They are skilled at helping patients work through complex and specific issues. They suggest solutions you might not have thought of and call you out when you need it, making your problems much easier to tackle.

Recap & Final Thoughts

Forgiveness means letting go of resentment, not justifying an action or person.
It is easier to stay angry than to forgive.
It is not stronger to stay angry than to forgive.
Forgiveness increases confidence.
Forgiveness improves communication.
Forgiveness clears the mind.
Forgiveness allows you to learn from others.
Forgiveness in some aspects of life help you forgive in others.
Forgiveness makes the world kinder.
“I don’t like myself because I am inferior,” can turn into “I love myself because I’m trying to grow.”
“If only this didn’t happen,” can turn into “This happened, how do we respond?”

Note that this writing isn’t citing scientific studies or telling anecdotes—it is presenting a set of ideas for the reader to consider. These beliefs do not cover the entire subject nor do they universally hold.

Balance

jarbus@tutanota.com (jarbus) — Tue, 09 Jul 2019 11:25:49 -0400

Choosing sides is fun. Be it politics or pop culture, nothing validates an opinion better than classifying opposing opinions as wrong. Sometimes, as seen by Holocaust deniers or the anti-vaccine movement, it’s possible people can be completely and utterly wrong. However, in the general case, both sides have merits and both sides have pitfalls. When trying to gain a perspective of the bigger picture, it’s best to position oneself somewhere in the middle.

This versus that

We live in a world with seven billion perspectives. Seven billion lifetimes of thought and contemplation. Of course, one cannot listen to every lifestory nor hear every person’s opinion. So we form into groups, with each important group representing something between the loudest minority or general majority of people with a particular set of opinions. This is a system able to properly distinguish emerging schools of thought while categorizing people into said schools, enabling people to efficiently organize, communicate, and compare ideas.

This does not imply that everyone in a group has well formed opinions. Quite the opposite in fact – these groups require people to give considerably less thought into their beliefs. Immersing oneself in a group for a few ideas can easily lead to naturally picking up other beliefs of the group, as many people will adapt for acceptance. This lower barrier to entry for taking sides is a core part of the conflict between groups, as society seems to reinforce beliefs by ridiculing opponents who don’t fully understand theirs – consider the plethora of “Person X DESTROYS believer of Y” videos online.

Given a large enough group however, there tend to be many people who have given careful time and consideration into why they believe what they believe. Thus, one cannot simply sweep a prominent idea under the rug because of a few bad apples. To do so is to throw away ideas which stood the test of time and the people behind them. If proposed a ridiculous sounding idea, merely dismissing it without consideration denies you of a potentially meaningful realization. Instead of “No, that’s wrong”, all parties benefit more from “Why?” The pursuit of knowledge would have ended a long time ago if we dismissed everything we didn’t believe.

Entertaining Thoughts

An invaluable skill that many successful people have is the ability to play with an idea without accepting it. This is essential for multiple reasons:

By entertaining an idea, you can learn key takeaways from views that may appear ridiculous at first
You can reinforce your beliefs by attempting (and failing) to convince yourself of an opposing view

Additionally, many people are absolutely rubbish at laying out a convincing argument. Requiring the persuasion of other people to change your mind will result in your mind rarely being changed. Sometimes, the best person to convince you of something is in fact, you.

Final Thoughts

In summary:

Don’t discredit an idea based on the people proposing it
Don’t discredit a person based on the group they belong to
Doing so inhibits learning for both parties
If someone can’t convince you of something, try convincing yourself

This is not a fully fleshed out idea of mine, nor is this idea exclusively mine. There are both counter arguments and room for expansion, however I think this is best left up to the reader.

jarbus@tutanota.com (jarbus) — Mon, 01 Jan 0001 00:00:00 +0000

jarviz

(jar)bus audio (vis)ualizer

built with webGPU, runs locally.

0:00 / 0:00

jarbus@tutanota.com (jarbus) — Mon, 01 Jan 0001 00:00:00 +0000

Jitter - Image Dithering Tool

jither: Jack's dithering tool.

Runs locally using web assembly.

Upload Image:

Dithering Strength

Pixelation Level 1x

jarbus.net on

Typing Speed Might Matter Now

Real-Life Superpowers

The Only Computer I Need

Mapping Control and Escape to Capslock

I Accidentally Wrote a Novella

Fiction: Growth and Decay

0x0d-I: March 13th, 2181, 10:00pm

0x01-I: April 2nd, 2061, 8:43pm

0x01-II: April 3nd, 2061, 3:00am

0x06-I: July 10th, 2111, 1:43pm

0x02-I: September 4th, 2071, 4:32am

0x02-II: September 4th, 2071, 12:31pm

0x04-III July 4th, 2094 3:02pm

0x04-IV August 12th, 2096 10:02am

0x04-VI February 18th, 2097 1:27pm

0x01-X March 22nd, 2071 2:01am

0x06-IX: October 5th, 2118, 1:11am

0x06-IV: September 1st, 2116, 1:02 am

0x06-V: August 3rd, 2117, 12:59pm

0x06-VII: October 3rd, 2118, 4:07pm

0x06-VIII: October 4th, 2118, 6:25pm

0x06-X: October 6th, 2118, 10:00am

0x02-X February 14th, 2081, 11:00pm

0x05-II April 1st, 2101, 2:32pm

0x07-III July 8th, 2123 4:15pm

0x07-II September 23rd, 2121 9:30am

0x0d-IV: May 4th 2187, 8:00pm

0x03 January 31st, 2088 7:05pm

0x04-V September 30th, 2096 11:00pm

0x07-VI December 25th, 2128 10:00am

0x0d-V: October 9th 2187, 1:00am

0x07-X February 04th, 2121, 10:00am

0x0d-V: December 8th 2187, 1:30pm

0x0d-VI: January 28th 2188, 11:30pm

0x08 ???

0x0e-II March 3rd, 2192, 6:44pm

0x0e-III October 4th, 2192, 10:32am

0x0e-IV May 5th, 2194, 5:01pm

0x0e-VII May 18th, 2198, 2:14pm

0x0f-I May 5th, 2201, 4:22pm

0x0f-II July 22nd, 2204, 11:47pm

0x0f-III August 2nd, 2204, 11:30am

0x09-IV November 5th, 2144, 1:02pm

0x0f-IX January 30th, 2211, 9:44am

0x0f-X, February 2nd, 2211, 8:50am

My AMD Framework 13 Upgrade Experience

Everything Matters

Low-Rank Factorizations are Indirect Encodings for Deep Neuroevolution

To the Students

NON-TECHNICAL

TECHNICAL

AI

NOT AI

Read Weird Things

Documentation

Primary Sources

Books and Textbooks

From REINFORCE to R1: an Abridged Genealogy of Reinforcement Learning

Table of Contents

Reinforcement Learning 101

REINFORCE

From REINFORCE to Actor-Critic

From Actor-Critic to Advantage Actor-Critic (A2C)

From Advantage Actor-Critic to Proximal Policy Optimization (PPO)

From Proximal Policy Optimization to Group Relative Policy Optimization

Eloquence and Wit from Will Durant

Introduction

Communism

Government

Women

Morals

Language

Appearance

Sumeria

Egypt

Babylon

Assyria

Judea

Persia