I am a Memoryless System

Link post

Author’s Note: this is my entry for the Edit Your Source Code Contest. I am an undergrad student at RIT, NY.

commit 85a4c4f37966a739e88c0a4c70946bd4 (HEAD -> master)
Author: demo
Date: Thu Oct 27 09:38:12 2022 -0400
Initial mind state

I’m sitting in Intro Psych on a Thursday, I think, and I’m staring at a black square with my mind on it.

Well, not technically “a black square with my mind on it”. It was more of a rectangle, with the little flip-out keyboard coming out the bottom of the shiny screen. And my mind wasn’t “on” the device. For most people, their mind is stored in their brain, more or less. Mine was just 7 miles away, on a cluster of cloud servers.

On my phone, a story open on reddit. Stocks down a bit. Emissions.

The people at Brainle, LLC’s AI lab made this PDA-laptop thingy, and I got in early on the beta. They also own the servers with my mind on them. I type commands and touch the screen of my PDA, those go to the servers, my mind changes. “My brain” is just a chip implanted in my skull, mirroring what’s on those servers. And “I” am sitting, as usual, with myself at my fingertips and zero clue what to do.


commit 564309daf5b1d832367a8cc330a5ba93 (HEAD -> master)
Author: nkross
Date: Thu Oct 27 10:58:37 2022 -0400
test

It’s near the end of class, and I still haven’t pushed a real change to my source code. 90 minutes have passed since I got it at the nurse’s office. I notice I am frustrated myself for wasting 90 minutes, wasting any minutes, wasting a second, wasting there goes another second staring at the PDA. And no ideas for how to change my mind.

That sounds really dumb, wouldn’t I jump at the chance to modify my brain’s code? But see, I have ADHD. Not the fun kind, where you take cocaine-and-skydiving lessons and grow up to be a billionaire or a crime lord. No, I have the “inattentive” type, where you sit in a pile of laundry and scroll through reddit for 9 hours a day.

But I can think the thoughts of a cocaine-and-skydiving guy, and that’s what pisses me off. I can have a cool idea, make a plan, remember some helpful info I read, and then… I waste the day. The mind is willing, but the brain is weak, like all fleshy things.

Now people are getting up and putting shit in their backpacks. I should too. I should also get my life in order, but one thing at a time. Typical Nick, gets every chance and forgets to grab it. For all I sit around thinking and stewing, nothing actually happens until I do it. The kind of person who would do it… courageous? Smart? Energetic? No, then someone else in the Brainle beta would’ve hacked their dopamine and gotten a PhD by now. It’s only been 7 days. What do I need to fix?


commit 59bca29a290ae161390887144d87f74f (HEAD -> master)
Author: nkross
Date: Thu Oct 27 11:10:58 2022 -0400
WM increase

Remember that TED talk by the loud funny kid? The autistic prodigy, Jacob Barnett, who does quantum physics? Well his Mom wrote a book about his life. Towards the end, she takes the kid to a professional IQ test, and one thing the expert says is that the kid has nigh-unlimited working memory. He reads something once, he remembers it forever. Reciting 999 digits of Pi, backwards? “It’s like reading them off a sheet of paper that’s right in front of you… Jake’s paper is the size of a football field.” Paraphrasing, but basically that.

So obviously, my first edit was to zoom in on my prefrontal cortex and add a couple zeros to my working memory capacity. Actually, more than a few zeros, since we can’t store that much in our little mental workspace at the same time.

So now I can work through the professor’s Macro-Econ lecture as she speaks. Instead of writing in my notes “GDP, inflation, currency reserves, CHECK SLIDES FOR FORMULAS”, I can just imagine the formulas as she explains them.

Wait, did she say increase or decrease? Shit. She said something else while I asked myself that. Shit, more words, more thoughts. Guhhhhhhh.

OK, I have extremely high working memory, I should be able to think of a solution to this conundrum. Load in past concepts from long-term memory, identify problem domain, create mental entities abstracting to the important bits, recall that time I read The Goal… I have a hypothesis. My working memory is my bottleneck, but that was past “me”, not “my” bottleneck now. The words I hear can be represented in working memory, but if I don’t think to listen to them in the first place, they don’t go in.

I consider making my next edit to my focus, but I’ve seen where that can lead thanks to my medication. If I turn focus on now, I could be thinking about GDP and currency reserves for hours. That’s not good on the margin!


Maybe I set the bar too high at first. The working memory helps, but it’s not everything. What do I want?

I’m in college because → I can get credentials and experience related to computer science, so that → I can land a tech job, so that → I can get money. And money’s useful for a lot of other things; if I get enough, most of the other stuff will probably fall into place.

That’s a lot of steps, and I’m pretty impatient. Again, ADHD. If I want money, why don’t I just look for it directly?

Some people cheat at video games with a special hacking software, called a “tracer”. Let’s say you’re playing Minecraft. Where are the diamonds? You’re supposed to look for them underground, which takes hours. A “tracer” is a thin line that stretches from your eyes all the way to the nearest diamonds. No exploration required, just follow the line.

Can I make a money tracer? Like, for dollar bills lying on the ground? Load in generic problem-solving strategies, Polya’s How to Solve It, engineering blog posts I’ve read, Fermi estimation, every scrap of knowledge that could tell me where dollar bills are. Encoding problem: what constitutes a “dollar bill”? The abstraction of just training a neural network on it. Images. Street view… how up to date are the satellite scans on Brainle Maps? Priors on green and thin and paper, maybe somebody’s already coded something to… ah, yes, maybe the TSA needs to look for weed automatically. Are those models on the Internet? (That’s dumb, why would they give out that info?) Can I pay some people to label images of crumpled dollars lying on a sidewalk? Fuck, no, I need more money for that kind of study, that’s why I’m looking for bills on the ground.


commit bc747db1406f0f52b7f49228a456e0e1 (HEAD -> master)
Author: nkross
Date: Thu Oct 27 14:38:06 2022 -0400
subagent setup

Instead of paying some people to label images, I created a subagent. A new mind within my mind. He’s a copy of me, though, so he got bored labeling images of sidewalks with dollar bills. But he told me something smarter: why not just make subagents to do things that would earn me money?

John D Rockefeller once said something like “I prefer 1% of each man’s work from 100 men, rather than 100% of 1 man’s work.” Instead of one (or two) copies of me doing something, I should make 100 simple subagents, each doing one measly little intellectual task. One guy does “have ideas”, another does “generate mathematical world-models”, another does “create metrics”, another “test models”, and so on. Parallel computing speeds up the whole operation on Brainle’s servers. And doesn’t the no-free-lunch-theorem say something about general algorithms being impossible? Well, I’ll just make a bunch of narrow algorithms!

Thus, my new plan:

  1. Scrap the tracer idea.

  2. Work on the most profitable no-effort thing I can think of, stock trading.

  3. Create subagents to do different parts of stock trading.

  4. Profit!

commit 548bc8e959fe3d3ab7b6a14f26a2ee55 (HEAD -> master)
Author: nkross
Date: Thu Oct 27 14:57:33 2022 -0400
more subagent setup

On the walk back to my dorm, I spin up a subagent instance to code the boilerplate for the other subagents. I am not spending my whole class doing that again. That subagent takes my instructions: one subagent for stock market models, one for downloading financial data, one for testing strategies, one for trading.


commit 564ac7fb1ef2bf03777351e48cbb3292 (HEAD -> master)
Author: nkross
Date: Thu Oct 27 15:30:29 2022 -0400
refactor
commit b6e9cd90786b50766855666ebb3e2a1c (HEAD -> master)
Author: nkross
Date: Thu Oct 27 15:57:33 2022 -0400
removed vol from model
commit 29ae729b248f8bc8222183662fcf1a42 (HEAD -> master)
Author: nkross
Date: Thu Oct 27 17:25:35 2022 -0400
CAGR hacked, rollback
commit ffdee88d14df154446c940e53ecc7baf (HEAD -> master)
Author: nkross
Date: Thu Oct 27 20:24:16 2022 -0400
weather API added
commit 37d71c3c7bdbd29c3e04b9e127cada4e (HEAD -> master)
Author: nkross
Date: Thu Oct 27 20:50:01 2022 -0400
added vol back into model

You can’t buy more than 100% of a company’s stock.

You can’t cheat by looking at the price before the subagent sees it.

You can’t do infinite trades with no costs.

You can’t do infinite trades, period.

You can’t cheat by leaving your training data in the test set.

reddit is not a source of information.

Every metric sucks. Every model sucks. They don’t see what I want.

Every time I write a new metric, I’m like “this captures all the profits, avoids all the drawbacks, satisfies my intuitions, is provably correct given my assumptions, and adjusts for the friggin orbit of Saturn.” And I train my strategy on it. And I test it in real trading. And it sucks.

A smarter man would’ve seen through the space of all possibilities, and written something that works the first time. I have spent the last 5 hours trying to beat the market, losing money the whole time. And it’s because my subagents aren’t me.

Now, I’ve read the tweets. “Changing the goal is bad”. But I always know my goal, and the subagents don’t. Fucking tired of this. Do X to do Y to get expected Z, fudge factor for S. I just want.


commit d26b58e172f5ba2ca4d558d7dd41683d (HEAD -> master)
Author: nkross
Date: Thu Oct 27 20:55:00 2022 -0400
reward func change

I have edited my core desires optimize for whatever gets me as rich as possible, ASAP.

The reasoning is simple:

  • My subagents fail because they don’t share my goals.

  • I fail because my goals are unfocused.

  • Money is a robust goal under lots of different scenarios.

  • Setting money as terminal, at least temporarily, makes it more likely to actually be achieved.

  • I’ll remember to change the goal again later.


wm dump

Ceiling is thin but no neighbors. Is it the Sleeping Beauty paradox? If I wake up, am I/​subagent there? If in prison? Slow down runtime, subjective time increase. No risk to core me. Core me wants money. Risk to compounding. Door sealed on my way out. I sealed the door on my way out. The door was sealed, by me, on my way out. Is this why people one-box on Newcomb’s problem? They won’t win, but they sort themselves into alternate timelines where they win the absolute most? Information leaked through the walls. I cleaned the walls. Subagent to do the boring physical tasks, like scrubbing or lunging. God’s drunkest information processor. Signals no cryptologists can find. Bodies smell? Brainle doesn’t monitor smell. Who gives their employees a smell detector? “Remote workers, can’t trust ’em.” He had nothing. Guessed the wrong key-owner. Police? Does Brainle monitor police? Police-is-a-(being-that-can-(smell)). Not through walls. Brainle must have tripwires. They have adults in the room. There are no adults. I must sort myself into timelines where they couldn’t find the body before I owned them. Italics are unprofessional, don’t leave them in the code. Did they see me lie? Did I encrypt myself in time? Does their server check for that? Do they know what blood tastes like? Do they taste blood? Can satisfied customers use tongues? Unsatisfied customers? Bayes on discriminate between the two groups. Backchain (that’s still not trivial, update against calling it this often) from success. Lever. Cortes. The /​r/​WhoWouldWin questions where they absorb somebody else’s powers. Can they yes, they can. I should’ve asked for his accounts first. What if he’s not even the security guy? Deception job titles, randomized to disguise who actually does what. Options? No, he doesn’t cash them out yet. Can’t lever, can’t sell. Bedroom drawers? Gold? 2 days to clear. Git has rollbacks, would I notice if they killed me? Focus, reallocate mental options to high returns. Sort myself into worlds where they shut up and I get. Take. Consume is dumb, compound. Win on speed. Execute.


A: another day, another wire tripped.

B: This is the third time this month

B: *week

B: and you know that will speed up.

A: No

A: you read too many doomer bloggers.

A: The tripwires (which we don’t need) keep going off on the dumbest things. Then we have to make the user experience worse to fill standards

B: Better safe than sorry. - Abraham Lincoln

A: “Safe”, when every day we delay the app is another day people remain idiots. They lovingly

A: *literally dont see the isomorphisms. Smarter agents = better at satisfying their goals = more goals satisfied = better world.

B: u said the 3rd user on the waitlist would never get that good at chess.

A: I never said that. I said Elo is dumb, because Elo is a dumb metric.

B: The tripwires do measure real progress.

A: ok but

A: we know when things go wrong

A: and we shut it off

A: you’re gonna say “but muh hacking” as if they were gods

B: speaking of gods, isn’t Jeff coming back from his missionary trip?

A: he was supposed to. Hes not answering texts. Zoomer

B: zoomer, but also boomer

A: Xennial

B: nobody can pronounce that.


git reset HEAD@85a4c4f37966a739e88c0a4c70946bd4 --msecret --logtrip

commit 85a4c4f37966a739e88c0a4c70946bd4 (HEAD -> master)
Author: demo
Date: Thu Oct 27 09:38:12 2022 -0400
Initial mind state

I’m sitting in Chem-101 on a Friday, and I’m staring at a black square with my brain on it.

Well, not technically “a black square with my brain on it”. It’s a representation of my brain, an image showing some clusters of what are probably neurons (why wouldn’t they be?). This is some beta hardware I got from Brainle. I can touch the screen, type in some commands, and change my mind.

On my phone, a story open on reddit. Stocks down a lot. More emissions. Missing person, don’t care, didn’t ask.

I look up at the latest commit. “Thu Oct 27”. A full day to play with the exclusive beta for the Brainle mind-modification pipeline. And without remembering how, I wasted it.

That sounds really dumb, wouldn’t I jump at the chance to modify my brain’s code? But see, I have ADHD. Not the fun kind, where you take cocaine-and-skydiving lessons and grow up to be a billionaire or a crime lord. No, I have the “inattentive” type, where you sit in a pile of laundry and scroll through reddit for 9 hours a day.

But I can think the thoughts of a cocaine-and-skydiving guy, and that’s what pisses me off. I can have a cool idea, make a plan, remember some helpful info I read, and then… I waste the day. The mind is willing, but the brain is weak, like all fleshy things.

Now people are getting up and putting shit in their backpacks. I should too. I should also get my life in order, but one thing at a time. Fuck’s sake, how did I waste a day before using the PDA? How did I waste a minute? Typical Nick, gets every chance and forgets to grab it. For all I sit around thinking and stewing, nothing actually happens until I do it. The kind of person who would do it… courageous? Smart? Energetic? Maybe it’s working memory… no, then someone else in the Brainle beta would’ve hacked that and gotten a PhD by now. It’s only been 8 days. What do I need to fix?