davidsong

@futureghost327 @VoidStateKate Depends. What do you do?

@bitplane @VoidStateKate Is there anything that might be practically useful for someone who doesn’t need to write code for anything?

Genuinely why are we calling it "artificial" intelligence. It is natural and the consolidation of every pattern we've ever created or made with our own intelligence compressed and consolidated into a coherent container. What is a better name for it? Enhanced Intelligence?

@futureghost327 @VoidStateKate Not interesting for you. It's for me. Maybe you'll benefit as a by-product, maybe not. But I'm doing the things I want to do 200x faster. And software archeology isn't the only thing I'm doing, I have a day job and other projects.

@bitplane @VoidStateKate Dude… being able to quickly write a code to open old files is not an interesting thing whatsoever. It’s just boring stuff we could already do but faster.

@robertskmiles If I had a whole doctor to myself I'd trust the doctor. Loop over claude over and over again, adding context and research as you go, you'll get a better answer than a doctor will give you in the tiny amount of time they have to spend on you.

If a doctor tells you something about your health, and Claude tells you something different, which is more likely to be right?

Yeah, about 200x slower. Programmers can now make 200x more things, or things that are 200x harder to do, and computers are doing way more cool shit because of this. When I started out coding I had to buy and read entire books to figure out how to do the most simple things, then the web and search engines went mainstream. Within 10 years we had mobile phones with computers in them. Pretty much every site and app you use (including your browser) is built on free software written by volunteers, who use the web and Google to figure out how to do it. AI is bigger than search engines or Wikipedia.

@bitplane @VoidStateKate So you used AI to write code to open old files faster? You could have done this without AI, just slower?

I agree. IMO evidence of infinity requires infinite evidence. Infinity is a religious choice - Cantor was a Calvinist, and he baked his God's infinite glory into the number line. The same sort of thing has happened to science too, it's the study of God's physical realm and nobody questioned it.

Physics again. Apologies to non-physicists. 🙁 Any finite quantity compared to infinity is infinitesimal. This is a logical contradiction and one of several reasons that infinity is bogus. Therefore, continuity is also bogus. So why is continuity still accepted in physics? It's because politicians (professional bullshitters) took over. Since Einstein's spacetime continuum requires infinity, infinity remains. 🙄 Late in his life, Einstein came to understand that continuity was illogical, but it was too late. Most of modern physics is bogus. 😠

I'm not using AI to look at them, I'm making code so anyone can look at them without AI. You know how this site crops and shrinks your profile pic after you upload it? Someone like me wrote the code for that, Twitter just used that code. Computers aren't magic, someone has to make them work. That's my hobby.

@bitplane @VoidStateKate It just sounds to me like a bunch of old files that you’re… opening to look at? And you need the AI because that’s… faster? Why do you need to open so many files so fast? I can only look at one thing at a time, really

I think you're right about the speed of light thing, things moving slower than it are moving at the speed of light internally. I disagree about determinism though. If nature was rule-following and discrete, brains would be mindless control organs - there would be no evolutionary selection pressure to align feelings with actions.

@TrueAIHound Moreover, we will get a completely deterministic but not linear universe ... we are not able to detect all influences, so we call effects where we did not observe the cause "random". Since we are part of the universe, we cannot observe all influences. But they are "there".

Physics: clarification I wrote previously: "If physicists truly understood that nature was discrete, not continuous, they would quickly realize that there's only one speed in the universe: the speed of light. Nothing can move faster or slower." How can that be, you ask? Don't we observe things moving slower than light all the time? It's important to realize that, in a discrete universe, every particle moves by making a series of quantum (discrete) jumps. Each jump occurs over a minute distance at the speed of light. If an object appears to move slower than the speed of light, it's because it is at rest part of the time. An object that is observed to move at half the speed of light is actually at rest half the time and jumping the other half. At the speed of light, it's all jumps with no rest periods. At ordinary speeds, the object is at rest almost all the time with a few jumps sprinkled in. We don't notice the tiny jumps because our visual system is extremely slow and coarse-grained in comparison. A discrete universe model changes everything. I believe there is big physics on the way within your lifetimes. Calling it 'big' doesn't come close to describing it.🤔

@JustinWaugh @yacineMTB Ah okay, so ML research? Sounds interesting. Solving continuous learning sounds like a tough problem to crack!

@bitplane @yacineMTB Async and continuous time learning: specifically building on device scheduled kernels that tick at dynamic rates and operate on channels of data (rather than data-flow sorta paradigm most ML is set up around)

Never use a codex prompt without /goal by the way

They contain: * Pages of comic books and books (cbr files) * Computer games for old consoles posted on the Internet * Software that runs on old machines * Stuff that was attached to emails or newsgroup posts in the 2000s * Tons of downloads on archive.org from the early web * Files that old people (like me) have in their old backups * Files that were on magazine coverdisks you might find on ebay I wanted to be able to open and write to them for another project I'm working on. I'm an archivist and computing historian, it's a hobby. I do it for the same reason anyone creative does anything, to scratch an itch and because I can. I also made this: bitplane.net/dev/python/pil… Which gets the loading screen from the old computer games I played as a kid. I'll make it so they show up as icons in my file browser eventually. And this: bitplane.net/log/2026/03/am… Which lets anyone view the icons like the ones I made for my own games when I was in school in the 90s. And this: bitplane.net/dev/rust/amiga… Which lets me open .lzx files containing software from the 1990s. And this: bitplane.net/dev/python/cff… which lets me read documentation from CDs in the 1990s. And this: bitplane.net/sh/rip Which rips those CDs and uploads them to the Internet archive for future historians. And this: bitplane.net/log/2026/04/ne… Which lets people open tons of old picture formats in Python. And this: github.com/bitplane/qemou… Which opens my old Amiga hard drive so I can get my old games off of it. And tons more stuff. They're not all AI written, but they're accelerated by AI.

@bitplane @VoidStateKate But like… I don’t NEED a Lego machine to sort my sock drawer. If I’m real with you, I don’t sort my socks AT ALL. It’s not necessary. You see what I mean? Like what the fuck is in these files and what do you do with them when you open them?

@JustinWaugh @yacineMTB Nice. What are you researching, if you don't mind me asking? This was mine: bitplane.net/log/2026/05/ra…

@bitplane @yacineMTB They're autonomous research loops: generating variants, running sweeps, collecting candidates, tuning, and re-running. Just checked, roughly ~56% of the time is LLM decode, 44% code execution. The top 2 are 5.5B and 7.5B input tokens and 11M and 14M output tokens each so far.

They're files that have other files inside them. Some of them are very old and don't open in programs like 7zip, some are new and don't open either. You need WinRAR. If you want to change what's in any of them, you can't do it on a phone, a raspberry pi, your set-top TV box or an e-reader. But now you can, but last week you couldn't The point is that this is something that is high effort for very little reward. The fact that it was so difficult to make is what makes it impossible - nobody's gonna spend years making it work just because. 6 months ago it would have been an impossibly difficult problem for the same reason it's impossible to build a lego machine to sort your sock drawer - it's doable, but why would you? In May 2026 you can do things just to see if they can be done.

@bitplane @VoidStateKate What is it doing exactly? Just opening and saving files? To what end? It’s…. Storage?

@valigo As soon as it was good I started using it to do more complex things, they broke in new and interesting ways. Like the composite mic that's a soundboard for proximity chat in games, and messes up when I switch between bluetooth, wired, usb and internal laptop mic.

I just thought of a very good "no sound on Linux" joke, but then I realized that I didn't have any sound-related issues on Linux in many years at this point. Or rather ever since PipeWire became mainstream. When was the last time you had sound issues on Linux? The joke was "on Linux, nobody can hear you scream" btw, but typing it out now it's kinda cringe ngl.

@JustinWaugh @yacineMTB What have the 190 hour one got done, is it waiting on results most of the time? I ran for 16 hours but I'd spent weeks getting a plan together and the output was 40,000 lines of code that needed about 10 review/refactor passes until it was right.

@yacineMTB I've had 4 running continuously for ~1 week now.

@futureghost327 @VoidStateKate Think of it a different way. If you had a technical expert who could do any work using a computer for you, what would you ask them to do?

It opens and saves RAR files - every version of a file format that evolved and grew for 30 years. It's also a technical document about them. They're like ZIP files. You might download from the Internet archive, find them on old CDs/DVDs, in comic book collections, or old games and software uploaded to the web. It lets you get to the files inside, or pack files into a new RAR file. It's written in Rust so future hackers are unlikely to be able to share a file that takes over your computer when you open it. It's free software so historians can use it to process collections on any kind of computer, without anyone's permission. Doing this would have taken years without AI assistance, so until now nobody did. The blog post documents how I did it, so other people can make their own things if they want to. Obviously the things I want to make are nerd shit because that's me. And I've been doing a lot of it recently, about 200x more than usual thanks to AI doing research and development.

@futureghost327 @VoidStateKate An experiment where the output was about 5 years of full time work, that was done in a month of evenings and weekends. Even if you don't care about the sort of work it is, or the social good of it being a donation to historians and engineers, the amount of work done is impressive

@bitplane @VoidStateKate What the fuck am I looking at?

@steveruizok This is really slick. Nice work.

stroke and erase

@futureghost327 @VoidStateKate I did this over the last month bitplane.net/log/2026/05/ra…

@bitplane @VoidStateKate What is it useful for? I haven’t found a use in my life

@CoastalFuturist That's pretty funny

Makes me SICK when women walk up to my wife and I when our 1 year old is smiling at them and says “he’s such a little flirt” I used to let it go but I’ve started saying “no he’s not he’s a year old that’s weird”

@yacineMTB I'll need to weigh them. 🖐️ 🖐️ WEHEEYYYY!

I would like to have a consultation. About lactation

Lactation consultant