cousin_it
cousin_it on Hacker News
I'll go out on a limb and say that future AI models, running on today's smartphones, will be better at programming and formal math than today's frontier models running in datacenters. And by "future" I mean less than 5 years from now, almost certainly.
The Jonckheere Rolls-Royce is my #1.
If humans lost economic leverage (no jobs) but kept military leverage (usefulness as soldiers), that'd probably be utopia. We could all be a leisure class like the aristocrats of the past, and could defend that state of affairs.
Unfortunately we aren't in that world. AI+robots will deprive humans of both economic and military leverage. Most likely, powerful "blobs" of a few owner humans + a lot of automation will inherit the Earth and kick most humans to the curb.
>
And the study and the creation of those formalisms is not something automatisable.Lol again. Have you tried it? Go ask your favorite LLM to invent a machine-checkable formalism for something that you vaguely describe to it. It'll oblige. If it's one of the top models and you give it a good budget, it'll do a good job. Very likely it'll do a better job than you.
Consciousness isn't required for problem solving. People said for decades that chess requires Genuine Human Creativity. When chess fell, people started saying that Go requires Genuine Human Creativity. When Go fell, people started saying the same about math; starting from simple problems and going up to Millennium problems. Now that a Millennium problem fell, people are still saying this. What can I say? Some people can't extrapolate from past evidence to likely future.
In my comment that started the thread, I pointed out that "coherence" in the article is considered valuable because it's a proxy for "ease of modification", but AIs can estimate "ease of modification" directly (by looking at how hard it is for them to make changes to a codebase) and learn from that in a closed loop. No round-trip through humans required, contrary to what the article says.
This is obviously pasted from an LLM, so yeah, bye.
(Sighs)
Goedel's theorems don't stop AI from finding and formalizing math results. They certainly don't stop AI from making conjectures or writing definitions. They have nothing to do with any of that.
The theorems are about inability of formal theories to prove certain sentences involving provability (the first is about "this sentence is not provable", the second is about "this theory is consistent"). Most mathematicians never touch any such sentences in their entire career. All the life's work of someone like Gauss, an all-time great, can be formalized just fine.
The article makes claims about the future, such as:
> We cannot make an AI that improves software coherence beyond the human ceiling without leaving the bubble
That's what I was replying to.
I think you maybe didn't read my comment carefully. The article says AI will have to talk to outside the datacenter to do this, I'm saying it won't.
>
Have you seen it reliably do this?Lol, no. I don't need to. I can predict such things.
For example, I predicted about 15 years ago that AI would decisively beat humans at math, because math is formalizable. Most people didn't believe it until this summer, and some don't believe it still.
To people who don't understand that AI improves over time, my prediction skill seems magical.
This is very high effort, but the central argument is wrong.
The central argument is that AIs, living inside datacenters, won't be able to evaluate the "coherence" of software "beyond the human ceiling". Okay; why do we need this "coherence"? The article argues that it's needed for ease of adaptation. But ease of adaptation (as measured in tokens required to add new features, or whatever) can be directly observed by AI as it works. AI can make conclusions about it, and carry them over from one program to another, without running into any human ceiling. So it can stay within its datacenter and make software that's faster than human-written, smaller, and more adaptable too. The whole essay dissolves. Sorry!
Yeah. I want to note quickly that the currentScript method can do a bit more than that. For example, the page head can include script.js with the code for rendering the site shared header and footer. Then in the page body you can have inline scripts using the currentScript trick to add the header and footer in place, kinda like including shared templates. And it runs during page load, before even onLoad, so the page renders with header and footer straight away.
When I first learned that XSLT was going away, I got curious how to achieve the same thing in JS (client-side templating that happens during page load, without any "flash of content"). It turns out there's a very nice way to do it, by using inline scripts that append elements next to document.currentScript. Lean Rada wrote about it here:
https://leanrada.com/notes/inline-rendering-currentscript/ and I wrote about it as well:
https://vladimirslepnev.me/write
The more I watch movies, the more I want longer shots, wider shots, and more people in the same shot. And if there's a cool thing happening, then show the action and reaction in the same shot - don't cut at the exact moment a punch connects or a car crashes. And don't play nonstop music in every scene, let the actors act.
Can we make a simpler thing: staff and durations same as before, but put solfege syllables (relative to current key signature) inside the note heads?
Edit: it seems a bunch of people already had this idea, it's even in Musescore. It might need some more work to make it readable at small sizes, though.
This looks cool! I wonder if the same feeling can be achieved without a fixed grid though. Some kind of pointilist or halftone rendering, but with free positioning of points.
>
Madrid has 3.4Mln people and is the second-largest city in the EU. How big do you want it to become?Bigger, obviously. Because the number of people who want to live in Madrid or visit it is higher than that. This can create a lot of economic surplus. Part of that surplus can be used to pay off locals (e.g. with housing in the newly built stock) so they win too.
The real problem, as far as I can tell, is that there's no socially accepted system for paying off locals. So locals end up resisting growth, because there's nothing in it for them except higher rents, and everyone loses.
Yeah. I like how Kat Woods phrased it: "You can give the slowest child sprinting lessons, but there are still nine chairs."
I think looking cool matters a lot in live music, maybe even more than the actual music. Blowing into a harmonica while carrying a backpack full of cutlery looks really dorky. But two people with drums+whistling and guitar+vocals can be made to look cool.
The first time I saw it, I thought about playing a drumkit and whistling a bassline at the same time, with someone else on guitar and vocals, making a full rock sound with two people. But then I was too lazy to try it.
I think it's more than "you need to know them". You need to be a friend to them and do things for them in turn. Play a part in the mutual help network. If many people free ride on it, it dies.
I don't really like the concept of "guests" being applied to people living somewhere for years, working, paying taxes, making families and friends, sending children to school and so on. I think it would be more humane if this concept was minimized, and instead the concept of "no taxation without representation" was applied. Fast track people as much as possible. Maybe one year as a voiceless "guest" is okay for a human being, but not more than that.
Yeah. I've seen the same puzzling thing in game development. People don't seem to want to make good games! They want a hat that says "I made a game". That done, they want a hat that says "my game is successful", so they go on subreddits asking for promotion advice. Honest comparison with existing state of the art is seen as annoying; advancing the state of the art isn't seen as a goal at all.
When I was learning DSP, I was surprised by the fact that generating pink (1/f) noise, sample by sample, is not mathematically easy at all.
One practical approach is passing white noise (every sample independent random) through a "pinking filter", which is a sum of a bunch of lowpass filters at different frequencies, so that the sum of their cutoff "knees" approximates the frequency curve of pink noise. Another approach is generating in chunks using Fourier transform from a desired shape. Given the ubiquity of 1/f noise, you'd think there would be a simpler and more direct algorithm, but no.
The article makes an interesting point. Like, sure, AI slop is nasty; but also maybe the job itself was bullshit all along (making powerpoints, or coding yet another todo app and so on), and you were only making peace with it by adding a human element which wasn't strictly necessary. Now with AI that illusion is stripped away, and you see the job for what it really is: just how little it means.
A bit off topic, but I'm guessing Aristotle himself would say that chasing "status in the workforce" (as opposed to leisure, riches, wisdom, power etc) is itself a good example of slavish disposition. Status in the bosses' eyes, amirite?
No mention of GANs? As far as I remember, their images felt much more like "sampled from training data" than "averaged from training data". The idea was that you train a "discriminator" that tells generated images apart from real ones, and make the "generator" try to fool it. But somehow it lost to diffusion models, and now all AI imagery looks like slop, when GAN outputs (though imperfect) didn't look nearly as slop. I don't really know what happened though.
Is HN showing post and comment times as recent when they really aren't? I see an exchange between Rochus and jefftk "1 hour ago", ending with a "thanks, posted" and that blogpost is from July 25. Today is July 29.