Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

When AI Builds Itself: Our progress toward recursive self-improvement

anthropic.com

Post reply on HN
Print view View original

731–738 of 738 posts

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#731

Post by overgard »

verdverm wrote: post #702
overgard wrote: post #690

Earlier quoted context omitted.

In my case, where I see it most often is when the LLM has to rework something multiple times, and the feedback loop is vague (especially when all I have to give it is "no error messages, but it's still broken"). It seems like after the third or fourth try it just kinda goes off the rails. I find that the one-shot quality tends to be a little better, if the slot machine happened to work correctly that time.

You shouldn't be using an LLM directly (web chat style). A proper harness allows an agent to see the errors itself and correct as needed. You can the correct it at higher, more meaningful levels.

My experience is in doing this with Claude/Codex/OpenCode with a pretty rigorous setup (AGENTS.md/CLAUDE.md for specific subfolder rules, strict compile/test/lint rules. This isn't me copy-pasting from web chat.

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#732

Post by lobocinza »

wordpad wrote: post #717
cyanydeez wrote: post #648

Earlier quoted context omitted.

theres also no evidence that any of the manual labor will benefit. we are nowhere near the type of scifi utopia that gives us replicator food. aside from capitalism moving money up and living condotions down, AI is going to accelerate the gap between rich and everyone else.

Starvation is already not a thing anymore.

https://news.un.org/en/story/2026/04/1167402#:~:text=Hunger%...

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#733

Post by lobocinza »

josephg wrote: post #565
tasuki wrote: post #531

Earlier quoted context omitted.

> if AI coded software works correctly, maybe it doesn't matter? The problem isn't the amount of code, it's how fitting/unfitting the abstractions are. Wrong abstractions are bugs in waiting. If there's much code with wrong abstractions, future change becomes difficult. Source: me, I've created many bad abstractions and they led to much pain...

Yeah. Its kind of strange - claude is great at some tasks, but it seems really rubbish at coming up with good abstractions a lot of the time. I've often caught it making a conceptual mistake (like "X cannot do Y") - then spending hundreds of lines working around an issue that doesn't actually exist. Its also really bad at inventing and leaning on invariants. I make rules in my code all the time - "by the time we get…

They have no illative sense like we do.

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#734

Post by aleqs »

overgard wrote: post #731
verdverm wrote: post #702

Earlier quoted context omitted.

You shouldn't be using an LLM directly (web chat style). A proper harness allows an agent to see the errors itself and correct as needed. You can the correct it at higher, more meaningful levels.

My experience is in doing this with Claude/Codex/OpenCode with a pretty rigorous setup (AGENTS.md/CLAUDE.md for specific subfolder rules, strict compile/test/lint rules. This isn't me copy-pasting from web chat.

I've been building a general linter tool to help keep repos in a consistent and clean shape when working with AI [0]. You can define repo and file layout, structure, hygiene rules and have them checked in pre-commit, in ci, or manually. It also integrates and plays well with AGENTS.md - allowing exporting agent instructions from the alint rule config [1].

[0] https://github.com/asamarts/alint

[1] https://alint.org/agent-friendly-linter/

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#735

Post by __patchbit__ »

NackerHughes wrote: post #719
wordpad wrote: post #717

Earlier quoted context omitted.

Starvation is already not a thing anymore.

Then why are people starving?

Israel is starving 2 million plus people by collective punishment area access denial and the blackmarket for food trades USD$250 for a dozen eggs.

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#736

Post by throw394689 »

__patchbit__ wrote: post #735
NackerHughes wrote: post #719

Earlier quoted context omitted.

Then why are people starving?

Israel is starving 2 million plus people by collective punishment area access denial and the blackmarket for food trades USD$250 for a dozen eggs.

This is not true. As of Jan 2026 the price of eggs in Gaza is 25 Shekels for 30 eggs:

https://www.un.org/unispal/wp-content/uploads/2026/02/WFP-Pa...

You’re just repeating old outdated bits of propaganda.

Also, nice that you’re estimating that there’s been no decline in population size over the past 3 years. At least you slipped a bit of truth out in an attempt to demonize them.

Re: When AI Builds Itself: Our progress toward recursive self-improvement

#737

Post by Folcon »

bitwize wrote: post #115

After several months with their top engineers and state-of-the-art AI on the job, Anthropic managed to "reduce flickering by 85%" on their TUI Claude Code client, which is built in fucking React and rendered by drawing the entire chat conversation each time (hence the flicker). I think they've since eliminated it completely by slapping some double-buffering around it (since "our client is actually a real-time game en…

You know, of the many criticisms I see people make, this is my canary, if claude code actually starts getting better, that would definitely remove one of the biggest question marks in my mind, that piece of software has so much strategic value for them, their entire premise is "our software gets better by itself", how is it that I still see weird UI bugs and glitches?

It's a game engine? Fine, get some good gamedevs on the team then, this is a non problem in gamedev land, heck Casey Muratori did a whole bit about performance improvements to editors, so they should be good there

Not to disagree with your point, I very much think the fact that Emacs and vim do this so well is not doing them any favours, but I'm trying to meet them where they are

Post reply on HN