TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Can LLMs invent better ways to train LLMs?

61 pointsby hardmaru11 months ago

10 comments

brokenmachine11 months ago
Can monkeys with typewriters invent better ways to train monkeys with typewriters?<p>Yes, but you may need a lot of monkeys.
luke-stanley11 months ago
The project sounds quite interesting but I&#x27;m not sure running it is going to work! The code `gpt_model = &quot;gpt4_20231230_1106preview&quot;` is not using a valid model name as best as I can tell, so it seems unlikely to work - from <a href="https:&#x2F;&#x2F;github.com&#x2F;SakanaAI&#x2F;DiscoPOP&#x2F;blob&#x2F;main&#x2F;scripts&#x2F;launch_evo.py#L15">https:&#x2F;&#x2F;github.com&#x2F;SakanaAI&#x2F;DiscoPOP&#x2F;blob&#x2F;main&#x2F;scripts&#x2F;launc...</a> Unusually, the issue section doesn&#x27;t exist so I can&#x27;t provide feedback to them that way. But luchris429&#x27;s repo does have it so will do so there. Maybe it&#x27;s dead code. Still, it&#x27;s wrong.
评论 #40671032 未加载
jasfi11 months ago
They are very useful when ideating with a human. On their own they could veer off into uncertain territory, and likely make mistakes obvious to humans.
teo_zero11 months ago
I&#x27;m sure LLMs can optimize the training of other LLMs (either by inventing new ways or fine tuning existing ones). But we can&#x27;t predict whether this will result in a giant&#x27;s leap in the field, or just small increments. That&#x27;s the definition of singularity, isn&#x27;t it?
seydor11 months ago
can LLMs optimize anything?
评论 #40672033 未加载
评论 #40666440 未加载
评论 #40667767 未加载
rulalala11 months ago
Would that not be a form of self consciousness?
bugbuddy11 months ago
Betteridge&#x27;s law of headlines: no
评论 #40666184 未加载
评论 #40665965 未加载
ChuckMcM11 months ago
A better question is &quot;Can LLMs invent <i>anything</i>?&quot;<p>Don&#x27;t misunderstand, building systems models using existing system response as a way of analyzing those systems is a useful methodology and it makes some things otherwise tedious things not so tedious. Much like &quot;high level&quot; languages removed the tedium of writing in assembly code. But for the same reason that a compiler won&#x27;t emit a new, more powerful, CPU instruction in its code generator, LLMs don&#x27;t generate previously unseen system responses.
评论 #40666524 未加载
评论 #40666215 未加载
评论 #40666497 未加载
评论 #40666775 未加载
评论 #40666226 未加载
pineaux11 months ago
So it does seem to work. That&#x27;s not clickbait then?
评论 #40666777 未加载
hlkcrcck11 months ago
Of course, they can invent anything. A better question is how efficient? Because even with brute force you can invent anything: <a href="https:&#x2F;&#x2F;libraryofbabel.info&#x2F;" rel="nofollow">https:&#x2F;&#x2F;libraryofbabel.info&#x2F;</a>