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

Sakana AI's Recursive Self-Improvement (RSI) Lab

sakana.ai

Post reply on HN
Print view View original

21–28 of 28 posts

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#21

Post by cheevly »

laughingcurve wrote: post #19
artninja1988 wrote: post #15

Earlier quoted context omitted.

I think sakanas papers are one of the more creative, not just gunning for incremental benchmark improvements. But yeah I agree that they can be a bit (or very) hypey. But regardless, I want to see more of their kind of research than endless benchmark chasing. All the best to David Ha and the team!

Can you explain how 'recursive self-improvement' functions without 'endless benchmark chasing'? I mean, RSI is literally that. What do you think they're improving on? How would a model self-improve without some metric/data of some kind to check? When you have metrics+data, that is a benchmark. And yes, simulations and or soft-verification like LLM judges are still a kind of benchmarking. Maybe its not a static benchm…

[flagged]

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#22

Post by algo_trader »

tmaly wrote: post #11

I heard about the ShinkaEvolve on a podcast where the guest had used it to evolve an agent harness for a less capable model. I ended up borrowing the ideas from it for one of my own personal projects.

Do you actually understand why evolution methods are beneficial?

SGD generates a stronger learning signal,is more efficient, and scales better. Using it end-to-end makes it stronger yet.

Yet somehow mixing in a weaker blunt evolution stage improves the result?

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#23

Post by dominotw »

laughingcurve wrote: post #3

I think the most impressive thing about Sakana.ai is their relentless pursuit of whatever is hype right now. Genuinely it take a lot of work and talent to be this hype-motivated and completely ignore anything except what is popular on X at any given time. Note: RSI is an incredibly important topic -- I just don't care to listen to Sakana on this matter -- they are the epitome of "hypebeast" https://www.urbandictionar…

recursive self-hype

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#24

Post by laughingcurve »

cheevly wrote: post #21
laughingcurve wrote: post #19

Earlier quoted context omitted.

Can you explain how 'recursive self-improvement' functions without 'endless benchmark chasing'? I mean, RSI is literally that. What do you think they're improving on? How would a model self-improve without some metric/data of some kind to check? When you have metrics+data, that is a benchmark. And yes, simulations and or soft-verification like LLM judges are still a kind of benchmarking. Maybe its not a static benchm…

[flagged]

This is a failure to engage the clear arguments and claims made. If you don't want to debate something, why bother commenting at me? Was I supposed to just cede ground and accept your framing wholesale? I'm putting forth very clear and open-to-rebuttal assertions which is what you should do.

Edit: Nevermind, realized he's just an uneducated troll just looking for his kicks. Comment flagged.

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#25

Post by laughingcurve »

cheevly wrote: post #20
laughingcurve wrote: post #18

Earlier quoted context omitted.

counter point: https://sakana.ai/asal/ "For the past 300,000 years, Earth has had only one form of advanced intelligence on it: humans. With the recent advent of AI foundation models, some believe we are at the dawn of a new kind of intelligence. As AI continues to evolve, we may witness the proliferation of diverse intelligent lifeforms coexisting with us." I am not even going to link more than one thing I think I'v…

Yeaahh, that you assert that makes your point says a lot.

In multiple threads now you've claimed I was wrong and then ended the conversation without any assertion, claim, or argument I could rebut. You've ended those conversations with snarky comments clearly designed to shut down debate. This is not how hackernews benefits.

Edit: Nevermind, realized he's just an uneducated troll just looking for his kicks. Comment flagged.

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#26

Post by bigbadfeline »

cheevly wrote: post #20
laughingcurve wrote: post #18

Earlier quoted context omitted.

counter point: https://sakana.ai/asal/ "For the past 300,000 years, Earth has had only one form of advanced intelligence on it: humans. With the recent advent of AI foundation models, some believe we are at the dawn of a new kind of intelligence. As AI continues to evolve, we may witness the proliferation of diverse intelligent lifeforms coexisting with us." I am not even going to link more than one thing I think I'v…

Yeaahh, that you assert that makes your point says a lot.

If you assert that it doesn't... it says even more.

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#27

Post by pjjpo »

laughingcurve wrote: post #3

I think the most impressive thing about Sakana.ai is their relentless pursuit of whatever is hype right now. Genuinely it take a lot of work and talent to be this hype-motivated and completely ignore anything except what is popular on X at any given time. Note: RSI is an incredibly important topic -- I just don't care to listen to Sakana on this matter -- they are the epitome of "hypebeast" https://www.urbandictionar…

I've never considered them a hypebeast, just a scam to get investor money as easily as possible. They seem to be quite successful really.

Re: Sakana AI's Recursive Self-Improvement (RSI) Lab

#28

Post by laughingcurve »

pjjpo wrote: post #27
laughingcurve wrote: post #3

I think the most impressive thing about Sakana.ai is their relentless pursuit of whatever is hype right now. Genuinely it take a lot of work and talent to be this hype-motivated and completely ignore anything except what is popular on X at any given time. Note: RSI is an incredibly important topic -- I just don't care to listen to Sakana on this matter -- they are the epitome of "hypebeast" https://www.urbandictionar…

I've never considered them a hypebeast, just a scam to get investor money as easily as possible. They seem to be quite successful really.

Dang, I thought I was being harsh I wouldn't say scam but maybe all AI is a scam to some.
Post reply on HN