HN Explorer

Google Has Achieved RSI

1 points by AlexWApp ↗ HN
My Twitter feed is flooded with claims that Google has achieved recursive self-improvement (RSI).

I am curious how people here interpret this. Does what Google has demonstrated actually qualify as RSI in the stronger sense of an AI system improving the process by which future versions of itself are developed, or is “RSI” being used too loosely?

4 comments

[ 0.24 ms ] story [ 19.2 ms ] thread
jryan49
What did they demonstrate can you share some links? I'm not on twitter.
runjake
There's various low quality rumors on X, but here's one quoting Dario that RSI is start to happen across the industry, including at Anthropic:

https://x.com/kimmonismus/status/2098786231534362767

Edit: And here it is directly from Dario:

https://darioamodei.com/post/we-must-pace-the-frontier

"My first concern is that, since roughly this summer, AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI. This dynamic is called recursive self-improvement, and it is starting to happen across the industry, including at Anthropic, as we and others have described. Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all."

Edit 2: An article citing Altman and Musk backing Dario on this post: https://www.ft.com/content/31220b59-b0c6-401c-a146-2b7b5d138...

Elon on X: https://x.com/elonmusk/status/2098789109980332057

Altman on X: https://x.com/sama/status/2098811563415150910

jryan49
Three options, they think we're reaching RSI or they are proposing we "slow down" because the models aren't getting better at the speed they expected and want an excuse for it. Or they want to make it illegal for other companies to build AI so they can corner the market.
geremiiah
Presumably they only have empirical results and if so, wouldn't you need to train multiple successive models to even test such a hypothesis? Even then, it's not like an inductive proof. An improvement in one generation does not imply an improvement in the next (unless you define improvement to mean exactly that).