logoalt Hacker News

codewithcheese • today at 3:24 PM • 1 reply • view on HN

RL is not scalable to super intelligence. Super intelligence should be able to learn continuosly without verifiable rewards.


Replies

StevenWaterman • today at 3:32 PM

You don't need RL all the way to ASI. You just need RL far enough that the model becomes good enough at AI research that it discovers something better than RL.

➕ show 1 reply