John Schulman
TalkRL: The Reinforcement Learning Podcast
00:00
The Relationship Between AGI and RL
I expect us to see multiple models that are proclaimed as AGI and then only after, only after interacting with them in a while, you do realize it's not quite there. I'd say that RL is a useful, um, like component of training AGI or an almost essential component. It lets you optimize any objective for the agents' behavior. So if we really, uh, want to optimize the agents behavior for a specific objective, RL is kind of the only framework that lets you do that.
Transcript
Play full episode
