Episode 28: Sergey Levine, UC Berkeley, on the bottlenecks to generalization in reinforcement learning, why simulation is doomed to succeed, and how to pick good research problems
Generally Intelligent
00:00
Large Scale Regression Learning (RL) in Atari Games
In RL, you need to do this kind of factual reasoning pretty implicitly. And so you need to represent these sub-optimal behaviors. But in a language model, you don't need to,. they're often quite bad at a counterfactual reasoning. So there's something interesting here.
Transcript
Play full episode
