Max Schwarzer
TalkRL: The Reinforcement Learning Podcast
The Balance Between Prior Knowledge and the Rest of the Algorithm
The generalization ability of our function approximators is that holding us back. I think we can get to about human normalized score two, so twice as good as BBF on average, just by fixing up residual problems with function approximation. If you really want to go beyond that though, yeah, you're going to need prior knowledge. So probably that's going to mean figure out how to use LLMs for this realistically given how the field is looking right now.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
