
Episode 24: Jack Parker-Holder, DeepMind, on open-endedness, evolving agents and environments, online adaptation, and offline learning
Generally Intelligent
00:00
Bayes Opt Online Oral Training?
In PBTU, what do you think are the Bayes blockers now? Like the kind of like local questions or things in the way? Yeah, I think that we kind of touched on one, but I think the greediness definitely is one of them. It definitely is only optimizing for one step ahead, and the methods that we've been using. And it also is really tricky to actually solve problems that we'd introduced, which is the online hyper- because you do have this challenge of like, if you have a few examples of something not working early on, you might not explore again for a very odd period.
Transcript
Play full episode