Episode 24: Jack Parker-Holder, DeepMind, on open-endedness, evolving agents and environments, online adaptation, and offline learning
Generally Intelligent
00:00
Are We Getting More General Agents?
I think if you can learn world models directly on the internet that produce any environment for an RL agent that it can't currently solve, that's interesting in some way. And then the agent learns to solve those environments, but they're somehow grounded to things we care about. It can't take like the race track one, be like, what kind of experiments until it's all done on top? I mean, that's... Oh yeah, you could. Yeah, how would you? But that's a very high level goal.
Transcript
Play full episode
