2 - Learning Human Biases with Rohin Shah
AXRP - the AI X-risk Research Podcast
00:00
The Reward Functions With Ta
The way that we use this is that we basically both the planet modula and the reward function n in or architecture, is trained by andtinb i radiant descent. And so there you're just infering the reward functions with ta, given your already learned model of what the biaces are. Once we have the plant, once after we've done this training, the planning module is then frozen, and it now has already incoded all the bioces. Then we use gradient descent to learn the reward functions on new tasks for which we don't already have the rewards.
Transcript
Play full episode
