6 - Debate and Imitative Generalization with Beth Barnes
AXRP - the AI X-risk Research Podcast
00:00
Is Debate Really a Good Idea to Train Ema Systems?
i and paul don't think that we will actually use debate to train ema systems. It's just like easier to understand and think about and like, intelible, practical to do human experiments with than idea. But like, has basically same set of limitations. I'm not thak competent. There could easily be something else. i feel like maybe there on additional problems that are sort of as big and of a similar type to this,. Like knowledge that's only justified fo our intuition. And it feels plausible that the pieces of debat plus amtinte of generalization capture allf the kind of ways of knowing things.
Transcript
Play full episode
