6 - Debate and Imitative Generalization with Beth Barnes
AXRP - the AI X-risk Research Podcast
00:00
Is It Necessary?
Lime technique is based on the idea that if we know everything a model knows, then it's easy for us to catch deception or treachery. But debate doesn't work well because knowledge looks less like this kind of relatively explicit boast up reasoning and not more things that are just reasonable in training data. Lik go sating. So so i interrupted a, what? Well is, yes. I ask, what problems does it solve and does it not solve? Yes. And als o, sort of, more generally, i his kind of strategys stealing argument, which is, like, anything the mol knows that it could be leveraging to advance some goal,. We can also leverage
Transcript
Play full episode
