How to Make Language Models Truthful

I think this paper is quite simple but both like very insightful has been discussed a lot on Twitter uh and less wrong in the past few weeks. We currently basically don't know how to get these models to tell the truth um so we have some techniques for like okay maybe we can train models to imitate truthful text or something like this. This is sort of the default approach that we have for making models truthfulUm so the thing is like models can also be more capable than humans and I think this will become certainly true in the future in which case I think that standard techniques for making modelstruthful just will break down. Even with current techniques for make models truthful it doesn't always work

Transcript

Play full episode

Transcript

Episode notes

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app