The Inside View cover image

Collin Burns On Discovering Latent Knowledge In Language Models Without Supervision

The Inside View

CHAPTER

How to Train a Probation Better on the Misaligned AI

The claim or the hope i mean these are conjectures right so this is about like you know supposed we didn't have these systems these are properties i would predict and i would predict that if we literally had this gptn fine-tuned untruthful stuff versus gptnfine-tuned on the misaligned system we'd be able to identify which is the truthful gptn. So basically you're like doing some like meta um some like meta level thing about like having a yes or no but instead of having two statements we have like two models and whenever you have two things that contradict each other you're just going to like look at the edan states without looking at

00:00
Transcript
Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
App store bannerPlay store banner