Neel Nanda on Avoiding an AI Catastrophe with Mechanistic Interpretability

[@mycrtfellonme]

Always happy to see more interviews! Love hearing the interesting perspectives of your guests! ☺️

[@antigonemerlin]

That explanation of groking was mindblowing, and also kind of scary. So these models have some capabilities that perhaps it doesn't think are good enough, but _it's still there_ ... Makes you wonder what's just beneath the surface.

[@carsonellis2992]

Need Nanda 🔥🔥

[@samyz3829]

ye LP

[@JazevoAudiosurf]

more AI content

[@DoktorUde]

sorry but who can take this guy seriously?

[@radicalrodriguez5912]

what's not to take seriously?

[@triton62674]

The guy was literally top at Cambridge every year in maths (out of 250!), he's one of the few people that should be taken seriously if anything.