Chain of thought monitorability: A new and fragile opportunity for AI safety

(arxiv.org)

132 points | by mfiguiere 4 days ago ago

65 comments