proofs getting cheap is great, but the spec is still the weak spot. if a model drafts both the spec and the proof, a reviewer can sign off on something that proves the wrong thing. who writes the specs in that Anthropic setup?
You are viewing a single comment's thread from: