Reposted by 🍐
but theorems output as the collateral exhaust fumes of model evals give you facts, not insights. At least not directly. Only by studying those facts intensively can you relate those facts back to genuine insight. And that's the heart of the fissure between labs and the mathematical community.
