Comment by fph

So a fair headline would be "OpenAI tried formalizing more of their proofs, and found 33% of them had a mistake".