OpenAI's math claims fall apart under human scrutiny

OpenAI posted a repo of AI-generated mathematics (49984923), and the mathematicians pushed back hard. The AHM issued a statement on the October 6 release (50003677). Then OpenAI withdrew three math papers (50003107). One top comment: 'So much for the "it's lean verified" defense.' Another: 'Proof by authority works until human mathematicians actually run the code.'

The surrounding threads show a field working out what it wants. 'The Mathocalypse' (49997718) calls the agents statistically guided brute-forcing that produces no understanding. 'Math 2.0' (50002008) argues the community needs to value progress more holistically, and the top replies come down to one word: 'Taste.' The Cleo thread (49982445) draws a parallel between a Stack Exchange account that may have been sock puppets and how AI labs present unsolved-theorem proofs.

The key bit: one commenter notes that without a thriving mathematical community to point out errors, the broken proofs would have stayed broken. Automated math erodes the very reviewers who catch its mistakes.