Aakash GuptaLead source
OpenAI's new math model spends about 3 hours of compute per result. A human expert needs days to weeks to verify one. OpenAI just published 722 manuscripts at once, and the boycott everyone's laughing at is actually a labor dispute. Run the numbers. Even at one week of expert review per manuscript, 722 papers works out to roughly 14 years of full-time mathematician labor. OpenAI already retracted three of them the day after release over a sign error, so the checking clearly matters. The company itself says unverified results may contain mistakes. So who does the checking? Not OpenAI. The field does it, unpaid. That's the mechanism behind the fury. Generation costs collapsed to three hours a result while verification is still priced in human career time, and the release was structured so…Read original on X