OpenAI's AI-generated mathematics release faces validation backlash

OpenAI published 722 manuscripts from an unreleased model covering 372 result families and about 4,000 attempted problems. Follow-up criticism highlights incomplete Lean verification, mismatches between prose and formal code, limited prompt disclosure, authorship concerns, and withdrawals or revisions; nevertheless, at least one result reportedly catalyzed a stronger bound, suggesting potential value as a research aid.