OpenAI's New Model Cracks Ten Math Problems, Raising Authorship Questions

OpenAI says its next-gen model Astra's internal version has solved ten long-standing math and computer science problems. These include high-dimensional sphere packing, binary codes, non-sofic groups, and Connes's rigidity conjecture, many unsolved for over a decade.

The computing cost is about $2,000 based on Sol API rates. OpenAI stresses that authorship should reflect how results were made: if an AI-generated proof is claimed as human work, it ignores the system's role and misrepresents human effort. Human researchers helped write the paper and verified proofs in Lean, but the math arguments came from the AI system.

This raises questions about AI's role in research and academic integrity. OpenAI believes transparent labeling is key, and this might push updates in authorship rules. The papers are now public for review.