OpenAI published on October 6-7, 2026 a catalog of 722 manuscripts organized into 372 result families, covering number theory, algebraic geometry, analysis and theoretical computer science, after testing its internal model on approximately 4,000 problems. About 300 of these results, or 42%, have Lean formalizations, a proof assistant that verifies each logical step. Three manuscripts were retracted shortly after publication due to errors, illustrating that AI-generated results are not synonymous with correct results. Some manuscripts address Millennium Prize Problems, and mathematicians such as Terence Tao and Alex Kontorovich have examined results connected to the Riemann hypothesis. Critics have raised concerns about verifiability, as OpenAI has not shared the underlying model or the complete prompts used.
Source: Read the original article
