OpenAI researchers say AI models solving research mathematics, with ChatGPT helping solve a 42-year-old open problem in optimization theory

Ernest Ryu, a former UCLA math professor, solved a 42-year-old open problem about Nesterov's method in optimization theory using ChatGPT in twelve hours spread across three evenings, after spending more than 40 hours on it without AI.

OpenAI's internal models have produced more than ten genuinely new solutions to Erdős problems (a collection of open questions left by Hungarian mathematician Paul Erdős) worthy of publication in academic journals, according to Sebastian Bubeck.

Math serves as a benchmark for AI progress because proofs require long, consistent reasoning and a single error anywhere in the chain destroys the entire argument—a demand that Bubeck argues trains systems in the kind of error-spotting and self-correction capability needed for artificial general intelligence.