AI Model Solves Complex 20-Year Math Problem Designed by Polish Researcher

Polish mathematician Bartosz Naskrecki, who spent nearly 20 years designing a complex research-level math problem for the FrontierMath benchmark, was surprised when AI model GPT-5.4 solved it correctly once in 11 attempts. The problem, spanning 13 pages and involving advanced fields like number theory and algebraic geometry, was previously considered beyond AI's conceptual grasp. Naskrecki praised the AI's solution as impressive and nearly human, though human verification remains necessary.