AI cracks two-decade-old math problem, shocks mathematician

Polish mathematician Bartosz Naskrecki, from Adam Mickiewicz University in Poznań, is amazed as an AI program successfully solves a maths problem he has been working on for nearly 20 years in advanced mathematics research.

The problem, which was part of the FrontierMath benchmark, was intended to test areas of mathematics like number theory, topology, combinatorics, and algebraic geometry to try to push the boundaries of artificial intelligence in advanced mathematics. Naskrecki thought that AI was only capable of being a “very advanced calculator".

The problem, crafted to require deep reasoning and creativity, typically takes highly trained mathematicians weeks to approach. In recent experiments, GPT-5.4 attempted the problem 11 times and succeeded once, producing a solution that Naskrecki calls “very nice, clean, and almost human".

He described the experience as exciting and slightly unsettling, signifying a breakthrough for the AI technology in its ability to cooperate with expert mathematicians.

“This is my singularity moment,” Naskrecki said on X, reflecting about how AI understands his ideas at the same level as the top experts in the world.

Experts, however, are quick to point out that AI technology has yet to replace human mathematicians, as many attempts, computer power, and human verification are needed, and many problems are out of AI’s reach.