AI solves 372 previously unsolved mathematics problems

OpenAI has solved 372 previously unsolved mathematical problems, a development described as “totally and globally revolutionary” by KTH mathematics professor Svante Linusson, SVT reported.

Linusson said AI models had progressed rapidly. Early in the year, they could solve some relatively simple unsolved problems. By spring, they were producing research at the level of a doctoral thesis, and by early summer they could tackle problems people had worked on for 80 to 100 years.

The announcement on October 6 has unsettled mathematicians around the world. The 372 solutions are expected to be checked by human mathematicians.

Linusson said the profession’s focus could shift from solving specific problems towards more exploratory work. As mathematics becomes more accessible and is used by more people, he expects demand for mathematicians to remain strong, although their methods and tasks will change.

He also raised questions about how universities should assess doctoral theses and choose candidates for academic jobs when AI can produce sophisticated solutions. “We have to rethink how all this should be done,” he said.