timothy gowers, a professor at the university of cambridge and a fields medalist, recently revealed an astonishing experience on his blog: using the unreleased chatgpt 5.5 pro, he solved within one hour a long-standing open problem in combinatorial mathematics that had puzzled the mathematical community for decades. for years, academia generally believed that large models’ ability to handle advanced mathematics was limited to “recitation”—that is, retrieving literature or mimicking known derivations. however, this test completely shattered that prejudice. gowers found that this internal beta model not only identified concise arguments overlooked by human experts but could even independently construct highly original proof frameworks without relying on existing theoretical support.
the target of this breakthrough was a challenging additive number theory problem concerning the upper bound estimate of the diameter of sumsets. previously, isaac rajagopal, a student at mit, had shown that this upper bound grew exponentially. guided by gowers, the model initially improved the upper-bound estimate in just 16 minutes, then autonomously identified key propositions for verification. after about an hour of self-correction, it ultimately submitted a complete proof. upon reviewing it, isaac rajagopal remarked that the logical structure was impeccably sound, with core ideas both original and ingenious—achievements that would make any human mathematician proud, even after weeks of effort.
as ai demonstrates “doctoral-level” original research capabilities, academic ethics and educational systems face profound challenges. gowers points out that these results fully meet the standards required for publication in top-tier journals; however, the preprint platform arxiv explicitly refuses to accept ai-generated content, potentially leaving important breakthroughs stranded in a “communication bottleneck.” meanwhile, medium-difficulty open problems once used to train doctoral students can now be solved by ai in under an hour, forcing human researchers to seek ever deeper and more difficult questions. with introductory research increasingly outsourced to ai, where exactly lies the core competitive edge of human mathematicians? this technological revolution is redefining the boundaries of human intelligence.
