Trending on X: OpenAI's internal model autonomously cracks an 80-year-old math conjecture
What Happened
OpenAI's unreleased internal model produced a novel proof for an open problem in combinatorial geometry tied to Paul Erdős. Fields medalist Tim Gowers called it "a milestone in AI mathematics." The result emerged not from a one-shot prompt but from an extended autonomous reasoning loop, and OpenAI says the same harness is being pointed at other open problems.
My Take
Watch what this does to "AI can't really reason" as an enterprise objection. Boards have been clinging to the idea that LLMs are stochastic parrots — a peer-reviewable math result puts a stake through that. The more interesting consequence is internal: every CEO who told their CFO "we don't need AI for hard work yet" just had their timeline compressed by a year. Expect a new round of strategy-deck rewrites by Q3.
Read Original Source