OpenAI 模型连续破解数学猜想，Erdős 问题成 AI 试验场
为什么偏偏是 Erdős 的问题？一个流浪数学家留下的 1000 道题，意外成了衡量 AI 数学能力的标尺。
OpenAI 未公开模型推翻了 80 年前的 Erdős 猜想，这是 AI 首次产出历史级数学证明。数学家们随即跟进改进，而一个由英国数学家创建的 Erdős 问题列表网站，成了各大 AI 公司的竞技场。
正文摘录
A "Wandering" Mathematician's Legacy, Now Being Fought Over by AI Companies .jpg) On May 20, 2026, OpenAI released an announcement that shook the entire mathematics community. One of their not-yet-public AI models had disproven a conjecture proposed by Hungarian mathematician Paul Erdős in 1946. It was the first time AI had produced a historically significant mathematical proof — not by tackling some fringe problem, but by finding a counterexample no one had thought of on a classic problem that had stumped mathematicians for 80 years (see "[AI Proves Core Mathematical Conjecture for the First Time: An 80-Year-Old Classic Problem, Solved by OpenAI](https://mp.weixin.qq.com/s?biz=MzA3MzI4Mj…