Mathematician Terence Tao described this as “perhaps the clearest example so far” of an AI solving an open mathematical problem. However, earlier proofs may already have existed and could have influenced the model’s response. Tao nevertheless confirms that the proof produced by GPT-5.2 Pro is “quite different” from the previous ones.
At the same time, Tao warns against a distorted perception of AI’s capabilities: negative results are rarely published, while positive outcomes tend to go viral. A new database by Paata Ivanisvili and Mehmet Mars Seven systematically documents AI attempts on Erdős problems.
The actual success rate is only one to two percent, concentrated on easier problems. Even so, Tao sees AI as a useful aid in this area, though Erdős problems of average difficulty remain unsolved so far. The first autonomously confirmed solution of an Erdős problem by an AI dates to January 4, 2026.
ES 
EN