🤖 Terence Tao on 'Math 2.0': What comes after OpenAI's 'mathocalypse'
In October 2026, OpenAI published 722 manuscripts (covering 372 topics) in the openai/math GitHub repository, presenting solutions or partial results for unsolved mathematical problems—including the Riemann, Birch–Swinnerton-Dyer, and Hodge conjectures. Approximately 50% of the works have already been formalized in Lean, with 3 manuscripts retracted and 14 revised within the first 24 hours. In response, Terence Tao delivered a public lecture titled 'Math 2.0' at Caltech, stating that proofs are no longer scarce and that the growth point is shifting toward understanding, knowledge organization, and the application of mathematics to real-world problems.
🌍 The mass release of AI-generated proofs (approximately 4,000 problems, averaging 3 hours per result) and the parallel First Proof (1stproof.org) initiative with transparent verification are forming a new standard for evaluating AI in mathematics: moving from 'found an answer' to 'verified and understood.' This changes the requirements for proof assistants like Lean, peer review procedures, and how laboratories report not only successes but also failures and computational costs.
👤 The key question raised by Tao is verification: exactly what has been proven, which premises were used, where the result is formalized, and where it requires human interpretation. All artifacts are open: the openai/math repository, the text and slides of Tao's lecture, and the video on YouTube. The status of each topic—whether it is a 'solution' or a 'partial result'—is best determined by consulting primary sources rather than sensational headlines.
Source 1: https://terrytao.wordpress.com/2026/10/10/math-2-0/ Source 2: https://en.wikipedia.org/wiki/OpenAI_bulk_release_of_mathematical_papers