Terence Tao, one of the world's leading mathematicians, published a blog post on July 21, 2026 about a potential counterexample to a famous unsolved math problem. In the post, he mentioned that he had used ChatGPT while working through the material (Tao's blog).

The problem is called the Jacobian conjecture, and it dates back to 1939. It involves polynomial maps, which are functions built from basic arithmetic operations on variables — think of them as recipes that take numbers in and produce numbers out using addition and multiplication. The conjecture essentially says that if such a map does not distort space in a certain way, it must be reversible. Despite its simple-sounding statement, the conjecture has resisted proof for over 80 years and has attracted many claimed proofs and counterexamples that later turned out to be wrong. A confirmed counterexample would settle the problem once and for all — in the negative.

Tao's post does not claim to verify or refute the counterexample. He describes the work as a "digestion" of the result, meaning he is trying to understand and explain the mathematics rather than serve as a formal reviewer. His blog has long been a place where he works through math in a semi-public, exploratory way, and this post fits that pattern.

What makes this post unusual is the role of ChatGPT. A commenter on the post noticed that Tao did not correct ChatGPT even once during their conversation (Tao's blog). Coming from the comment section of a post by a Fields Medalist — the highest honor in mathematics — that observation is worth pausing on.

The facts support a limited claim: Tao used ChatGPT as a conversational partner while working through the mathematics, and the exchange went without him needing to correct the model's output. Exactly what ChatGPT contributed — checking calculations, reconstructing arguments, or just serving as a sounding board — is not specified. And the observation of zero corrections comes from a commenter, not from Tao himself. It is one data point, not a thorough test.

The context matters. The Jacobian conjecture lives in an area of mathematics that demands precise, step-by-step symbolic reasoning — the kind of careful algebra where a single sign error can derail an entire argument. Language models have not historically been good at this. Earlier versions routinely made algebraic mistakes that a first-year graduate student would catch right away. The fact that a mathematician of Tao's standing used ChatGPT on this kind of material and found nothing worth correcting in-line tells us something about how far these tools have come.

It is also a single conversation, noted by a commenter, on one specific problem. Broad conclusions would be premature. Mathematicians have been trying out language models for proof assistance, literature search, and informal exploration over the past couple of years, with mixed results. Tao has been open about experimenting with AI tools in his research, and this post fits that pattern of public, practical exploration rather than promotion.

The broader question the post raises, without directly asking it, is whether these tools are crossing a threshold in mathematical conversation. They are not yet replacing formal proof-checking software like Lean or Coq, which provides machine-verified certainty. But they may be becoming reliable enough for a working mathematician to treat as a genuine thinking partner on hard problems. The gap between "no corrections needed" and "the output was actually correct" is significant, and the available facts do not close it. What we have is an observation, not an evaluation.

In my view, the most telling detail is how ordinary this was. Tao used a language model while doing mathematics, mentioned it in a blog post, and a commenter noticed that no errors came up. No press release, no benchmark, no staged demo. The tool is becoming part of the workflow, and the workflow is becoming part of the documentation. That is a quieter and more lasting sign of adoption than any announcement.

The counterexample to the Jacobian conjecture, if it holds, resolves one of mathematics' persistent open questions. Whether ChatGPT played a real role in understanding it or was simply a clean conversational mirror remains an open question of its own.