Terence Tao Uses ChatGPT to Explore a Jacobian Conjecture Counterexample

The Jacobian Conjecture has resisted proof for over eight decades. It asks whether a polynomial map with a nowhere-vanishing Jacobian determinant must be invertible. No general proof exists, and claimed proofs have repeatedly collapsed under scrutiny.

Terence Tao published a ChatGPT conversation in which he uses the model as a working interlocutor to probe a potential counterexample. The conversation is technically dense: Tao steers the dialogue, checks the model's reasoning, and pushes back where the logic is shaky. The model does not solve the problem. What it does is hold enough algebraic structure in context to make the exchange useful.

For engineers and technical founders watching AI capability, the signal here is not that LLMs do research-grade mathematics autonomously. They do not. The signal is that a mathematician at Tao's level found the interaction worth publishing. That is a calibration point.

The practical implication for builders: LLMs are increasingly usable as reasoning scratchpads for hard, structured problems — not because they get answers right, but because articulating a problem to a model that can reflect back semi-coherent structure surfaces gaps in the human's own framing. Tao's session shows this pattern operating at an extreme end of the difficulty spectrum.

What this is not: a claim that the conjecture is resolved. No counterexample has been verified. The conversation is exploratory, and the mathematical community has not validated any result emerging from it.

What it does confirm: the gap between LLMs as autocomplete tools and LLMs as technical thinking partners is closing, even in domains where formal correctness matters most. The pace at which working mathematicians adopt these tools as part of their visible workflow is itself a leading indicator worth tracking.

Source