On Sunday, July 20, 2026, mathematician and Anthropic researcher Levent Alpöge dropped a bombshell: a counterexample to the Jacobian Conjecture, an 87-year-old problem in algebraic geometry, generated by Anthropic’s Claude Fable 5 model. By Monday afternoon, Terence Tao — arguably the greatest living mathematician — had posted a detailed digestion of the result on his blog. By Tuesday night, Tao had published something stranger: his full, unedited ChatGPT conversation in which he worked through the counterexample, line by line, with OpenAI’s model.

The Hacker News thread hit 832 points and 483 comments within hours. The internet did what the internet does: it marveled at the AI, it marveled at Tao, it debated whether machines had “really” done mathematics. All of that misses the point.

The genuinely new thing that happened this week was not that an AI produced a mathematical result. It was that the world’s most trusted mathematical mind felt compelled to perform his verification in public, using a rival AI, and that performance — not the counterexample itself — became the object of public attention. We are watching the birth of a new scientific ritual, and it will change who gets to claim authority in the age of machine-generated knowledge.

The Counterexample Arrived Without a Peer-Reviewed Paper

Alpöge’s announcement did not come through a journal. It came through a preprint, a blog post, and a shareable Claude conversation. The counterexample — a polynomial map in three complex variables with constant Jacobian determinant 1 that is locally but not globally invertible — was, by all appearances, correct. But correctness in mathematics is not a matter of appearances. It is a matter of proof, and proof requires verification.

Traditionally, verification happens slowly. A paper is submitted, referees are assigned, months pass. The process is opaque, clubby, and — for all its flaws — it has kept mathematics remarkably free of the replication crises that plague other fields. What happened this week was different. The verification happened in real time, in public, and the person doing it was not an anonymous referee but the most famous mathematician on the planet, narrating his thought process through a chatbot interface.

One algebraic geometer, reached by Slack while his advisor was still asleep in California, put it this way: “I woke up to a counterexample that wasn’t peer-reviewed, explained by a blog post, and verified by a chatbot. I’m not sure if I’m living in the future or a parody of it.”

Tao Didn’t Just Check the Math — He Performed the Checking

This is the part the “AI solves math” headlines miss. Tao did not simply read Alpöge’s preprint and issue a statement. He opened ChatGPT, fed it the problem, and walked through the counterexample step by step, publishing the transcript. He used one AI to interrogate the output of another AI, and he let the world watch.

Why? Tao is perfectly capable of verifying a three-dimensional polynomial map on his own. The decision to do it publicly, with a chatbot, was not about efficiency. It was about trust. In a world where AI systems can generate plausible-looking mathematics at the frontier of human knowledge, the old verification infrastructure — a handful of overworked referees — cannot keep up. The new verification infrastructure is the public performance of expertise by a trusted figure, amplified by the very same AI tools that created the need for verification in the first place.

This is not a criticism. It is a description of a shift that is already underway, and that will accelerate. When the next AI-generated breakthrough arrives — in mathematics, in drug discovery, in materials science — the question will not be “is it correct?” but “who performed the verification, and did we get to watch?”

The Spectacle Is the Point, and That’s Uncomfortable

There is something deeply unsettling about this, and not for the reasons the AI skeptics usually cite. The discomfort is not that machines are replacing human creativity. It is that the rituals we use to establish truth are becoming indistinguishable from content.

Tao’s ChatGPT conversation was, in one sense, a rigorous piece of mathematical work. In another sense, it was a piece of media — a shareable artifact that generated engagement, discussion, and a great deal of ad revenue for the platforms that hosted it. The 832 upvotes on Hacker News are not a measure of the counterexample’s correctness. They are a measure of how compelling the performance was.

This is not a slippery slope argument. It is an observation that the boundary between verification and spectacle has collapsed, and that the people who thrive in the new regime will be those who can do both: check the math and put on a show. Tao, to his credit, seems to understand this instinctively. The rest of the scientific community is going to have to learn it, and fast.

The Winner This Week Wasn’t Anthropic or OpenAI — It Was the Ritual

Much of the commentary has framed this as a competition: Claude found the counterexample, ChatGPT helped explain it, which AI is smarter? That framing is a category error. The AI models are tools. The real winner this week was the emerging norm that when a machine produces knowledge, a human being with sufficient stature must publicly wrestle with it before anyone believes it.

That norm is fragile. It depends on the existence of figures like Tao, whose reputation is so secure that watching him think out loud is enough to settle the question. What happens when the verifier is less famous, or when the AI’s output is too complex for any single human to check? What happens when the performance of verification becomes a genre, with its own conventions, shortcuts, and grifters?

We do not have answers to these questions. But we have a case study, dated July 20–22, 2026, that makes them impossible to ignore. The Jacobian Conjecture fell this week. Something else rose in its place — a new way of deciding what counts as true — and we are only beginning to understand what it will demand of us.

Sources

THIS IS WHY I LOVE THE INTERNET!

So much to learn from seeing arguably the greatest living mathematician use AI:

https://t.co/mYtwD84A25](https://x.com/leonnoel/status/2080066269848682766)