Terence Tao shared his ChatGPT transcript on a math conjecture. The thread tried to unmask it as him

A transcript instead of a paper

The shared ChatGPT conversation that hit 452 comments on Hacker News isn't a summary or a press release — it's a raw back-and-forth between a mathematician and the model, working through a counterexample to the Jacobian Conjecture, a long-standing open problem in algebraic geometry about whether a certain class of polynomial maps must be invertible. Fields Medalist Terence Tao is the widely understood author, based on both the transcript's tone and its follow-up on Hacker News's earlier related discussion of the same counterexample.

What drew readers in wasn't the underlying mathematics — most commenters admitted they couldn't follow the technical substance — it was the register of the questions. One line, quoted repeatedly in the thread: "I've activated Pro. Can you continue to look for a potential geometric explanation of the X_3 ~ A3 miracle that avoids coordinates or other unmotivated constructions?" That's not how a curious amateur talks to a chatbot. It's terse, technically loaded, and assumes the reader (human or model) already has the context — the phrasing of someone who has spent decades inside a field talking to peers who don't need things spelled out.

"It just shows how much you can squeeze out of these models"

The thread's dominant reaction wasn't skepticism about whether an LLM can do real mathematical reasoning — it was fascination at what emerges when an expert, rather than a novice, drives the conversation. One commenter put it: "I know this is 'just' next token inference, matrix mult and just software, ie there's no 'intelligence' there BUT, looking at this convo … damn! The fascinating thing is the LLM is not just following, it's contributing ideas a domain expert has to actually evaluate." Another added that the value here isn't really about the math — it's a rare window into how a world-class researcher structures a problem in real time, redirecting the model, discarding dead ends, pushing toward a specific kind of proof (geometric, coordinate-free) rather than accepting whatever the model offers first.

Several in the thread noted the transcript works as an unusually candid kind of publicity for both the mathematician and the tool — "great PR at least," as one put it — letting outsiders watch expert-level reasoning happen instead of reading about it secondhand in a paper's polished final form.

Ask it to guess who it's talking to

The thread's best aside came from a commenter who'd tried the meta-experiment directly: ask ChatGPT, from within the same or a similar transcript, to guess the identity of whoever it's talking to. Without hints, the model reportedly guesses "math researcher or paper author" on its own — and if fed light context clues ("this was shared over the internet," "it's somewhat famous"), it narrows further. The model isn't just answering technical questions in these sessions; it's picking up on exactly the same register cues — word choice, unexplained jargon, the confidence to redirect its own answers — that tipped off the Hacker News crowd within minutes of the transcript going up. Whether or not Tao is confirmed as the author outside the thread's own inference, the transcript makes a case on its own terms: expertise doesn't just change what questions get asked of a model, it's legible in the asking.