Terence Tao, widely regarded as one of the greatest living mathematicians, publicly shared his ChatGPT session working through a three-dimensional counterexample to the Jacobian conjecture. The conversation link, originally surfaced on X by Dan Shipper, drew immediate attention after Andrew Conner highlighted how it reveals the connections forming in Tao’s mind. Within hours, the shared link circulated across social platforms, reaching audiences far beyond the mathematics community.
TL;DR: Terence Tao posted a public ChatGPT conversation exploring a counterexample to the Jacobian conjecture, a decades-old unsolved problem in algebraic geometry. Andrew Conner’s X post praising the session as a window into Tao’s reasoning process sent thousands of readers to the shared link, sparking broad discussion about how AI tools participate in advanced mathematical exploration.
What Is the Jacobian Conjecture and Why Does It Matter?
The Jacobian conjecture is a famous unsolved problem in algebraic geometry, first proposed by Ott-Heinrich Keller in 1939. It concerns polynomial maps between coordinate spaces and asks a deceptively simple question: if a polynomial transformation has a constant nonzero determinant of its Jacobian matrix, must the transformation be invertible? The conjecture remained open for over eight decades.
Why does this matter? The conjecture connects to deep questions about the structure of polynomial rings and the relationship between geometry and algebra. Resolve it, and the implications ripple across several areas of pure mathematics. It fails.
A three-dimensional counterexample, which Tao explored in his ChatGPT session, would demonstrate that the conjecture does not hold in the way mathematicians long suspected. The problem is so significant that it appears on Steven Smale’s list of mathematical problems for the twenty-first century, marking it as one of the field’s defining open questions.
Tao’s public engagement with a counterexample — working through it interactively with ChatGPT — signals that new computational tools are entering spaces once reserved for chalkboard proofs and private seminars. The counterexample itself involves specific polynomial mappings where the Jacobian determinant condition holds but invertibility fails, a construction that requires careful algebraic verification.
How Did Terence Tao Use ChatGPT for This Problem?
Tao used ChatGPT as an interactive thinking partner, walking through the components of the three-dimensional counterexample step by step. The shared conversation link, posted publicly, shows him querying the model about specific algebraic structures and checking his understanding of the polynomial mappings involved. He did not ask ChatGPT to solve the problem outright.
Instead, the session reads like a structured dialogue. Tao poses targeted questions about the Jacobian matrix, the determinant conditions, and the properties of the polynomial transformations under discussion. ChatGPT responds with explanations, and Tao refines or redirects the conversation based on those responses. This is working out loud.
Haruhiko Okumura, a professor at Kyoto Sangyo University, shared Tao’s accompanying blog post titled “A digestion of the Jacobian conjecture counterexample” alongside the ChatGPT link, directing further attention to Tao’s written analysis on his WordPress blog. The blog post and the ChatGPT session complement each other — one formal, one exploratory.
The approach matters because it demonstrates a specific use pattern: a domain expert using a language model to externalize and verify intermediate steps in a complex reasoning chain. Tao’s session is not a prompt-engineering showcase. It is a working mathematician doing mathematics with a tool that can hold context and respond to nuance.
What Does the ChatGPT Conversation Reveal About Tao’s Thinking?
Andrew Conner’s X post captured the central appeal of the shared session: “It’s so lovely reading a slice of how his mind works, the connections he’s making.” The conversation transcript exposes the intermediate stages of mathematical reasoning that normally remain invisible. Published proofs show the destination. This shows the route.
Readers can observe Tao testing hypotheses about the polynomial structure, asking ChatGPT to confirm or elaborate on specific algebraic properties, and pivoting when a line of inquiry does not produce clarity. The connections Conner refers to are the links Tao draws between the Jacobian determinant condition, the invertibility question, and the specific construction of the counterexample. Each step is explicit.
This transparency is rare. Mathematicians at Tao’s level typically share polished results through papers or lectures, not raw exploratory dialogs. The ChatGPT session strips away the formal layer and presents reasoning as it happens — with false starts, clarifications, and course corrections intact.
The viral response to the shared link suggests an appetite for this kind of process-level visibility. People are not just interested in the mathematical result. They want to see how a Fields Medalist thinks when the camera is off and the tool is a chat window.
Can Large Language Models Actually Solve Advanced Mathematics?
The short answer is no — not independently, and not at the level of a Fields Medalist working on an open conjecture. ChatGPT did not produce the counterexample to the Jacobian conjecture. Tao was working through a known result, using the model to organize and verify his understanding of its components.
What the session demonstrates is a different capability: language models can serve as interactive scratchpads for experts who already possess deep domain knowledge. ChatGPT can explain algebraic concepts, recall definitions, and respond to targeted queries with enough accuracy to support a working mathematician’s thought process. The model participates. It does not lead.
This distinction matters. The conversation shows that the value of large language models in advanced mathematics lies not in autonomous problem-solving but in augmenting human reasoning. Tao’s questions guide every step. The model follows. When the model’s responses are insufficient, Tao redirects or seeks clarification elsewhere.
The broader implication for AI-assisted mathematics is that these tools occupy a specific niche: useful for exploration and verification, insufficient for independent discovery at the research frontier. Tao’s session is evidence of how a leading mathematician actually integrates a language model into serious work — carefully, interactively, and with full awareness of the tool’s limitations.
How Did the Mathematics Community React to the Shared Session?
The mathematics community responded with visible enthusiasm after Dan Shipper posted a link on X directing followers to Terence Tao’s public ChatGPT share link. The conversation quickly became what Digg described as a “curiosity object” for researchers and technologists alike, spreading rapidly across social media platforms. Andrew Conner, posting on X, captured the collective sentiment when he wrote that reading the session was “so lovely” because it revealed “a slice of how his mind works, the connections he’s making.” This public visibility into a Fields Medalist’s thinking process represents an unusual degree of openness.
Mathematicians rarely expose their intermediate reasoning steps. Tao’s willingness to share the full ChatGPT interaction — including false starts, corrections, and iterative refinement — gave colleagues and students an unfiltered look at how exploratory mathematical thinking actually unfolds. The session circulated widely on X, with users like Haruhiko Okumura sharing both the ChatGPT link and Tao’s accompanying blog post titled “A digestion of the Jacobian conjecture counterexample.” Reaction focused less on the AI tool itself and more on the human reasoning visible in the prompts.
What Is the Three-Dimensional Counterexample Tao Discussed?
The Jacobian Conjecture, first proposed in 1939 by Ott-Heinrich Keller, states that any polynomial map with a constant nonzero Jacobian determinant must have a polynomial inverse. For decades, mathematicians believed this held across all dimensions. Tao’s ChatGPT session focused on understanding a three-dimensional counterexample — a specific construction that demonstrates the conjecture does not hold in higher-dimensional settings. The counterexample required careful tracking of polynomial structures across multiple variables.
According to Geek Haus coverage, Tao used the ChatGPT session to digest and explain how this AI-assisted counterexample works. The conversation walked through the algebraic machinery needed to verify that the Jacobian determinant condition is satisfied while the invertibility condition fails. Tao’s blog post, dated July 21, 2026, served as a companion to the shared session. The three-dimensional case matters because lower-dimensional instances of the conjecture had been verified, making the higher-dimensional failure particularly instructive.
What Are the Limitations of AI in High-Level Theoretical Math?
The shared session makes clear that ChatGPT functioned as a reasoning aid rather than an independent mathematical solver. Tao drove the conversation, posed the critical questions, and evaluated the correctness of each step. The AI model contributed by holding context, suggesting algebraic manipulations, and serving as an interactive scratchpad. However, it did not originate the mathematical insight or verify the counterexample independently. This distinction matters significantly for anyone evaluating AI capabilities in specialized domains.
Large language models operate by predicting likely next tokens based on training data. They lack formal proof verification mechanisms built into their architecture. When Tao needed to confirm a step, he relied on his own mathematical judgment rather than accepting the model’s output at face value. The session demonstrates that AI can accelerate the drafting and exploration phases of mathematical work. It cannot replace the rigorous verification that theorems require. Researchers who treat AI output as automatically correct risk building on unverified claims.
How Should Researchers Approach AI as a Mathematical Collaborator?
Researchers should treat AI models as interactive reasoning partners rather than authoritative sources. Tao’s session provides a template: the mathematician maintains intellectual ownership of the problem while using the AI to offload mechanical calculations, explore variations, and externalize intermediate steps. This approach keeps the human accountable for correctness while extracting productivity gains from the model’s ability to process and recall information rapidly. The key is maintaining a clear division of labor.
Effective collaboration with AI in mathematics requires domain expertise sufficient to catch errors. Tao could identify when ChatGPT produced a flawed algebraic step because he already understood the underlying mathematics deeply. A less experienced user might accept incorrect reasoning without recognizing the problem. The session suggests that AI tools are most valuable when the human collaborator brings strong theoretical grounding and uses the model to extend their working memory rather than substitute for it. Institutions adopting these tools should train researchers accordingly.
Frequently Asked Questions
Did ChatGPT actually solve the Jacobian Conjecture counterexample?
No. The counterexample was an AI-assisted construction, not an AI-generated solution. Tao’s blog post, dated July 21, 2026, describes it as a “digestion” of the counterexample, meaning he worked through and explained an existing result. ChatGPT served as an interactive reasoning aid during the exploration process, but Tao himself drove the mathematical reasoning and verified each step.
Why did Terence Tao make his ChatGPT conversation public?
Tao shared the session via a public ChatGPT link, which Dan Shipper then surfaced on X. Andrew Conner noted on X that the session revealed “how his mind works” and “the connections he’s making,” suggesting the shared link offered educational value beyond the mathematical result itself. The public link allowed anyone to read the full exchange, including Tao’s prompts and the model’s responses.
What specific AI model did Terence Tao use for this session?
The sources do not specify which ChatGPT model version Tao used during the session. The conversation was shared via a standard chatgpt.com share link, as confirmed by Haruhiko Okumura’s post on X. OpenAI’s share feature preserves the full interaction history but does not display the model identifier in the shared view.
Does this mean AI is now doing professional-level mathematics?
Not independently. The session demonstrates that a Fields Medalist used ChatGPT productively as a reasoning collaborator on a serious mathematical problem. However, Tao controlled the direction of the conversation, evaluated correctness, and produced the final mathematical explanation himself. The AI accelerated exploration but did not replace expert mathematical judgment.
Summary
- AI as reasoning aid, not solver: Tao’s session shows ChatGPT functioning as an interactive scratchpad for a working mathematician, not as an autonomous problem-solver. The human expert remains essential for direction and verification.
- Public sharing has educational value: By publishing the full ChatGPT session, Tao gave the mathematics community a rare look at exploratory reasoning by a top-tier researcher. Andrew Conner’s reaction on X highlights how valuable this transparency is.
- Domain expertise is non-negotiable: Effective use of AI in theoretical mathematics requires the human collaborator to possess deep subject knowledge. Without it, errors in AI-generated reasoning go undetected.
- The counterexample is three-dimensional: The Jacobian Conjecture fails in three dimensions, and Tao’s blog post from July 21, 2026 provides a detailed digestion of why this construction works.
- Collaboration patterns are emerging: Researchers can study Tao’s prompting style and interaction pattern as a model for how to integrate AI tools into serious mathematical work without ceding intellectual control.
Read the original sources: Digg coverage, Andrew Conner on X, and Tao’s blog post via Geek Haus.