Terence Tao on AI: His Prediction and 2026 Update

Quick answer: Terence Tao sees AI as a useful research tool whose value depends on verification and human understanding. In June 2023 he predicted that properly used AI, combined with proof checkers, search and math software, would be a trustworthy co-author in research by 2026. In 2026 he called current models “ready for primetime” in math, but also signed a September declaration by 25 Fields Medalists warning that AI companies’ goals are misaligned with mathematics, and in October argued for a “Math 2.0” that values understanding and good questions over mass-produced proofs.

Sources checked November 22, 2026.

Terence Tao, a UCLA professor who won the Fields Medal in 2006, has spent three years experimenting with AI tools in public and writing about them on his blog. His views are practical and carefully qualified, and they shifted during 2026: from “this now saves time” to “we need to protect what makes mathematics valuable.” Here is what he predicted, how he uses AI, and what he says now.

November 2026 Update

On November 20, 2026, Tao summarized three new results produced with heavy AI assistance:

ResultWhoStatus as Tao described it
Finite-time blowup for the unforced 3D Euler equationsAlpöge, Buckmaster and CoiculescuHuman-readable proof, formalized in Lean
Finite-time blowup for unforced 3D Navier–StokesAlpöge and BuckmasterLean formalization; paper “technically human-readable,” but a professionally written proof is still needed
Generalized Riemann HypothesisAlpöge, with an internal AI modelAn enormous Lean-only proof; verification under way

Tao did not endorse the results. He stressed that they still need to be digested and checked, writing that he imagines “the digestion process will take some time.” The next day, the Lean and Mathlib communities certified the formal statement and proof of the Riemann Hypothesis result (see AI math breakthroughs of 2026).

What Did Terence Tao Predict About AI in 2023?

In a June 12, 2023 essay for Microsoft’s AI Anthology, Tao wrote that when integrated with tools such as formal proof checkers, internet search and math software, “2026-level AI, when used properly, will be a trustworthy co-author in mathematical research.”

Did it come true? Partly, and on his own terms. AI tools did make useful contributions inside supervised workflows in 2026, from the OpenAI model that disproved Erdős’s unit distance conjecture in May to machine-checked proofs. Those examples do not show dependable performance across all research, and the prediction never promised autonomous AI. The qualifier “when used properly” carries the weight.

How Tao’s Views Developed: A Timeline

DateWhat Tao said or didSource
June 2023Predicted that “2026-level AI” would be a trustworthy co-author when used properlyMicrosoft AI Anthology
Sept 2024Compared an early OpenAI reasoning model to advising “a mediocre, but not completely incompetent, graduate student”Social media post
Sept 2024Launched the Equational Theories Project, a large collaboration of people and machine toolsTao’s blog
Nov 2025Coauthored a paper testing Google DeepMind’s AlphaEvolve on 67 problemsarXiv
Mar 4, 2026At an IPAM event at UCLA, said current models are “ready for primetime” and save more time than they wasteOpenAI Academy report, Mar 6
July 24, 2026Public lecture “Mathematics in the Age of AI” at the International Congress of Mathematicians, PhiladelphiaSimons Foundation report, Aug 13
Sept 11, 2026Signed “A Severe Misalignment of AI in Mathematics” with 24 other Fields MedalistsTao’s blog
Oct 9, 2026Caltech public lecture “Math 2.0”Lecture slides and blog
Nov 20, 2026Blog summary of AI-assisted Euler, Navier–Stokes and GRH resultsTao’s blog

How Tao Actually Uses AI

TaskTool’s roleTao’s role
Learning backgroundSummaries of unfamiliar areasWorks through the mathematics himself
Code and calculationsWrites quick programs and plotsDecides whether an idea is worth pursuing
Construction searchAlphaEvolve evolves candidate examplesChooses problems, checks and explains results
Formal verificationAI helps write Lean codeChecks that the formal statement means what it should

A concrete case: in his 2026 blog post “A digestion of unit distance constructions,” explaining the AI-found constructions, Tao disclosed that AI helped him get initial summaries of the algebraic number theory involved, while he reorganized the mathematics into a readable exposition. In an October 2026 UCLA profile he described AI as “a firehose of low-quality water” that needs filtering, and said it is “safe to use AI up to the point where you can verify.” For more examples from other researchers, see how mathematicians use AI, and for the tool itself, AlphaEvolve explained.

What His ICM Lecture Said About Proofs and Understanding

At ICM 2026, Tao said AI is getting good at producing and checking proofs, but very long machine proofs may be correct while nobody understands them. Results need explanation and context to become part of mathematics, and current tools are, in his words, “very opaque about their process.” He insisted that the mathematical community, not AI companies, should decide what is acceptable: “We set the rules on what’s acceptable or not.” In an accompanying essay he argued that AI could force mathematics to reconsider its values and practices, an upheaval comparable to the foundational debates of the early 20th century.

Why does a checked proof still need a human? Lean confirms that a proof establishes the statement it was given. It cannot confirm that the statement matches what mathematicians meant, or explain why it is true. A proof of a slightly wrong statement can check perfectly.

The September Declaration

On September 11, 2026, Tao and 24 other Fields Medalists, including Peter Scholze, Maryna Viazovska, Martin Hairer and June Huh, signed “A Severe Misalignment of AI in Mathematics.” It accepts that AI capabilities have improved dramatically, but objects to companies racing to mass-produce solutions to open problems, often announced in a rush without write-ups or credit to earlier work. The signers argue this can displace the human understanding and training that problems are meant to create.

What “Math 2.0” Means

In an October 9, 2026 public lecture at Caltech, Tao argued that “Math 1.0” assumed proofs were scarce, and AI has ended that. His main points:

  • AI is jagged. It is superhuman on some tasks and weak on others; success stories are over-reported, so evaluations should report failures and computing costs.
  • Problems are lighthouses. The work done while trying to solve a famous problem is often worth more than the answer. Solving it too early, without understanding, can leave the surrounding field barren.
  • Verification is necessary but not sufficient. Someone still has to understand the result.
  • Expand the frontier. He wants AI used for large projects, exposition and libraries like Mathlib, not only for solving existing problems.

His concern is not that mathematics will run out of questions; it is the health of the process by which good problems, understanding and new mathematicians develop.

What Students Should Take From His Views

Tao’s own rule, use AI only “up to the point where you can verify,” is a good one for learners. Our recommendation: use AI to find explanations and practice problems, but solve problems yourself first, then compare. The skills Tao values most, asking good questions, checking arguments and explaining ideas, are exactly what you build by learning math properly. See Is Math Still Worth Learning If AI Can Do It?

Common Misconceptions

“Tao thinks AI will replace mathematicians.” No. He expects the job to change, with humans setting rules, choosing problems and explaining results.

“Tao is an AI skeptic.” No. He uses AI regularly and predicted its rise; he is critical of hype, of rushed announcements and of proofs without understanding.

“His 2023 prediction meant AI would work alone.” No. He specified AI combined with proof checkers and other tools, “when used properly.”

Watch: Terence Tao on AI

If you learn better by watching, these videos are worth your time.

ICM 2026 Public Lectures – Terence Tao (Simons Foundation). Tao’s public lecture at the 2026 International Congress of Mathematicians.
No More Heroes – Terence Tao Talks to Numberphile About AI (Numberphile). A relaxed conversation about AI and how mathematical research is changing.
What Is Math For in the Age of AI? – Live from ICM 2026 (Quanta Magazine (The Joy of Why)). A podcast recorded at ICM 2026 on the purpose of mathematics when machines can prove theorems.

Check Your Understanding

Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.

Answer each question and press Submit to see your result right away.

1. What did Tao predict in June 2023?
Found a bug? Let us know!
2. What does Lean confirm about a proof?
Found a bug? Let us know!
3. What was “A Severe Misalignment of AI in Mathematics”?
Found a bug? Let us know!
4. In Tao’s “Math 2.0” lecture, why are open problems like lighthouses?
Found a bug? Let us know!
5. What rule did Tao give for using AI safely?
Found a bug? Let us know!

Frequently Asked Questions

What does Terence Tao think about AI?

He sees AI as a useful research tool whose value depends on verification and human understanding. He uses it regularly, but in 2026 he warned against mass-produced proofs nobody understands and argued that mathematicians, not AI companies, should set the rules.

What did Terence Tao predict about AI in 2023?

In a June 2023 essay for Microsoft’s AI Anthology, he wrote that “2026-level AI, when used properly, will be a trustworthy co-author in mathematical research,” when combined with proof checkers, search and math software.

Does Terence Tao use AI?

Yes. He uses AI to summarize background, write code, search for constructions with AlphaEvolve and help formalize proofs in Lean. He coauthored a 2025 paper testing AlphaEvolve on 67 problems.

What is Terence Tao’s Math 2.0?

A vision he presented at Caltech on October 9, 2026: since AI makes proofs abundant, mathematics should value understanding, exposition, good questions and large collaborative projects, not just solving problems.

What did Terence Tao say about the November 2026 AI proofs?

On November 20, 2026 he summarized AI-assisted results on the Euler equations, Navier–Stokes and the Generalized Riemann Hypothesis, stressing that they still need to be digested and verified.

Keep Exploring

Sources and Further Reading

Related to This Article

What people say about "Terence Tao on AI: His Prediction and 2026 Update - Effortless Math"?

No one replied yet.

Leave a Reply