AI software completes a formal proof of Fermat’s Last Theorem. ~ David H Bailey. https:// mathscholar.org/2026/09/ai-sof tware-complete-a-formal-proof-of-fermats-last-theorem/ # AI4Math
We need to separate the "are models good at math?" question from the "do they steal unpublished ideas from users?" question. Of course they're not totally unrelated, but one can concede they have gotten good at attacking math problems while still believing they take ideas from users.
"if we strip away the humans from mathematics the subject will loose its soul. Math is a deeply human endeavor." - words to thrive by
A Lean 4 formalization of Scott's continuous lattices (1972). ~ Lars Warren Ericson. https:// arxiv.org/abs/2606.30782v2 # LeanProver # ITP # AI4Math
AI has solved one of math’s $1 million Millennium Prize Problems. ~ Konstantin Kakaes. https://www. quantamagazine.org/ai-has-solv ed-one-of-maths-1-million-millennium-prize-problems-20260908/ # AI4Math
This is exactly what those companies are after: tricking everybody into the perception that 'all math is solved by AI'. Here's my reply to this.
web.mathi.uni-heidelberg.de/media/Meanin...
Given current events, I think it is worthwhile to start collecting useful online resources with regards to AI and mathematics in general. I will start a list below, but ask contributors to contribute further links in the comments. Note that while I personally do not agree with 100% of the content of every single link […]
This is Jessica. If you’ve ever reviewed a heavily AI-generated paper, you know how unrewarding it can be. For example, last week I spent a morning reviewing a paper for a well-respected general science journal, which sounded interesting but was … Continue reading →
3/3 Some thoughts on the use of AI * Intelligence of this kind should not be treated as an ordinary consumer product whose conditions are set entirely by a few companies. A provider should not be able to collect people’s ideas and information, control all evidence about how they were used, and then exploit that advantage against its own users. Once intelligence becomes basic infrastructure for soc…
2/3 Some thoughts on the use of AI * The more important question is therefore what someone contributes to human mathematical understanding. This includes explaining why an argument works, separating the main idea from technical details, connecting a result with other areas, finding the right questions, teaching new methods, and helping other mathematicians make use of them. Such contributions may…
1/3 Some thoughts on the use of AI A mathematical publication used to bring together three things. It announced a result, identified the people who had produced it, and added something to our human understanding of mathematics. It could therefore serve at the same time as a record of knowledge, a basis for assigning credit, and evidence of a mathematician’s ability. AI breaks this connection. * An…
Ii's near-impossible to show direct evidence for stealing ideas, because AI's are black boxes. But circumstantial evidence (which @gro-tsen.bsky.social 's thread catalogues is piling up, no? And OpenAI explicitly did not rule out the possibility that it made use of unpublished ideas from users.
Robert Wegner, PostDoc at Karlsruhe Institute of Technology I have the impression that recent advances in the deployment of AI for the purpose of mathematical research have made many people uncomfortable. I want to present my personal view on these matters, which is more optimistic. Let me get the following assumption out of the way: […]
Mathathon Organizers: Alvan Arulandu, Sathvik Rodrothu, Caiman Moreno-Earle, Avni Garg, Brian Zhao. On September 4th, we announced Mathathon, an event allocating compute credits across 100 teams of mathematicians tasked with solving and presenting an open problem in 40 hours. As of September 10th, we have received over a thousand applications from undergraduates, Ph.D. students, postdocs, […]
A coalition of current and former Caltech mathematicians Note: The Mathathon organizers have published a response, available here. On October 30th, 2026, Caltech will begin hosting the Caltech Mathathon, a math hackathon in which participants are encouraged to solve open research problems by prompting LLMs. Two prominent AI companies, Anthropic and OpenAI, will supply participants […]
So it's getting clear that some of the math abilities AI models have become ❝superhuman❞ at is “not giving credit where credit is due”, like “failing to cite previous work on the subject that inspired you” and, worse, “stealing colleagues' ideas”.
🧵⤵️
the controversy thickens—now other openai results are being questioned 😈 mathstodon.xyz/@andreasthom... mathstodon.xyz/@andreasthom... mathstodon.xyz/@andreasthom...
What is mathematics now, and what should it be? ~ Jeremy Avigad. https:// proofsandprompts.com/2026/09/0 7/what-is-mathematics-now-and-what-should-it-be/ # AI4Math
(Auto)formalization, but why? ~ Seewoo Lee. https:// proofsandprompts.com/2026/09/0 8/autoformalization-but-why/ # AI4Math # LeanProver # ITP
On the Navier–Stokes Millennium Prize Problem. ~ OpenAI. https:// openai.com/index/navier-stokes -solution/ # AI4Math
AI may have just solved a million-dollar math problem. The field will never be the same. ~ Joseph Howlett. https://www. scientificamerican.com/article /ai-may-have-just-solved-a-million-dollar-math-problem-the-field-will-never-be-the-same/ # AI4Math
Clay Institute won't call Navier-Stokes solved after OpenAI proof claim. ~ Marcus Schuler. https://www. implicator.ai/clay-institute-n avier-stokes-openai-proof-claim/ # AI4Math
Wrote up my thoughts on the whole OpenAI Navier–Stokes Millennium Prize Problem story, and how it highlights the still confusing question of what using my data "to improve model performance" actually means simonwillison.net/2026/Sep/8/o...
RE: https:// mastodon.social/@tristanbuckma ster/117233413705701198 Many of you have already heard about the controversial backstory involving the race between OpenAI and Tristan Buckmaster and Levent Alpöge (who works for Anthropic). If not, read this: https:// mathstodon.xyz/@tristanbuckmas ter@mastodon.social/117233413735526010 and this: https:// mathstodon.xyz/@tao/1172335276 38291447 (2/2)
News flash: OpenAI claims formal proof that Navier-Stokes solutions can blow up. They used 10,000 agents simultaneously, who sent 2.7 million messages and used ~130 billion output tokens. They will not claim the $1,000,000 Millennium Prize for this result. This prize is not going as expected. The first winner refused to take it, saying the math establishment is corrupt. The second possible winner…
Orr Shalit, Professor at the Technion There is nothing new under the sun. Yes, we are living through a technological revolution of dramatic consequences. We who have lived through the birth of personal computers, the rise of the internet, the invention of smartphones, the disruptive eruption of social media, the promise of mass open online […]
Seewoo Lee, postdoc at EPFL. I want to share some thoughts on (auto)formalization based on personal experience, and propose some possible positive and negative usages of it. Disclaimer: I mostly focus on formalizations in Lean, due to my limited experience and knowledge. But most of the discussion can be applied to formalizations using other proof […]
There’s some exciting very recent work by Alpöge and Buckmaster, building upon prior work by Córdoba and Martínez-Zoroa, in the general topic around the infamous global regularity problem for the incompressible three-dimensional Navier-Stokes equations. It is now widely expected that it should be possible to construct smooth initial data and smooth forcing term that would […]
I’ve posted an AI-assisted preprint proposing an almost-all-primes lower bound of c(log p)^3 / log log p for strictly ordered Type-II Erdős–Straus solution counts. A restricted prime-side proof is checked in Lean; independent human review and originality checks remain outstanding. Does anyone know prior work already implying this bound? References welcome. https:// zenodo.org/records/22556477 # Nu…
RE: https:// mathstodon.xyz/@tao/1172309839 38174038 @ tao writes: "For the first time in the history of mathematics, we are faced with the situation that open problems in pure mathematics -- one of the primary fuels for mathematical discovery -- are being mined as a non-renewable resource". Very important thought! But maybe it's renewable if one is willing to solve an already solved…
Verification, formalisation and canonisation: formal theorem proving in the age of AI. ~ Sidharth Hariharan. https:// thefundamentaltheor3m.github.i o/beamers/tifr-2026/slides.pdf # AI4Math # ITP
Li's criterion for the Riemann Hypothesis, in Lean 4. ~ Nicholas Bulka. https:// github.com/nicholasbulka/li-cr iterion-rh-equivalence-lean # LeanProver # ITP # AI4Math
The global Kronecker–Weber theorem in Lean 4. ~ Naganori Yamaguchi. https:// github.com/n-yamaguchi-0729/Kr oneckerWeber # LeanProver # ITP # AI4Math
Erdős problem #501 in Lean 4. ~ Elliot Glazer. https:// github.com/elliotglazer/erdos5 01 # LeanProver # ITP # AI4Math
MathINsiD — a bridge layer between bibliographic and formalized mathematical knowledge You can browse mathlib or Palomar and see what is formalized. You cannot take a published paper and ask whether it has been. MathINsiD indexes that direction: it starts from the literature, not from a repository. ~5M zbMATH Open records — 97.7% of zbMATH — with formalizations attached to the papers they formaliz…
FLT: Anthropic has beaten me to it (one of their internal models, using the prove2.me platform, has formalized a complete proof of Fermat’s Last Theorem (FLT) in Lean). ~ Kevin Buzzard. https:// xenaproject.wordpress.com/2026 /09/04/flt-anthropic-has-beaten-me-to-it/ # AI4Math # LeanProver # ITP
Formalizing Fermat's Last Theorem. ~ Anthropic. https://www. anthropic.com/research/formali zing-fermats-last-theorem # AI4Math # LeanProver # ITP
Kevin Buzzard's thoughts on Fermat's Last Theorem being autoformalized in Lean.
xenaproject.wordpress.com/2026/09/04/f...
Proof of the Lyons-White conjecture. ~ Colin Defant, Ken Ono. https:// arxiv.org/abs/2608.27708v1 # LeanProver # ITP # AI4Math
FrontierMath Erdős (A benchmark of Erdős problems formalized in Lean). ~ Tom Adamczewski, Greg Burnham. https:// epoch.ai/files/frontiermath-er dos.pdf # LeanProver # ITP # AI4Math