Researcher in mathematics and machine learning @Inria and @ENS_ULM. Tweeting about maths and AI. Feedback: https://xitter.cf/t.co/BUSsBZQ7EL

Paris
Joined April 2012
𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗱𝗲𝗲𝗽 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗰𝗼𝘂𝗿𝘀𝗲🌻𝘀𝘂𝗺𝗺𝗮𝗿𝘆 🧵Thread with links to all slides, videos, notebooks, and @PyTorch code. Solutions are now available on the repo of the course: github.com/dataflowr/noteboo… 1/15
4
127
2
489
75,927
On the Origin of False Proofs by Means of Bugs in Proof Assistants by Yannick Foster @InHenriPoincare automath.dma.ens.fr/ To learn a bit more about @leanprover and @CoqLang
2
6
516
What happened to test time scaling? One year ago, we used absurdly large pass@k for problems you can now one-shot. Is test time scaling still used? Or was it used to get better datasets to train new models? Or not used at all, just a matter of burning tokens...
2
11
1,283
Marc Lelarge 🌻 retweeted
OpenAI’s faster Fourier transform is less a mathematical breakthrough and more a malicious exploit of mathematical traditions. argmin.net/p/faster-fourier-…
66
118
56
1,153
365,750
Bonjour Mission pour la Place des Femmes au @CNRS, il faut un compte email cnrs.fr pour pouvoir signaler sur votre plateforme: mpdf.cnrs.fr/ressources/ Je n'en ai pas donc voici mon signalement.
Aux étudiantes et étudiants, aux personnels universitaires et au public en France : il est important de connaître l’affaire Karim Adiprasito. Le tribunal disciplinaire de l’Université hébraïque de Jérusalem a reconnu ce mathématicien coupable de harcèlement sexuel, notamment d’actes à caractère sexuel qualifiés d’indécents, envers sa doctorante, Johanna Steinmeyer. Il a retenu l’exploitation du rapport de pouvoir et la création d’une dépendance affective. Son appel a été rejeté en août 2026. Il s’agit d’une condamnation disciplinaire universitaire. La procédure a duré près de trois ans. Le tribunal d’appel affirme qu’Adiprasito a bénéficié d’une procédure équitable et relève des demandes répétées apparaissant comme des démarches dilatoires. En janvier 2026, Johanna a été retrouvée morte chez elle, à 29 ans, avant la conclusion de l’appel. Raz Kupferman, professeur et mathématicien de premier plan du département, impliqué dans le traitement de cette affaire, a précisé à la demande de ses proches que la cause de son décès n’avait pas été déterminée. Il ne faut ni parler de suicide ni établir un lien de causalité avec cette affaire. Adiprasito a été interdit d’accès au campus pendant cinq ans et condamné à verser l’équivalent de deux mois de salaire brut aux héritiers de Johanna. Le tribunal d’appel a jugé cette sanction clémente. Johanna est morte à 29 ans. Lui poursuit son activité dans une fonction prestigieuse de directeur de recherche au CNRS, à l’Institut de mathématiques de Jussieu–Paris Rive Gauche, financée par l’argent public, donc par les contribuables français. Les institutions françaises doivent répondre : quand ont-elles été informées ? Quelles protections ont-elles mises en place ? Son rôle d’encadrement est-il réexaminé ? Les mathématiques sont importantes. La vie humaine, la dignité et les droits des femmes le sont davantage. Aucun accomplissement scientifique ne justifie l’abus d’autorité.
1
20
3,659
I always learn new things at ENS.
6
645
As correctly mentioned in your technical paper, 3 papers solved the first "open" problem on Gaussian ellipsoid fitting, in August. AI scooped by humans?
mathematicians and muse spark collaborated to solve 6 open problems in math: 1. The Strict Threshold for Gaussian Ellipsoid Fitting 2. Finite-Time Blow-Up of Radial Negative-Energy Solutions for the Mass-Critical Biharmonic Nonlinear Schrödinger Equation 3. Semiabelian Groups Need Not Be Monomial 4. Tightness of the Cycle-Based Relaxation for Completed Length-Three Alpha-Cycles 5. String Two-Point Function = Height Function on a Curve 6. On Solvable Evolution Algebras and a Conjecture by García-Martínez and Pérez-Rodríguez
2
5
1
57
6,081
Graph alignment is all you need slides presented at the workshop organized by @n_keriven @vaiter and Catherine Matias at @_CIRM github.com/mlelarge/chaining…
13
91
5,766
Marc Lelarge 🌻 retweeted
And then we attacked them with AI. More than 30 now have a complete proof or an explicit counterexample. Some have already been checked by mathematicians, a few formalized in Rocq. We welcome reviewers! github.com/graph-theory-AI/G…
More than 800 conjectures in graph theory. If you still have Fable credits, give it a try! mlelarge.github.io/graph-con…
1
7
7
549
Marc Lelarge 🌻 retweeted
Mechanical Turk 2.0
We’re giving scientists, mathematicians, and engineers free access to our frontier models—starting with 10,000 researchers and expanding to 100,000 through 2027. ChatGPT for Academic Researchers is built to accelerate discovery across disciplines.
8
18
4
267
14,419
"In addition to our inner motivation(...)for doing mathematics, we are driven by considerations of economics and status.(...)More than the knowledge, people want personal understanding. And in our credit-driven system, they also want and need theorem-credits." WP Thurston (1994)
Any argument along the lines of "this proof is correct but has negative value" is dumb. Obviously we can just ignore the proof! If the problems is incentives (people don't want to find the right proof any more, glory is already harvested) then let's change the incentives.
3
850
Marc Lelarge 🌻 retweeted
Checking that a major mathematical proof is correct can take years. Formalization—converting the mathematical reasoning into a form computer proof assistants like Lean can verify—can help. Last month, Claude completed the first formalized proof of Fermat’s Last Theorem, one of the most famous theorems of all time. This was a project experts thought would take many years. It is the largest Lean proof ever written. Fermat’s Last Theorem was first proven in 1995 by Sir Andrew Wiles, more than 350 years after it was conjectured. Our proof, which totals over 13 million lines of code, provides machine verification. More importantly, it proves over 29,000 other theorems that the proof requires, across many areas of math which had never before been formalized. We see this as a major step in the long process of firming up the core of mathematical knowledge, building on work from three centuries of mathematicians and hundreds of contributors to Lean and Mathlib. We are optimistic that AI-assisted verification of mathematical proofs will help reduce the burden of refereeing mathematics in an era where more proofs are being produced than ever before. You can read about the process on our Science Blog: anthropic.com/research/forma… And see the complete proof on GitHub: github.com/anthropics/fermat…
667
1,876
795
14,104
4,764,272
Marc Lelarge 🌻 retweeted
Congrats to @Jacob_Tsimerman and welcome to the team, it's going to be fuuuun!!!
Jacob Tsimerman, who won a fields medal, perhaps math's most prestigious prize, just announced at a press conference that he is "pivoting toward AI safety" and will be going to OpenAI soon
11
41
14
602
156,527
Human mathematicians are being outcounterexampled, by Kevin Buzzard.
Sharing my slides "Vibe Proving on a €200 Budget" and practical LLM workflows, tools, prompts, agents, and deployment practices for research. slides: docs.google.com/presentation… repo: github.com/mlelarge/ai-resea…
1
4
35
3,833
Jacobian conjecture, T. T. Moh in Algebra and geometry (1995)
hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final ((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)
3
7
133
17,578
Marc Lelarge 🌻 retweeted
I had the honor of giving a keynote at the International Conference on Machine Learning in Seoul last week titled “What will be left for us to work on?” I addressed the widespread anxiety about how we should adapt as AI capabilities increase. I was thrilled by the talk’s reception, so I have made my slides available, annotated with a lightly edited transcript: cs.princeton.edu/~arvindn/ta… I made three arguments. First, the "AI as Normal Technology" framework is a correct and useful as a way to think about AI’s impacts, unless and until there is some future discontinuity such as through recursive self-improvement. Second, even though we should take recursive self-improvement seriously, there is no milestone that companies might achieve in the lab that will suddenly put us all out of work. Third and finally, jobs of the future will be radically different, and a lot of adaptation will be needed. I shared my thinking about what this might look like and ended with a vision of human/AI “co-superintelligence”.
33
134
20
857
203,980
When AI generates thousands of lines of formalization, we need tools to understand them. Glad to see that @ejgallego is working on this with Verso Blueprint! Very nice talk by @Leonard41111588
If you are in Seoul for ICML Workshops, come to check our latest work with Guillaume Baudart: Library Before Proof: Making LLM-Generated Rocq Usable by Mathematicians (Friday #2408) See you there! #ICML
2
11
1,929
Very nice talk by Remy Degenne: Can we accelerate formal mathematics with AI?
If you are in Seoul for ICML Workshops, come to check our latest work with Guillaume Baudart: Library Before Proof: Making LLM-Generated Rocq Usable by Mathematicians (Friday #2408) See you there! #ICML
7
1,122