Founder/Chair, AMI Labs; Professor, NYU; Partner, 224 Ventures; Ex-Chief AI Scientist, Meta. Researcher in AI, ML, Robotics, etc. ACM Turing Award Laureate.
New York
Joined June 2009
- Tweets28K
- Following803
- Followers1.4M
- Likes33.4K
Yann LeCun retweeted
Several comments:
1. image/video SSL methods differ by whether they use reconstruction or not. Joint embedding architectures Amy or may not have a predictor (JEA vs JEPA). They also differ by their anti-collapse method and data corruption/augmentation methods.
2. DINO *is* a kind of JEPA (technically a JEA). The difference with other JEPA models is the anti-collapse criterion (based on contrative clustering+ stopgradient/EMA), and the data augmentation scheme (image transformations, which is task specific),
3. V-JEPA 2.1, like DINO, is from Meta-FAIR (corporate lab), involved significant engineering teams, large/proprietary datasets, considerable computing resources, lots of fine-tuning, and number-pushing on many benchmarks.
4. Many JEPA papers are applications and/or fine-tuned versions of DINO, V-JEPA or I-JEPA.
5. LeJEPA, LeWM, and most other papers with "JEPA" in the title are *research* papers focused on the methodology, often done by students in academic labs, involving no engineering support, very little computing resources, and small/public training sets. They are generally tested on toy problems, simulated environments, or small/public benchmarks.
6. You simply cannot compare 5 with 2/3/4. 5 gives you ideas and new methods, not models. 2/3 give you well-engineered models, with lots of empirically validated tricks. Only rarely will 2/3 give you a new, well-principled method with room to spare. 5 give you empirical results on new applications.
7. What is published in 5 are methods that will eventually become like 2 and 3, whenever someone with resources picks it up and scales it.
DO NOT CONFUSE UPSTREAM FOUNDATIONAL RESEARCH ON METHODS AND DOWNSTREAM SCALING -> ENGINEERING -> NUMBER PUSHING -> MODELS -> PRODUCTS.
The former makes the rest possible, but it takes time.
Yann LeCun retweeted
Interested in building what’s next in AI for Science? Join us at PolymathicAI in New York (NYU + Flatiron).
We are looking for: Interns, Postdoctoral Researchers, Research Scientists and Research Engineers.
Submit your interest here: forms.gle/EMfMccvPZ3S19BLD6
Yann LeCun retweeted
Watching all these Republican attempts to explain why they lost the election is painful, because they cannot come out and say one simple truth. They handed the party to a corrupt, criminal, dictator-loving, economic and strategic idiot--who proceeded to be all of these things.
Yann LeCun retweeted
- Je ne vois AUCUNE raison de continuer à entraîner des modèles d’IA toujours plus puissants.
- Guérir le cancer ?
- Oui bon. Mais à part guérir le cancer ?
- La possibilité de vieillir en bonne santé, sans oublier qui sont ses enfants, sans porter des couches ?
- Admettons. À part le cancer et la vieillesse en bonne santé ?
- L’espoir pour les personnes handicapées de retrouver la parole ou la vue ? De traiter la maladie de Charcot, la sclérose en plaques et Parkinson ? D’inventer de nouveaux vaccins pour qu’un parent ne voie plus jamais son enfant mourir ?
- Tu ne vas pas me faire tous les progrès médicaux un par un ! À part le cancer, la vieillesse en bonne santé, la fin du handicap ou de la mortalité infantile ?
- Des robots pour que plus personne ne soit obligé de nettoyer des toilettes, de déboucher les égouts, de porter des sacs de ciment sous la canicule, de risquer sa vie sur un chantier ?
- Je te concède tout ça ! Mais à part le cancer, la vieillesse en bonne santé, la fin du handicap ou de la mortalité infantile et la disparition des métiers pénibles ?
- La démultiplication des rendements agricoles pour que tout le monde sur la planète mange à sa faim ?
- Admettons. Mais à part le cancer, la vieillesse en bonne santé, la fin du handicap ou de la mortalité infantile, la disparition des métiers pénibles et des famines ?
- L’invention de nouvelles technologies vertes ? Une énergie abondante et propre ? Faire reverdir les déserts ? Dessaler l'eau de mer pour offrir de l’eau potable à tous ? Vaincre le réchauffement climatique ?
- Bon. Mais à part le cancer, la vieillesse en bonne santé, la fin du handicap ou de la mortalité infantile, la disparition des métiers pénibles et des famines, et le sauvetage de la planète ?
- Un monde d’abondance où chacun pourra se loger confortablement, se chauffer l’hiver, et ne plus regarder les prix au supermarché ?
- Ok mais à part le cancer, la vieillesse en bonne santé, la fin du handicap ou de la mortalité infantile, la disparition des métiers pénibles et des famines, le sauvetage de la planète et l’abondance ?
- L’exploration spatiale ? La résolution des mystères de l’univers ?
- Qu’elle s’occupe déjà de nos problèmes concrets…
Yann LeCun retweeted
In the United States, Brazil, Hungary, Serbia and Israel, "voters are turning out in droves to cut autocrats down to size. In country after country stricken by democratic decay, overwhelming turnout in elections is making a difference." trib.al/QeMQyNN
Yann LeCun retweeted
Yann LeCun: The Al industry is stuck because every lab is chasing LLMs. Real Al agents will need to predict the consequences of their actions.
"The Al industry is completely LLM-pilled, as they say."
"Everybody is working on the same thing, they're all digging the same trench."
"It's one big reason I left Meta, right, because Meta also became LLM-pilled."
"I cannot imagine that we can build agentic systems without those systems having an ability to predict in advance what the consequences of their actions are going to be."
Yann LeCun retweeted
1/
A 109M text model trained with good views matched next-token prediction with 42x fewer samples on WIT-1024 (.109 at 21.6M samples vs .111 at 900M).
Scaled up, it reached .184 mutual 10-NN with DINOv2, outscoring Qwen2.5-72B (.154) and Llama-3-70B (.151). 🧵
Yann LeCun retweeted
Join me and 2000 other AI engineers and builders at AI Dev in New York City, Nov 30-Dec 1!
A LinkedIn study found that over July-September, the job title AI Engineer has been one of the 10 fastest-growing occupations in seven major economies. AI engineering has become a core skill for all software work, and AI Dev is designed to help you develop it.
Our sessions give you hands-on practice with:
- Coding agents and harness design: Harness tuning, repository context, autonomous coding
- Agent memory and knowledge graphs: GraphRAG, vector search, persistent memory, hybrid retrieval
- Agents in production: Evals, observability, simulated testing, browser and voice agents
@ylecun and I will also talk about what’s next in world models and physical AI.
I hope to see you there!
Tickets: ai-dev.deeplearning.ai/
Yann LeCun retweeted
Democrats have won more votes than any party in midterm history: 66 million and counting.
Yann LeCun retweeted
Now we'll finally learn what's in the Epstein files. Mike Johnson is no longer in a position to shut down the House in order to prevent votes on measures to force their release. It's time. It's past time.
Yann LeCun retweeted
Introducing ID-WM: a hierarchical world model that decouples "where to go" from "how to get there." Finally, we can plan over hundreds to thousands of steps with WMs.
It beats the strongest baseline by 38.5%, single-level WMs by 135%. #NeurIPS2026 🧵
id-wm.github.io
Yann LeCun retweeted
Plot twist: I was Vapnik's Department Head between 1996 and 2002, and made sure he had the space, resources, and collaborators to do his best work.
To people who think I'm dogmatic: I famously disagreed with Vapnik about the best way forward in ML, but did what I could to promote his work, because research is about exploration and science is about results.
Yann LeCun retweeted
Elon Musk dropped nearly half a billion dollars on the midterms.
98% of the money he spent on specific candidates went to losers.
Yann LeCun retweeted
looped reasoning is completely overrated
Energy-Based Reasoning (EBR) runs circles around looped reasoning
paper dropping this week!
Yann LeCun retweeted
At @224ventureshq, one thing we look for when we speak to founders is speed of execution.
Execution is critical because of the rapidly changing market we are in. When you're starting out, it is often the case that you do not have a moat. You need customer traction first, to learn what matters, improve the product, and to start the data flywheel that improves model performance. This momentum compounds and helps teams build an advantage.
If you are in the position where you already have a moat, speed of execution is no less critical.
You very quickly wind up with the question: how quickly can this team execute, learn, and improve? At the earliest stage, that is how an advantage starts to get built.
Yann LeCun retweeted
If you are at INFORMS in San Francisco, come by room 211 where I will be talking about our latest JEPA research in 10min (8:20am)! Have some time to chat afterwards as well!
Yann LeCun retweeted
LLMs accumulate a lot of declarative knowledge.
That's how they can answer a large number of questions.
That's also why they require so many parameters and hence so much memory.
But one can replace the accumulation of knowledge by *understanding*
That what JEPA world models do.
That's one reason they can be useful with considerably less memory and training data than LLMs.
This paper merely shows that the underlying similarity structure of the data, whether text or image (e.g. co-occurence of words in text or objects in images) uniquely determines the shape of the representation point cloud, when the encoder is trained trained with a common anti-collapse criterion like SIGReg or VCReg (not by reconstruction).