Yann LeCun

I do not write posts on X. I tweet links to posts on other platforms. I like and retweet (occasionally) I comment on friends' tweets (rarely) Follow me on...⬇️⬇️⬇️

@giffmana @KinasRemek @ClementDelangue @atreides_sf I trolled OpenAI when they didn't initially release gpt2 because oooooh soooo dangerous 😅

@ylecun @KinasRemek @ClementDelangue @atreides_sf What is triton? What is clip? What is gpt2? What is tiktoken?

well let's be real, open model SOTA was distilled from claude

@4ldo9 @LiorOnAI 😅 For the record l, Randall picked the name LeWorldModel. And Larry Jackel picked the name LeNet back in at Bell Labs in 1989. Not me.

@LiorOnAI @ylecun @ylecun is really a boss, he called it "Le World Model" and not "Un World Model". It's like going to the Ritz an ordering un croque monsieur, on the menu it will be "Le Croque Monsieur"

Just read LeCun's latest paper. His team trained the first world model that can't collapse. Let me explain why this matters. It's called LeWorldModel. World models predict what happens next physically. Objects moving, falling, colliding. That's the base layer for robots that plan, cars that simulate before they steer, any AI that acts in reality instead of just talking about it. The catch is nobody could train these reliably. The models kept cheating. They'd map every input to the same output. Like a weather app stuck on "sunny" forever. Technically predicting. Completely useless. So teams piled on fixes. Frozen encoders, stop-gradient hacks, 6+ loss hyperparameters. A fragile stack too brittle for production. This team asked a different question. What if you make collapse mathematically impossible? An encoder turns each video frame into a small vector. A predictor takes that vector plus an action and guesses the next one. First loss: how wrong was the guess. Second loss: a regularizer called SIGReg that checks if vectors spread out like a bell curve. If they start looking the same, the loss spikes. The model can't cheat because the math won't let it. That simplicity is what makes the results possible. Six hyperparameters became one. 15M parameters. Trains on one GPU in hours. Plans 48x faster. Encodes with ~200x fewer tokens. Open-source. I could run this on my own hardware. Which changes who gets to build physical AI. Not just big labs anymore. Any team, any startup, any grad student. LeCun has pushed JEPA as the path forward. The criticism was always training instability. This paper removes that objection. Two directions compete in AI right now. Bigger LLMs with more compute. Or small models learning physics from raw pixels.

@kostea12 @ClementDelangue @atreides_sf Decency?

@ylecun @ClementDelangue @atreides_sf Why would they give back?

@NandinoAI @ClementDelangue @atreides_sf No. Closed Labs don't publish research papers on their mainstream models.

you are right on the mechanics but wrong on the framing. closed labs hire the researchers who built the open models, train on open datasets, then charge API fees. but open source also benefits from closed lab research papers and architecture innovations. it's not parasitism, it's a feedback loop where closed labs monetize faster and open source democratizes slower. the real question is whether that gap is widening or shrinking. right now it's shrinking.

@KinasRemek @ClementDelangue @atreides_sf Do you think OpenAI didn't profit from all the stuff published/open-sourced by FAIR, DeepSeek and all the others?

@ylecun @ClementDelangue @atreides_sf All? Not. Open models profit from reasoning introduced by OpenAI „O” model. OS benefit a lot just by reverse engendering.

@ClementDelangue @atreides_sf Let's be real, all closed models profit from open models WITHOUT GIVING BACK.

@atreides_sf Let’s be real, closed model sota was distilled from a mix of web and open-source

15 millions de paramètres. 1 seul GPU. LeWorldModel de Yann LeCun est un premier pas vers les « world models » capable de comprendre le monde physique 👉 l.numerama.com/IN0

If Trump wanted to put his face on every coin and bill in America, the Republican Party would lineup like trained seals to applaud.

The US has never had a more stupid president than Trump. That is evident to all outside the US & 58% of the Americans. How pathetic that the GOP is unable to stand up to this liar & madman!

Today we're introducing TRIBE v2 (Trimodal Brain Encoder), a foundation model trained to predict how the human brain responds to almost any sight or sound. Building on our Algonauts 2025 award-winning architecture, TRIBE v2 draws on 500+ hours of fMRI recordings from 700+ people to create a digital twin of neural activity and enable zero-shot predictions for new subjects, languages, and tasks. Try the demo and learn more here: go.meta.me/tribe2

Soon to be announced: first-ever American Bootlicker Award

« Il est temps de pointer du doigt ceux qui se disent ici "Patriotes" ou "souverainistes" mais qui roulent en vérité pour Trump, pour Poutine ou pour Xi Jinping. On les connaît ! La presse le révèle chaque jour : c’est l’AfD, le Rassemblement national, le Fidesz, Vox, Chega ou encore le PVV. Ils siègent à ESN, à l’ECR et chez les soi-disant Patriotes. Mais quand on vote contre le soutien à l’Ukraine, quand on vote contre les sanctions pour la Russie, c’est ça être patriote ? Quand on soutient la politique MAGA de Trump, quand on soutient les géants du web américain contre le DSA, c’est ça être patriote ? La vérité, c’est que ces partis ne défendent pas leur pays. Ils ne défendent pas leur peuple. Ils défendent des intérêts étrangers en affaiblissant l’Europe. Et je le rappelle au passage : de tels agissements dans nos États membres seraient passibles de condamnation pour haute trahison, d’intelligence avec l’ennemi ou d’atteintes aux intérêts fondamentaux de la Nation. » — @ValerieHayer

Frankly I’m shocked that not many more American startups and big tech have noticed the massive opportunity and gap in the market for American open-source AI

After @Pinterest @Airbnb @NotionHQ @cursor_ai, today it’s @eoghan @intercom publicly sharing that they’re finding it better, cheaper, faster to use and train open models themselves rather than use APIs for many tasks. And hundreds of other companies are doing the same without sharing. Ultimately, I believe the majority of AI workflows will be in-house based on open-source (vs API). It took much more time than we anticipated but it’s happening now!

Reminder that the United States could have been the world leader in 5G technology instead of China if we had just given *one guy* a green card when he needed one.

💥🧨 French Senator Claude Malhuret DESTROYS the ENTIRE Trump administration over Iran, corruption and incompetence ! We NEED more leaders like him ! Watch and share 👇

If there is precisely one thing you watch today, make it this. French Senator Claude Malhuret. A microphone. And the most magnificently savage dismantling of the Trump administration ever delivered in a language they almost certainly don’t speak. He covers Iran. He covers corruption. He covers the kind of staggering, industrial-scale incompetence that would get you fired from managing a car park. And he does it with the calm, unhurried certainty of a man who has read every page of the indictment and found it, if anything, worse than expected. France has never pretended to like these people. But this is contempt elevated to an art form. The kind of refined, aristocratic disdain that takes centuries of civilization to produce and approximately ninety seconds to deploy. Malhuret sounds like he is four seconds from the button. Not out of panic. Out of sheer, exhausted disgust. Honestly? Understandable. Watch it. Share it. The adults are speaking. Gandalv / @Microinteracti1

Trump is so unpopular in Germany that even leaders of the far-right Alternative for Germany (AfD) are shunning him. trib.al/27HTl6B