That is Yann LeCun’s argument in one line. He is Meta’s chief AI scientist, he won the Turing Award for deep learning, and he thinks the thing everyone is currently building is a detour. JEPA is what he proposes instead.
This page explains both — the argument, and the architecture — without the maths.
Nobody shows them a billion miles of footage. They sit in a car, and after roughly twenty hours of practice they can drive — in rain, at night, past a cyclist they have never seen before.
Self-driving has taken decades, tens of billions of dollars, and enormous fleets collecting data, and it still isn’t there. LeCun’s point isn’t that engineers are bad at it. It’s that the teenager is doing something categorically different, and we are not building that thing.
That last number is the interesting one. In four months a baby has seen roughly 960 hours of video through its own eyes — about the size of a standard AI video dataset. And by then it already understands that objects still exist when hidden, that solid things don’t pass through each other, that some things move on their own and some don’t.
The teenager isn’t learning to drive from twenty hours of data. They’re learning to drive from twenty hours plus sixteen years of understanding how the world works.
That accumulated understanding is what LeCun calls a world model. It is the thing that lets you know, without trying it, that driving into the barrier would be bad. Today’s systems have to be shown. That, in his view, is the whole problem.
These get flattened into “LeCun says LLMs are useless,” which is not what he says. He says they are useful and they are not the road to human-level intelligence. Here is the actual chain.
An LLM is trained to produce likely text, not true text. When it doesn’t know, the most likely-sounding continuation is still a confident sentence. Truth was never the target.
Each word carries some chance of being wrong, and the model is committed to it once it’s out. Over a long answer that risk multiplies. There’s no step where it steps back and checks the whole plan.
Humans work out roughly what they mean, then say it. An LLM must emit a token at every step — it has no private, variable-length place to think first. Chain-of-thought is thinking out loud, which helps, but it is still thinking in words, in public, one word at a time.
A hard question and a trivial one go through the same number of layers. Humans spend more thought on harder things. That flexibility — deciding how long to think — is missing.
Language is a compressed summary of experience, written by people who already share a reality with the reader. Learning only from the summary means never learning the thing being summarised. Babies get there without reading anything.
Something that can answer “if I do this, what happens?” fast enough to try a hundred options in your head before moving. That, he argues, is what common sense is — and it is the part nobody has built.
His position paper is called “A Path Towards Autonomous Machine Intelligence” (2022). JEPA is the piece of that path he has actually specified in detail.
If there is one thing to take away, it is this. Everything else in JEPA is machinery around this single decision.
Suppose you show a model the first half of a video and ask it to predict the rest. A generative model — the Sora or GPT approach — tries to produce the actual next frames, pixel by pixel. That is a hopeless target, for two separate reasons.
Judged on reproducing the input exactly. Must model everything, including what doesn’t matter. Can’t represent “many things could happen” except by blurring them together.
Judged on predicting a compressed description. Free to discard detail. “Many things could happen” collapses naturally, because many futures share one gist.
That’s the whole name, by the way. Joint-Embedding: both halves get embedded into the same space. Predictive: one half predicts the other. Architecture: it’s a shape you can build lots of models in, not a single model.
Concretely, for images (this is I-JEPA, the first real one). Take a photo, hide most of it, and make the model guess the gist of the hidden parts from the bit it can see.
Note what never appears: no labels. Nobody tagged these photos. The image supervises itself by having a piece of it hidden. That’s what self-supervised means, and it’s why this can eat raw video off the internet.
If you score a model on predicting its own summaries, it can win by making every summary identical. Predict a constant, score a perfect zero, learn nothing. This is called collapse, and avoiding it is most of the practical difficulty.
LeCun prefers not to ask “what happens next?” at all. He asks “is this a plausible next?” — scoring a pair as low energy if they fit together and high if they don’t.
The difference matters. There are a hundred plausible next seconds, so producing one is guesswork. But judging whether a proposed one is sane is a well-posed question — and judging is exactly what you need when you’re trying out possible actions in your head. Collapse, in this language, is a model that calls everything plausible.
The world model is one box in a larger design LeCun sketched for a complete agent. Worth seeing, if only to know how much of it is still a drawing.
The actor runs in two modes, borrowed openly from Kahneman’s System 1 and System 2.
See it, do it. One pass, no deliberation. Fast and cheap, and roughly what every AI system today does — including LLMs.
Propose a sequence of actions, run them forward in the world model, add up the cost, adjust, repeat. Only then move. This is planning, and it’s what’s missing.
The nice detail is how they connect: once Mode 2 has solved something enough times, a Mode 1 reflex can be trained to shortcut it. That’s learning a skill — a new chess player calculates every line; a strong one just sees the move.
Between the 2022 paper and now, four real models. The pattern is clear: each one makes the representations better, and until the most recent one, none of them did any planning at all.
| model | year | what it added |
|---|---|---|
| I-JEPA | 2023 | Images. Proved the idea works: guess the gist of hidden patches from a visible one. Learned good features faster than the methods it was compared against. |
| MC-JEPA | 2023 | Motion. Learns how things move alongside what they are, and is arguably the first thing with a layered structure like the proposed hierarchy. |
| V-JEPA | 2024 | Video — about 3 seconds of it. The same masking trick punched through a short stack of frames. |
| V-JEPA 2 | 2025 | Scale, and finally a loop that closes: 1B parameters, 22M videos, ~16 seconds of context — plus a robot arm that plans with it. |
Everything before it learned to see. V-JEPA 2 is the first that uses what it sees to decide, and it does it in the shape the 2022 paper described.
The recipe: freeze the video encoder. Train a second network that takes “here’s the scene, here’s the move I’m considering” and predicts the gist of the scene afterwards. Now give it a photo of the goal. At run time, it tries out candidate action sequences internally, scores each by how close the imagined end state lands to the goal, and executes the winner — on a real robot arm, with no training data for that specific task.
That is Mode 2 planning, running on a JEPA world model. Small, clumsy, on a tabletop arm — but it is the loop the whole argument was about, closed for the first time.
V-JEPA hides a patch in the middle of a clip and reconstructs it. That’s not the same as watching ten minutes and predicting the next one. Real anticipation across time is the open problem.
The plan calls for layers — seconds at the bottom, plans at the top. What exists has two levels at most, and nobody knows how many you actually need, or whether it should vary.
The configurator, the memory, the learned critic, breaking a goal into subgoals — all still prose.
No sound, no touch, no language in the loop. A world model for a creature that only sees.
Genuinely unsettled, and it’s worth holding both sides rather than picking a team.
Predicting pixels really is wasteful, and predicting representations really does work better — that part is now just an experimental result, not an opinion. Hallucination really is structural. Nothing in production plans in any deep sense. And a robot arm that reaches a goal it was never shown is a hard thing to wave away.
He has been predicting a wall for years and language models keep clearing it. Reasoning models that think before answering have closed part of gap 3 from inside the paradigm he called a dead end. And text is simply a far richer signal per byte than video — the blog’s own author notes that language models are currently just better, and that whether vision matters at all is speculative.
One framing that helps: LeCun isn’t mainly claiming LLMs will stop improving. He’s claiming there’s a thing they structurally don’t do — hold a model of the world and try things out in it before acting — and that no amount of scaling text prediction produces it. He could be right about the missing piece and wrong about how much it matters, or right about both.
The fair summary in 2026: the core technical bet has been vindicated in a narrow way, and the grand architecture around it is still mostly unbuilt. Ten percent of a long argument has been demonstrated. The rest is a drawing — a detailed, specific, falsifiable drawing, which is more than most positions in this debate come with.