We've published a paper that explains our views on AI competition between the US and China.
The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: https://t.co/TgJBeodWYK
Yann LeCun closed $1.03B for AMI Labs on March 10. Three days later, this paper dropped from his NYU collaborators.
15M parameters. Single GPU. A few hours of training.
LeWorldModel is the first JEPA that trains end-to-end from raw pixels. Two loss terms: predict the next embedding, keep the latent space Gaussian. Previous JEPAs needed exponential moving averages or pretrained encoders to avoid representation collapse. LeWM doesn't.
Six hyperparameters down to one.
The numbers are the story. Foundation-model-based world models require hundreds of millions of parameters and serious compute to plan a control task. LeWM plans up to 48x faster while staying competitive on 2D and 3D benchmarks. The whole thing fits on a laptop GPU.
Look at the trajectory. Yann announced his Meta departure in November 2025 after 12 years and called founding FAIR his "proudest non-technical accomplishment." On March 10, 2026, AMI Labs closed the largest seed round in European history at a $3.5B pre-money valuation. Bezos, Nvidia, Samsung, and Toyota all wrote checks.
Three days later: a paper showing that JEPA-from-pixels is no longer fragile and no longer compute-heavy. The engineering scaffolding that made it look like an academic curiosity is gone.
The authors sit at Mila, NYU, Samsung SAIL, and Brown. None at Meta.
Yann's bet was that the path to machine intelligence runs through world models, not language models. He left a public company to build it. Each JEPA paper from his network resets the assumed cost structure for that bet. This one makes world modeling laptop-cheap.
Meta still has the GPUs. The architecture left.
Karpathy threw a grenade at every senior engineer who still treats LLMs as a toy.
his actual words: the worst thing an expert can do right now is reject them.
most experts read it as a threat, but it's advice.
his framing:
> the gap between "AI tools are bad" and "AI tools are useful when used right" is professional discipline, not capability
> agents have cognitive deficits. they fail in ways nothing in the training set anticipated
> the experts who reject LLMs lose to experts who learn to wrangle them
> "models have so many cognitive deficits. but you can route around them"
routing around the deficits is what CLAUDE.md was invented for.
Karpathy himself wrote 4 rules. across 30 codebases they took my Claude error rate from 41% down to 11%. solid drop.
but his rules pre-date the slop era going public. I bolted on 8 more, tuned to the failure modes that surfaced after January. got it down to 3%.
a CLAUDE.md does not raise Claude's IQ. it lowers his slop floor. that is the entire game.
open the article underneath.
the model is not the bottleneck. your config is.
Andrej Karpathy could have packaged this into a $2,000 masterclass.
Instead, he uploaded it to YouTube for free.
Three hours covering how modern LLMs actually work:
tokenization, neural nets, RLHF, hallucinations, tool use, reinforcement learning, and systems like AlphaGo and DeepSeek.
This isn’t about prompts.
It’s about understanding the machine behind the magic.