AIBRIEF
His new Paris lab, AMI (Advanced Machine Intelligence) Labs, will pursue JEPA (Joint Embedding Predictive Architecture), a world model architecture that predicts in embedding space — a learned geometry of meaning — rather than over the next word
Yann LeCun has launched a Paris-based research lab called AMI Labs, with a reported $1.03 billion seed round at a $4.5 billion valuation.
The lab is the public vehicle for LeCun's architectural bet against token-prediction LLMs. His proposed path is JEPA (Joint Embedding Predictive Architecture), a non-generative model that learns world dynamics by predicting in embedding space rather than producing text or pixels.
LeCun argues frontier LLMs are limited: a model can describe that pushing a book under a water bottle moves the bottle but lacks an internal representation of the physics behind it. His proposed alternative is a world model that learns the effect directly.
The funding figure comes from a curated AI newsletter rather than a regulatory filing, so it should be read as a reported number. Hacker News and Reddit threads on the launch have been openly skeptical, with practitioners noting that frontier models already handle parts of LeCun's water-bottle example. The Nebius Science interview behind the launch is cited but not independently verified.
The next 12 months will give the bet its first clean test: whether AMI Labs ships an open JEPA model and a robotics or physical-reasoning benchmark that beats text predictors at the frontier.