Background
Advanced Machine Intelligence (AMI Labs) is a Paris-based AI research company founded in 2025 by Yann LeCun — Chief AI Scientist at Meta — and entrepreneur Alexandre LeBrun. The company raised a $1.03B seed round at a $3.5B valuation, co-led by HIRO Capital, with participation from Toyota Ventures, Daphni, Founders Future, and others.
AMI Labs' mission is to build AI systems that understand the real world — not just language. Where current large language models excel at text prediction, AMI targets a deeper form of intelligence: systems that can reason about physics, causality, and the dynamics of the physical environment.
LeWorldModel (LeWM) is the first Joint Embedding Predictive Architecture that trains stably end-to-end from raw pixels using only two loss terms. Authored by Yann LeCun, Lucas Maes, and collaborators at Mila and Meta FAIR.
It achieves 48× faster planning than foundation-model-based alternatives by predicting entirely in latent space — no pixel decoding required. A compact ~15M parameter model competitive with much larger systems.
A World Model is an AI system that learns an internal representation of how the world works. Rather than mapping inputs to outputs via pattern matching, a world model builds a compressed, structured understanding of reality — enabling it to predict, plan, and reason about situations it has never directly observed.
Yann LeCun has argued that world models are the missing ingredient for truly general AI. Current LLMs have no grounding in physical reality: they cannot mentally simulate pouring water into a glass, or anticipate the consequences of pushing an object off a table. World models aim to fix this by learning from sensory experience — video, audio, proprioception — rather than text alone.
AMI Labs' architecture, described internally as Alpha Intelligence, is designed around this thesis: a hierarchical latent-space model trained on multimodal real-world data, capable of building and querying an internal world representation at inference time.