Yann LeCun’s new venture, AMI Labs, is developing world‑model systems based on the Joint Embedding Predictive Architecture (JEPA) he pioneered, which trains neural networks to forecast future states in an internal representational space rather than outputting raw pixels or text. The company, headquartered in Paris with additional sites in Montreal, Singapore and a New York office near LeCun’s meeting spot, went public in March 2026 and now employs roughly sixty people, many of whom previously worked with him at Meta where he spent twelve years, seven as chief AI scientist. AMI’s current focus is industrial AI: models that understand and predict the behavior of physical systems such as manufacturing plants or turbojet engines, enabling early anomaly detection (e.g., spotting a sudden strange noise before failure) and robotic reasoning akin to a cat anticipating that pushing a vase off a counter will cause it to fall. LeCun argues these world models will eventually supersede large language models as the core of AI, emphasizing open‑source components and practical deployment rather than speculative existential risk.