Meta's Yann LeCun to Launch Physical AI Startup After Declaring LLMs 'Dead End' (observer.com)

Yann LeCun, Meta’s chief AI scientist and a Turing Award winner, is reportedly leaving Meta to start a new AI company focused on "world models" — systems that perceive and reason about the physical environment rather than only generating language. His exit follows Meta’s recent reorganization that has funneled resources into a superintelligence push (TBD labs) and deprioritized long‑term exploratory work at FAIR, which LeCun helped found and which contributed to early Llama models. The move underscores internal shifts at Meta — including LeCun now reporting to Alexandr Wang — and marks another leadership loss for FAIR after Joelle Pineau’s departure. Technically, LeCun’s startup signals a broader pivot in AI research away from text‑only large language models (LLMs) toward embodied, multi‑modal models that learn physics, causality, spatial relations and action planning. LeCun has long argued LLMs are a “dead end” for human‑level AI, advocating systems that ground perception, prediction and control in real‑world dynamics. His effort joins parallel initiatives (Fei‑Fei Li’s World Labs, DeepMind’s Genie work, Nvidia’s Cosmos) and could accelerate research into simulation‑based training, sensorimotor grounding, and models that combine vision, dynamics and decision making — a shift with major implications for how the community funds, benchmarks and architectures future AI systems.