Yann LeCun, Meta’s chief AI scientist and a pioneering figure in artificial intelligence, is leaving the company to launch a new AI startup, Meta confirmed to Business Insider. LeCun announced the transition on Facebook, revealing he is founding a venture focused on world-model research—a long-held passion of his that centers on building AI systems capable of understanding and predicting the physical world through sensory data, rather than relying solely on text-based language models. Meta will partner with LeCun on his new venture, though the specifics of the collaboration remain undisclosed. The move comes amid a period of turbulence within Meta’s AI division. Over recent months, the company has aggressively recruited top talent from rival labs, restructured its AI efforts under the newly formed Superintelligence Labs, and appointed former Scale AI CEO Alexandr Wang as its head. However, the restructuring has sparked internal tensions. Reports indicate friction between newly hired high-profile researchers—often brought in with significant compensation—and long-time Meta employees, with some of the latter threatening to leave. The company’s recent reorganization, which split AI operations into four dedicated teams—research, training, product development, and infrastructure—was aimed at accelerating progress in large-scale AI, especially in response to competition from OpenAI, Google DeepMind, and Anthropic. Despite these efforts, Meta’s Llama 4 release was met with lukewarm reception, both internally and externally, raising questions about the company’s direction. The departure of key figures has also added to the instability. Earlier this week, Souminth Chintala, the creator of Meta’s widely used open-source AI framework PyTorch, left after 11 years to join Mira Murati’s Thinking Machines Lab. LeCun’s exit is not entirely unexpected. For years, he has been a vocal advocate for alternative approaches to AI, criticizing the industry’s overreliance on large language models. Instead, he has championed JEPA (Joint Embedding Predictive Architecture), a method that enables AI to learn about the world through visual and sensory inputs, aiming for deeper understanding and generalization. While Meta has increasingly prioritized scaling language models and commercial AI deployment, LeCun’s vision has long emphasized foundational research and building systems that perceive and reason about reality. His departure marks a significant shift in the company’s AI leadership and signals a potential divergence in strategic direction.