Yann LeCun, Chief AI Scientist at Meta and a pioneer in deep learning, is stepping down from his role to launch a new AI startup focused on world-model research, a field he has long championed. The move, confirmed by a Meta spokesperson to Business Insider, marks a significant moment in the company’s AI strategy, especially as Meta continues to restructure its AI division amid internal challenges. LeCun announced the transition on Facebook, emphasizing his intent to build a company centered on developing AI systems capable of understanding and predicting the physical world through sensory data—aligning with his JEPA (Joint Embedding Predictive Architecture) framework, which prioritizes perception and world modeling over text generation. Meta will reportedly partner with LeCun on the venture, though specific details about the collaboration remain undisclosed. LeCun’s exit comes at a time of turbulence within Meta’s AI organization. The company has recently reorganized its AI efforts into four dedicated teams—research, training, product, and infrastructure—under the newly formed Superintelligence Labs, led by former Scale AI CEO Alexandr Wang. This restructuring followed a major push to compete with OpenAI, Google DeepMind, and Anthropic in the race for large-scale AI models. However, the shift has triggered internal friction, particularly between newly hired high-profile researchers and long-time employees, with some existing staff reportedly threatening to leave due to disparities in compensation and integration. The recent release of Llama 4 also received lukewarm feedback, both internally and externally, underscoring growing concerns about Meta’s AI momentum. LeCun’s departure is not entirely unexpected. He has been a vocal critic of the industry’s overreliance on large language models (LLMs), arguing that they represent a narrow path to artificial general intelligence. Instead, he has advocated for systems that learn from raw sensory input—like images and video—to build internal models of the world, a concept he believes is essential for true AI understanding. This vision has increasingly diverged from Meta’s current focus on scaling and commercializing LLMs, especially as the company accelerates its productization of models like Llama. His exit follows the departure of other key figures, including Souminth Chintala, the creator of PyTorch, who left to join Mira Murati’s Thinking Machines Lab. Industry experts see LeCun’s move as a pivotal moment in the evolution of AI. His new venture could help recenter the field on foundational research in perception and world modeling, potentially offering a counterbalance to the current LLM-driven paradigm. As AI development becomes increasingly dominated by corporate interests, LeCun’s shift toward independent, research-driven innovation may signal a broader reevaluation of how intelligence is built. With a legacy spanning decades and foundational contributions to neural networks, his new startup is likely to attract top talent and significant attention, possibly reshaping the trajectory of next-generation AI.