Meta’s Chief AI Scientist, the renowned computer scientist Yann LeCun, is embarking on a bold new journey, departing the secure environment of Big Tech to pursue a far more uncertain, yet potentially transformative, endeavor. Although his soon-to-be former employer, Meta, will remain involved with his forthcoming enterprise, the company has opted to participate solely as a collaborator rather than as a financial backer, signaling a deliberate choice to maintain a balance between cooperation and independence.
In a public statement posted on LinkedIn last month, LeCun revealed that he would be leaving Meta after an impressive twelve-year tenure marked by groundbreaking contributions to artificial intelligence. During that time, he played two pivotal roles: for five years, he served as the founding director of the Fundamental AI Research group—better known by its acronym FAIR—Meta’s flagship research organization devoted to advancing the theoretical and scientific foundations of AI. For the subsequent seven years, he guided Meta’s overall research trajectory as its Chief AI Scientist, shaping many of the company’s most significant developments in machine learning and computational reasoning.
This year, however, Meta restructured its artificial intelligence division to emphasize a new and ambitious priority: the pursuit of what it refers to as ‘superintelligence.’ The reorganization led to the formation of the Superintelligence Lab, an advanced research unit now directed by Alexandr Wang, the founder of the data infrastructure company Scale AI. This shift reflects Meta’s evolving vision—from exploring fundamental AI science to an intensified quest for systems that could one day meet or surpass human cognitive capabilities.
In an earlier announcement made in November, also via LinkedIn, LeCun disclosed the nature of his next professional chapter. He explained that he is establishing a startup dedicated to continuing the Advanced Machine Intelligence (AMI) research trajectory that he has been developing over several years in collaboration with colleagues at FAIR, New York University, and other institutions around the world. In his own words, the ultimate goal of this new enterprise is to catalyze what he calls the next major revolution in artificial intelligence: the creation of systems with a deep, physical understanding of their environment—machines capable of retaining long-term memory, engaging in complex reasoning, and orchestrating multi-step plans to achieve sophisticated objectives.
Despite his departure, LeCun emphasized that his relationship with Meta will not be entirely severed. Speaking during the AI-Pulse event held in Paris on Thursday, he noted that Meta would remain a strategic partner in his new venture, though it has made a conscious decision not to invest equity in the project. According to LeCun, this arrangement aligns with the mutual recognition that the scope of his new research extends well beyond Meta’s immediate corporate priorities.
During the same talk, LeCun elaborated on one of the conceptual pillars of his upcoming work: the development of ‘world models’—advanced machine learning architectures that enable intelligent systems to construct and operate upon abstract representations of the world. These models are designed to help humans and machines alike make better decisions and predictions based on nuanced and dynamic understandings of physical and social environments. This approach diverges markedly from the Large Language Models (LLMs) currently dominating the AI landscape, which generate responses purely from linguistic data without any true comprehension of the underlying reality such language describes.
LeCun mentioned that Mark Zuckerberg, Meta’s CEO, views this architectural direction with enthusiasm and regards it as a plausible pathway toward the next generation of intelligent systems. Yet both men have come to appreciate that the potential range of applications for such world-model-based systems far exceeds what Meta can reasonably prioritize, given its focus on consumer-scale AI products. As LeCun explained, the variety and breadth of real-world problems that could benefit from this kind of intelligence—spanning robotics, autonomous navigation, scientific discovery, and beyond—are so vast that pursuing them effectively requires an independent structure, unbounded by the strategic aims of a single corporation.
Meta, on its part, has redirected much of its research energy toward the still-theoretical concept of superintelligence: artificial minds that might someday reason with the depth, flexibility, and intuition of human cognition—or even surpass it. The establishment of the Superintelligence Lab in June formally codified that ambition. Zuckerberg, previewing Meta’s vision, has spoken of creating what he calls ‘personal superintelligence’—AI assistants or companions that could one day understand users and their needs so deeply that they perform tasks beyond current human limitations.
As of now, neither Meta nor Yann LeCun has offered further comments or official statements beyond what has already been made public. Nonetheless, this transition marks a pivotal moment for both parties: for Meta, it represents a decisive turn toward the grand challenge of building superintelligent machines; for LeCun, it is an opportunity to explore the uncharted frontiers of machine reasoning and perception through entrepreneurial independence and scientific freedom.
Sourse: https://www.businessinsider.com/yann-lecun-ai-startup-meta-partner-investment-world-models-2025-12
