Yann LeCun, the Godfather of AI, asserts that the tech industry is fundamentally mistaken about its understanding of AI
This is not the first occasion on which LeCun has expressed his disapproval of large language models. The seasoned researcher has consistently argued against the notion that LLMs, regardless of their size, can result in superhuman intelligence.
Yann LeCun, often recognized as the pioneer of artificial intelligence and the former chief AI scientist at Meta, has delivered a significant warning to the tech industry, asserting that large language models, which serve as the primary foundation of the current AI surge, represent a dead end in the quest for superintelligence.
In a discussion with the Financial Times, LeCun remarked that while large language models are beneficial, they are inherently limited due to their reliance on language and their lack of comprehension of the physical world. He emphasized that to attain human-level or superhuman intelligence, systems must be capable of modeling the workings of the real world, rather than depending exclusively on text-based information. He further stated that LLMs are essentially a dead end in the pursuit of superintelligence, as reported.
This is not the first occasion on which LeCun has criticized large language models. The seasoned researcher has consistently opposed the notion that LLMs, even when scaled up significantly, can result in superhuman intelligence, and he has been outspoken regarding what he perceives as their inherent limitations.
LeCun has advocated for an alternative approach focused on what he terms world models, a category of AI systems intended to learn from physical reality rather than language alone. His favored architecture, referred to as V-JEPA, seeks to create internal representations of the world by learning from videos and spatial data. He has asserted that these systems possess the ability to plan, reason, and maintain persistent memory, which forms the foundation of what he calls Advanced Machine Intelligence, or AMI.
Although LeCun did not explicitly mention the reasons for his exit from Meta, he indicated that continuing at the company had become politically challenging. He acknowledged that Meta's CEO, Mark Zuckerberg, was supportive of his world model research, but he noted that the teams assembled for the company’s new superintelligence initiative were predominantly focused on large language models, characterizing them as entirely LLM-centric, as reported by the Financial Times.
LeCun was alluding to the group that transitioned to Meta from Scale AI. In June 2025, Zuckerberg declared the establishment of Meta Superintelligence Labs, a new division dedicated to the advancement of sophisticated AI systems. This initiative is spearheaded by senior leaders, including the former chief executive of Scale AI, Alexandr Wang. Earlier that same month, Meta announced its intention to appoint Wang as its chief AI officer, along with several of his associates, as part of a $14.3 billion investment in Scale AI, as reported by news outlets.
