In an interview with the UK's Financial Times, Yann LeCun, the former Chief AI Scientist at Meta Platforms Inc. (Meta), expressed strong criticism of Meta's AI strategy, cautioning that the company risks a significant exodus of top talent. LeCun specifically highlighted concerns regarding Alexander Wang, Meta's newly appointed head of AI, arguing that he lacks research experience, fails to grasp the fundamentals of conducting scientific research, and is out of touch with the genuine needs of researchers.

LeCun further contended that Meta's substantial investment in large language models (LLMs) represents a strategic misstep. He explained that while LLMs excel at tasks like text completion, they fall short in understanding the physical world, engaging in long-term reasoning, and executing complex planning—all essential components for achieving human-level intelligence. Instead, LeCun advocates for a shift toward developing "world models," which incorporate visual, auditory, and environmental data to simulate real-world dynamics. Such models, he argues, could enable machines to develop human-like concepts of space, time, and causality.

LeCun also disclosed that Meta has already launched a world model project, which is being trained on an extensive dataset of 1 billion hours of video. He anticipates that a prototype of this groundbreaking system will be ready for release by 2026. The project's primary goal is to enable machines to perform multi-step tasks within virtual environments and to achieve seamless integration with augmented reality (AR) and virtual reality (VR) hardware.