LeCun, set to depart from Meta at the end of the year, openly shared on a podcast his skepticism regarding the potential of large language models (LLMs) to pave the way for artificial general intelligence (AGI). He also expressed his disagreement with Meta's closed - door approach. LeCun has plans to launch a new venture, Advanced Machine Intelligence (AMI), which will be centered around world - model research and open - source initiatives. His ultimate vision is for AMI to emerge as a leading provider of intelligent systems in the future. LeCun highlighted a fundamental distinction between world models and LLMs. World models are designed to process high - dimensional, continuous, and noisy data, which he believes represents the correct path for constructing intelligent systems. In contrast, LLMs are limited to handling discrete, symbolic text data. He argued that relying solely on text - data training would prevent AI from achieving human - level intelligence. Reflecting on the evolution of AI learning methods over the past two decades, LeCun introduced the JEPA architecture. This innovative framework emphasizes "prediction in abstract representation spaces." He firmly believes that the current trajectory of LLMs alone will not lead directly to AGI. According to him, the real challenge in attaining human - level intelligence lies in first reaching the intelligence level of a dog. At the age of 65, LeCun has no intention of retiring. He stated that his life's mission is to enhance human intelligence. Given that Meta no longer supports long - term open research, his departure from the company has become an inevitable outcome.