Yann LeCun, a leading figure in artificial intelligence and a Turing Award winner, is stepping down as Meta’s chief AI scientist to launch a new AI startup focused on advancing machine intelligence. His departure comes amid speculation that he was being sidelined after Meta's CEO, Mark Zuckerberg, created a new Superintelligence Lab led by Alexandr Wang. LeCun has expressed dissatisfaction with the current direction of AI research, particularly the heavy emphasis on large language models (LLMs), which he believes fall short in addressing real-world challenges. In a LinkedIn post, LeCun outlined his startup's mission to develop AI systems that comprehend the physical world, possess persistent memory, and can reason and plan complex actions. He has been a long-time proponent of self-supervised learning, a method where models train themselves without human-labeled data. LeCun argues that self-supervised learning is essential for achieving the next significant advancements in AI, contrasting it with the limitations of LLMs that primarily learn from human-generated text. LeCun’s criticism of LLMs highlights their inability to generalize based on sensory experiences, unlike humans or animals. He points out that LLMs struggle with unpredictability and require vast amounts of training data to perform basic tasks. This limitation underscores the need for a different approach to AI, one that moves beyond the current reliance on LLMs and instead focuses on learning that mimics human understanding and interaction with the world.