Yann LeCun, the renowned computer scientist and chief AI scientist at Meta, has spoken out about the nature of intelligence and the current state of artificial intelligence, emphasizing that “intelligence really is about learning.” In a recent conversation, LeCun reflected on his decision to step down from his role at Meta, a move that marks a significant moment in the evolution of the company’s AI strategy. LeCun, widely regarded as one of the founding figures of deep learning, has long argued that true intelligence goes beyond pattern recognition and statistical modeling. He believes that systems must be able to learn from experience, reason about the world, and adapt to new situations—capabilities that current large language models (LLMs) still lack. “Intelligence really is about learning,” he said, underscoring the importance of developing AI that can understand cause and effect, build internal representations of the world, and learn from fewer examples. While LLMs have made remarkable progress in generating human-like text and performing complex tasks, LeCun remains skeptical of their ability to achieve genuine understanding. He pointed out that these models often hallucinate, lack common sense, and struggle with reasoning—issues rooted in their reliance on vast amounts of data rather than real-world interaction and causal reasoning. His departure from Meta comes at a time when the company is reshaping its AI ambitions, particularly following the recent $14.3 billion investment in Scale AI and the hiring of Scale’s co-founder Alexandr Wang. LeCun’s exit signals a shift in leadership and focus, with Meta increasingly prioritizing data infrastructure and model training over foundational research in areas like world models and embodied learning—fields LeCun has championed for years. Despite stepping back from day-to-day operations, LeCun remains deeply involved in advancing AI research. He continues to advocate for a new generation of AI systems that combine deep learning with structured knowledge, predictive modeling, and autonomous learning. He envisions AI agents that can explore environments, learn from feedback, and build mental models of reality—capabilities he believes are essential for achieving artificial general intelligence. LeCun also warned against overestimating the current capabilities of today’s AI. “We’re not close to building machines that truly understand,” he said. “We’re still far from creating systems that can learn like humans do—through curiosity, experimentation, and interaction with the physical world.” As Meta and other tech giants race to deploy ever-larger models, LeCun’s perspective serves as a reminder that progress in AI isn’t just about scale—it’s about rethinking how machines learn, reason, and interact with reality.