Yann LeCun argues that current large language models have reached their limits and that the next major advance in AI will be “physical AI”—systems that can understand and interact with the real world through sensory data and predictive modeling. He emphasizes the importance of open research and warns that concentrating AI power in a few hands is the greatest risk, while predicting that AI will augment human capabilities rather than cause mass unemployment.
The speaker, Yann LeCun, discusses the current state and future trajectory of artificial intelligence, expressing skepticism about the term “AGI” (Artificial General Intelligence). He argues that human intelligence is not truly general, so aiming for “human-level” AI is a misnomer. LeCun believes that while machines will eventually surpass human intelligence, this will not happen imminently; significant conceptual breakthroughs are still required. He emphasizes that current large language models (LLMs) have reached their limits and that a new paradigm is needed to achieve true intelligent behavior, particularly systems capable of predicting the consequences of their actions—something LLMs cannot do.
LeCun critiques the prevailing assumption that intelligence is primarily about language. He asserts that real intelligence involves understanding and interacting with the physical and social world, which is far more complex than language prediction. The next revolution in AI, according to LeCun, will be “physical AI”: systems that can process high-dimensional, continuous, and noisy sensory data (like video and sensor inputs), build predictive models of their environments, and plan actions accordingly. These systems will be fundamentally different from current generative architectures and will be able to learn from the real world in a way that LLMs cannot.
Reflecting on his time at Meta, LeCun attributes the rapid progress in AI over the past decade not to any single breakthrough, but to the openness of AI research—sharing papers, code, and findings freely. He warns that the recent trend toward closed, proprietary research, especially in Western companies, threatens to slow progress and cede leadership to more open research environments, such as those currently found in China. LeCun advocates for open-source AI platforms, arguing that only open systems can serve as repositories of global human knowledge and support cultural and linguistic diversity.
LeCun’s new venture, Advanced Machine Intelligence (AMI, pronounced “Abi”), aims to develop AI systems based on “world models” that learn from sensory data and physical interaction rather than language alone. These systems are designed to predict the outcomes of actions in the real world, enabling them to plan and reason more effectively. LeCun describes their progress in developing self-supervised systems that can understand and predict video content, detect impossibilities, and acquire a form of common sense. He stresses the importance of abstract, phenomenological models over overly detailed simulations, as the former are more practical for understanding and controlling complex systems.
On the societal impact of AI, LeCun identifies the concentration of AI power in a few companies or governments as the most pressing risk, rather than apocalyptic scenarios. He believes open, diverse AI systems are essential for democracy and cultural health. Regarding economic disruption, he predicts AI will boost productivity but not cause mass unemployment, as the pace of adoption is limited by how quickly people can learn new technologies. For students and workers, he advises focusing on fundamental skills and adaptability. Looking ahead to 2035, LeCun envisions AI systems that deeply understand the physical world, augmenting human intelligence and decision-making, with progress driven by a series of conceptual breakthroughs rather than a single dramatic event.