Meta AI Chief Yann LeCun Targets 'World Model' for Smarter Machine Intelligence

Yann LeCun, the chief artificial intelligence scientist at Meta, has publicly stated that contemporary AI systems, including widely deployed large language models, are fundamentally 'not smart'. This critique comes as his new start-up, Wayve, backed by NVIDIA, Microsoft, and British investors, seeks to revolutionise the field by developing AI with a more profound understanding of the world.

LeCun argues that current AI merely mimics human behaviour without genuine comprehension. He envisions a 'world model' approach, wherein AI systems would learn the mechanics of their environment through observation and interaction, akin to how infants develop understanding. This would enable them to predict outcomes, grasp cause and effect, and plan actions, moving beyond pattern recognition.

Wayve, a London-based company, is specifically focusing on applying this advanced AI paradigm to autonomous vehicles. The aim is to create self-driving systems that can navigate and react to unpredictable real-world scenarios with human-like intuition and reasoning, rather than relying on extensive pre-programmed rules or vast datasets alone.

The initiative represents a significant departure from the prevailing emphasis on scaling up existing large language models, which LeCun suggests have inherent limitations in achieving true artificial general intelligence. Instead, Wayve's work is posited as a foundational step towards AI that can reason, adapt, and learn in a more flexible and robust manner.