PARIS, July 5, 2026 – Yann LeCun, the legendary AI scientist who spent a decade as Meta’s chief AI officer, has a blunt message for the tech world: today’s most celebrated artificial intelligence systems, including ChatGPT and Gemini, are fundamentally stupid. Speaking at the VivaTech conference in Paris, LeCun argued that these large language models (LLMs) lack any real understanding of the physical world, making them incapable of solving basic real-world tasks like household chores.
LeCun, who left Meta in 2025 to found Advanced Machine Intelligence Labs (AMI Labs), told reporters that current AI systems are "not particularly smart" and "don't have an underlying understanding." He illustrated his point with a simple demonstration: holding a pen upright and letting go. "Even a toddler knows the pen will fall, but no human would try to predict which direction," he said. "An LLM, however, will generate a statistically plausible but almost certainly wrong prediction, because it cannot reason about physical reality."
This critique comes as AMI Labs announced a record-breaking seed funding round of more than $1 billion, backed by Nvidia and a fund managing Amazon founder Jeff Bezos’s private wealth. The investment signals a major bet on LeCun’s alternative approach: Joint Embedding Predictive Architecture (JEPA), a system designed to create abstractions of the real world. Unlike LLMs that regurgitate statistical patterns, JEPA aims to assess the outcomes of actions in unpredictable environments. "We don’t have robots as good at understanding the physical world as a rat," LeCun admitted, framing the challenge as a fundamental shift from text-based reasoning to embodied intelligence.
The timing is critical. As tech giants race to deploy AI in autonomous vehicles, robotics, and logistics, LeCun’s warning underscores a growing consensus that today’s models hit a hard ceiling. Investors are now pouring billions into "world model" startups that promise to bridge the gap between digital data and physical reality. With AMI Labs at the forefront, the next frontier of artificial intelligence may not be smarter chatbots, but systems that finally learn to navigate a messy, unpredictable world.