The technology behind ChatGPT, Google‘s Gemini, and virtually every AI chatbot flooding the market is fundamentally broken and will never achieve true intelligence, according to Yann LeCun—the pioneering researcher who helped invent the neural networks that made them possible. In a scathing rebuke of his own employer and the entire AI industry, LeCun says the hundreds of billions being poured into large language models represents a historic miscalculation, one he’s now abandoning Meta to correct.The timing of LeCun’s bombshell couldn’t be more dramatic. While tech giants pour hundreds of billions into LLM infrastructure, the 65-year-old AI researcher, widely recognized as one of AI’s three founding fathers, is reportedly launching a startup to pursue “world models”—a competing approach he believes is the only path to machines that can genuinely think. His imminent departure from Meta comes after CEO Mark Zuckerberg effectively sidelined him, restructuring the company around the very LLM technology LeCun has spent years warning against.
Why the godfather of AI thinks everyone got it wrong
LeCun’s critique cuts to the bone of today’s AI hype. Speaking at a Brooklyn symposium, he warned that LLMs are “sucking the air out of the room” while offering no path to human-level intelligence. His assessment is withering: current AI systems are dumber than house cats.The problem lies in how these systems learn. Language models train exclusively on text—words divorced from physical reality. A cat, LeCun argues, can plan intricate movements, understand cause and effect, and navigate complex environments. A four-year-old child absorbs more meaningful sensory data in a day than an LLM encounters in training datasets equivalent to hundreds of thousands of years of reading. Ask today’s most advanced AI to imagine a rotating cube, and it fails at a task any child handles effortlessly.
The billion-dollar divorce from Silicon Valley consensus
LeCun isn’t just any critic—he’s one of three researchers who won computer science’s Nobel Prize equivalent for inventing the neural networks underlying modern AI. Mark Zuckerberg personally recruited him in 2013 to build Meta’s AI division from scratch.But Meta’s recent moves tell the story of an irreconcilable split. The company appointed 28-year-old Alexandr Wang as chief AI officer over LeCun, brought in ChatGPT co-creator Shengjia Zhao as chief scientist, and slashed resources for LeCun’s research group. The message was clear: Meta is all-in on LLMs.LeCun’s planned startup will pursue “world models”—AI systems that learn through visual and spatial data, building internal representations of how the physical world actually works. Stanford’s Fei-Fei Li, Google DeepMind, and Nvidia are exploring similar territory, but progress is measured in decades, not quarters.The dispute reveals an uncomfortable truth: AI’s brightest minds cannot agree on the basic path forward, even as markets bet trillions on one approach prevailing.
