Yann LeCun, one of the world’s leading artificial intelligence researchers, has a blunt verdict on the technology behind ChatGPT and its rivals: “They’re not particularly smart.”
Speaking at VivaTech, France’s leading technology conference, LeCun argued that large language models (LLMs) “basically just accumulate knowledge” and “can regurgitate something” but lack any “underlying understanding.”
“Yann LeCun says LLMs like ChatGPT are 'not smart' and launches AMI Labs with $1bn to develop more flexible AI.”
“We don’t have robots that are nearly as good at understanding the physical world as a rat,” he said.
LeCun spent a decade at Facebook-owner Meta as chief AI scientist before leaving in 2025 to found Advanced Machine Intelligence Labs (AMI Labs) in Paris. His goal: build a more flexible AI system that can handle the messy, unpredictable real world – something he says LLMs are fundamentally unsuited for.
“They’re not a path towards human level or human-like intelligence, or even animal-like intelligence, because they cannot deal with real world data, they just are not built for that,” he explained.
To illustrate the limitation, LeCun held a pen upright on its tip. What happens when you let go? A toddler knows it will topple, but no one can predict the exact direction. An LLM, however, might try to generate a single prediction based on statistical patterns – an answer that would “almost certainly be wrong” because the system isn’t reasoning about physical reality.
AMI Labs is developing a system called Joint Embedding Predictive Architecture (JEPA), which creates abstractions of the real world to assess action outcomes. The system filters out useless information, leaving the AI to reason about what matters.
Investors have backed the vision. Earlier this year, AMI Labs announced it had raised more than $1bn (£760m) in seed funding – one of the largest such rounds in Europe. Investors include US computer chip giant Nvidia and the fund that manages the private wealth of Amazon-founder Jeff Bezos.
LeCun acknowledges that LLMs are “extremely good” at coding, mathematical problems and generating text – “well defined and predictable problems.” But for complicated real-world tasks like getting a robot to do household chores, he argues, a different kind of intelligence is needed – one that can deal with “a bewildering array of outcomes.” Whether JEPA is the answer remains to be seen, but the race for what comes after ChatGPT is already underway.