Yann LeCun holds a pen upright and lets go. A toddler would know it topples over, but no human would guess the direction. Yet a large language model like ChatGPT might try to predict exactly that — and get it wrong, because it is not reasoning about the physical world.
“We don’t have robots that are nearly as good at understanding the physical world as a rat,” says LeCun, one of the leading figures in artificial intelligence. He spent a decade as chief AI scientist at Facebook-owner Meta, but left in 2025 to found Advanced Machine Intelligence Labs (AMI Labs) in Paris.
“Yann LeCun says current AI is 'not smart' and has launched a startup to build a more flexible system.”
His goal is to move beyond systems like ChatGPT, Claude and Gemini. “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 told the BBC on the sidelines of VivaTech, France’s leading technology conference.
Large language models are extremely good at coding, maths and generating text, LeCun acknowledges. But he argues these are well-defined, predictable problems. “They basically just accumulate knowledge... They can regurgitate something, you train them to regurgitate, but they’re not particularly smart. They don’t have an underlying understanding,” he says.
AMI Labs is developing a new type of AI called Joint Embedding Predictive Architecture (JEPA). It creates abstractions of the real world that allow it to assess the outcomes of actions — filtering out useless information through difficult maths. In the real world, outcomes are bewilderingly varied, demanding a more flexible intelligence.
Investors see potential. Earlier this year AMI Labs announced it had raised more than $1bn (£760m) in seed funding — one of the largest such rounds in Europe. Backers include US computer chip giant Nvidia and the fund that manages the private wealth of Amazon-founder Jeff Bezos.
LeCun’s bet is that the next leap in AI will come not from bigger language models, but from systems that genuinely understand the physical world — even if they still lag behind a rat.