"We don't have robots that are nearly as good at understanding the physical world as a rat," says Yann LeCun, one of the leading figures in artificial intelligence. The comment, made on the sidelines of France's VivaTech conference, sets the tone for his latest venture — Advanced Machine Intelligence Labs (AMI Labs), a Paris-based startup he founded in 2025 after a decade as chief AI scientist at Meta.
LeCun's goal is to move beyond current systems like ChatGPT, Claude and Gemini. They have their uses, he says, but will never tackle complicated real-world situations, such as getting a robot to do household chores. "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 explains.
“Yann LeCun has launched AMI Labs with over $1bn to build AI that understands the physical world like animals.”
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 managing the private wealth of Amazon-founder Jeff Bezos.
Large Language Models (LLMs) like ChatGPT excel at coding, maths and generating text, LeCun acknowledges. But 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.
The real world, by contrast, presents a bewildering array of outcomes. LeCun holds a pen upright on its tip. What happens when you let go? Even a toddler knows it will topple over, but no human would guess which direction — it's unknowable. An LLM, however, might try to generate a single prediction based on statistical patterns from its training data. That prediction would almost certainly be wrong, because the system is not reasoning about physical reality — it is generating what appears statistically plausible.
LeCun says his company's system, called Joint Embedding Predictive Architecture (JEPA), is designed to handle such problems. It creates abstractions of the real world that allow it to assess the outcomes of actions. These abstractions involve difficult maths but filter out useless information. The question remains: can JEPA move AI beyond regurgitation and toward genuine understanding?