We don't have robots that are nearly as good at understanding the physical world as a rat, says Yann LeCun – one of the world's leading AI researchers – and his new start-up aims to change that.
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 mission: to build a more flexible artificial intelligence that can handle the messiness of the real world, something he argues systems like ChatGPT, Claude and Gemini will never achieve.
“Yann LeCun raises $1bn for new AI system that he says will surpass current models like ChatGPT.”
"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.
Investors are betting he is right. Earlier this year AMI Labs announced it had raised more than $1bn (£760m) in seed funding – one of the largest such rounds in Europe – with backers including US computer chip giant Nvidia and the fund that manages the private wealth of Amazon-founder Jeff Bezos.
LeCun's criticism of current AI is blunt. Large Language Models (LLMs) like ChatGPT are extremely good at coding, maths and generating text, he acknowledges. But he argues these are well-defined, predictable problems. "They [LLMs] 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.
To illustrate the limitation, LeCun holds a pen upright on its tip. A toddler knows it will topple, but no human would guess which direction it falls – it's unpredictable. An LLM, however, might try to generate a single prediction based on statistical patterns. That prediction would almost certainly be wrong, because the system is not reasoning about physical reality.
AMI Labs is developing a system called Joint Embedding Predictive Architecture (JEPA) that creates abstractions of the real world – filtering out useless information – to assess the outcomes of actions. It is designed to handle the bewildering array of outcomes that real-world problems present, like getting a robot to do household chores. LeCun believes this is the path to intelligence that rivals not just humans, but even a rat.