Yann LeCun has been saying for years that large language models are not the path to real intelligence. Now he has $1.03 billion and a Paris headquarters to back it up.
LeCun, the Turing Award winner and former chief AI scientist at Meta, launched Advanced Machine Intelligence Labs, known as AMI, just four months ago. This week the company announced it completed a seed round of $1.03 billion at a $3.5 billion valuation, the largest seed round in European history. Backers include Nvidia, Bezos Expeditions, Toyota Ventures, Samsung, and Mark Cuban among others. LeCun originally aimed to raise around $500 million. Demand pushed it past a billion.
What Is He Actually Building
AMI is working on what LeCun calls world models, an alternative approach to AI that he has been arguing for since at least 2022.
Here’s the simplest way to understand the difference. ChatGPT, Claude, Gemini, all the LLMs you use every day, learn by predicting the next word in a sequence. They’ve ingested enormous amounts of human-generated text and gotten extremely good at producing fluent, plausible language. But they don’t understand the physical world. They can describe a ball rolling down a hill in perfect prose while having no actual model of gravity, momentum, or cause and effect.
LeCun’s alternative is called JEPA, the Joint Embedding Predictive Architecture. Instead of predicting text word by word, it learns abstract representations of how the world actually works. The idea is that true intelligence requires understanding physical reality, not just patterns in language. Factories, hospitals, robots operating in open environments, these systems need AI that can reason about the real world, not just generate plausible-sounding sentences about it.
AMI’s CEO Alexandre LeBrun made a prediction that’s either honest or self serving depending on how you read it: “My prediction is that world models will be the next buzzword. In six months every company will call itself a world model to raise funding.” He said it with a smile. Make of that what you will.
Why Europe and Why Now
LeCun has been explicit about the political dimension of what he’s doing. AMI is headquartered in Paris with offices planned in New York, Montreal, and Singapore. He has described the company as one of the few frontier AI labs that are neither Chinese nor American, a deliberate framing that positions Europe as having a genuine shot at competing at the frontier of AI for the first time in the current cycle.
The founding team is stacked with former Meta AI researchers. LeCun serves as executive chairman. LeBrun, former CEO of medical AI startup Nabla, runs day to day operations. Key hires include Mike Rabbat, former research science director at Meta, and Saining Xie, former research scientist at Google DeepMind.
AMI’s first disclosed partner is Nabla, a digital health startup, which will get early access to AMI’s models for healthcare applications. Robotics, manufacturing, and wearables are other target areas. The company has no plans to generate revenue for the time being and by LeCun’s own account could take years before world models move from research to commercial applications.
Should You Care About This
If you use AI tools for everyday work the honest answer is probably not yet. World models are a long-term research bet, not something that changes your workflow this year or next.
But the broader story matters. LeCun is essentially arguing that the entire trajectory of the last five years, the one that produced ChatGPT and Claude and everything built on top of them, is a detour rather than the destination. That’s a big claim. $1.03 billion in backing from Nvidia, Bezos, Samsung, and a few dozen other serious investors suggests a meaningful number of people think he might be right.
The LLMs are not going anywhere. They’re getting better every few weeks and the productivity gains they enable are real. We reviewed Claude Pro recently and the gap between what these models could do a year ago versus today is staggering. Meanwhile OpenAI is pivoting away from half its product lineup to double down on the LLM approach that LeCun thinks is a dead end.
But the question of whether predicting text is actually the path to general intelligence has never been settled, and LeCun has spent his career arguing it isn’t. Now he has the money to find out. If you want to see what the current generation of AI can actually do right now while LeCun figures out the next one, automation tools like Make.com and n8n are the fastest way to put LLMs to work today.
