End of the Language Model Era? “AI Godfather” Yann LeCun Builds a New Kind of AI

The end of 2025 brought news that caused quite a stir in the technology world. Yann LeCun, one of the fathers of modern artificial intelligence and a recipient of the prestigious Turing Award, left Meta and announced the creation of a new startup. It is called AMI Labs (Advanced Machine Intelligence), and its goal is ambitious, to say the least: to build artificial intelligence that truly understands the physical world. Not just text, not just probabilities, but the entire world.

And investors believe in him. In March 2026, AMI Labs announced that it had closed a funding round of more than one billion dollars at a pre-money valuation of 3.5 billion dollars.

Why LeCun Left Meta

LeCun spent more than ten years at Meta. He joined in 2013 to establish the FAIR (Fundamental AI Research) research laboratory and became one of the company’s most prominent figures. But over time, their paths began to diverge. Meta decided to focus on large language models (LLMs) and catch up with OpenAI, Google, and Anthropic. LeCun went in a different direction. For years, he worked on what are known as world models, an approach to AI that does not rely on predicting the next word in a text but instead tries to understand how the world works.

“Meta’s strategy shifted, and it essentially had to catch up with the industry in LLMs and do the same thing as other companies. That doesn’t interest me,” LeCun told WIRED. In November 2025, he went to Mark Zuckerberg and told him that he could do it faster, more cheaply, and better outside Meta. Zuckerberg’s response? “OK, we can collaborate.” The departure was amicable. The door remained open.

What Are World Models?

This brings us to the heart of the matter. What exactly is AMI Labs building?

Large language models such as ChatGPT or Claude work by learning from enormous amounts of text and predicting which word or sentence will come next. They are useful. But they have a fundamental weakness: they hallucinate. They make up facts because their underlying nature is statistical and nondeterministic. LeCun put it this way: “The idea that you can scale up the capabilities of LLMs to the point where they have human-level intelligence is complete nonsense.”

World models work differently. Instead of text, they use data from the real world, whether from cameras, sensors, or industrial equipment. They do not try to predict every detail because the world is inherently unpredictable. Instead, they learn abstract representations of reality and make predictions within this abstract space. They can simulate cause-and-effect relationships, plan sequences of actions, and answer “what would happen if” questions.

The technical foundation is the JEPA (Joint Embedding Predictive Architecture), which LeCun proposed in 2022. While LLMs have learned language, world models are trying to learn the physics of the world.

The Team Behind AMI Labs

LeCun is not the only person behind the project. And that may be one of the reasons investors opened their wallets so quickly.

Yann LeCun serves as executive chairman while continuing as a professor at New York University. Alexandre LeBrun, the former head of healthcare startup Nabla, became CEO. He has had an extensive career in natural language processing. He worked at Nuance Communications, whose technology helped give rise to Apple’s Siri, and led the AI division at Meta.

Saining Xie, formerly a researcher at Google DeepMind, leads the scientific side as chief scientist. Pascale Fung serves as director of research and innovation. Michael Rabbat, Meta’s former vice president of research, leads the world models team. And Laurent Solly, Meta’s former vice president for Europe, oversees operations as COO.

It is a team that knows major technology companies from the inside. And it has decided to forge its own path.

Where and How AMI Labs Will Operate

The startup is headquartered in Paris, but it has operated globally from day one. It is opening offices in four cities: Paris, New York (where LeCun teaches at NYU), Montreal, and Singapore to gain access to the Asian market.

AMI Labs does not plan to generate revenue yet. This is a basic research project, not a startup that will launch a product in three months. LeBrun says so openly: “AMI Labs is a very ambitious project because it begins with basic research.” Even so, the company does not want to remain confined to the laboratory. The first partner to gain access to the early models is healthcare startup Nabla. It was there that LeBrun reached the same conclusion as LeCun: LLM hallucinations can have deadly consequences in medicine.

Other target customers include manufacturers, automakers, the aerospace industry, pharmaceutical companies, and robotics firms. LeCun can also envision consumer applications: “For a home robot to be truly useful, it needs common sense and a genuine understanding of the physical world.” He also mentioned a potential collaboration with Meta on Ray-Ban Meta smart glasses as one of the more near-term applications.

Open Research

One thing about AMI Labs will catch the attention of anyone following the AI industry: the company has committed to open research. It will publish scientific papers and release a large portion of its code as open source. “We think things move faster when they are open, and it is in our best interest to build a community and a research base around us,” LeBrun said.

LeCun goes even further. For him, openness is not just a strategy; it is a conviction. “I don’t think any of us, whether it’s me, Dario Amodei, Sam Altman, or Elon Musk, has the legitimacy to decide for society what is a good or bad way to use AI,” he told WIRED.

The True Future of AI, or Just Another Trend?

We might ask: isn’t this just another trend that will fade away within a year? CEO LeBrun himself warns against excessive optimism. “My prediction is that ‘world models’ will be the next marketing slogan. In six months, every company will call itself a world model to get funding,” he told TechCrunch. He says it with a smile because he considers AMI Labs to be a fundamentally different project.

Competition in this field is growing. Google DeepMind is working on world models. Fei-Fei Li’s startup World Labs raised more than one billion dollars last year. AMI Labs is entering this competitive field with more funding than any of its rivals and a team with decades of basic research behind it.

Time will tell whether world models truly surpass LLMs. It may take years. But one thing is certain: Yann LeCun has staked his reputation, his career, and a billion dollars of other people’s money on being right.