The Turing Award-winning pioneer has left Meta to launch AMI, aiming to achieve human-level intelligence by teaching machines how the real world actually works.
- A Billion-Dollar Pivot: Former Meta Chief AI Scientist Yann LeCun has launched AMI, a Paris-based startup backed by over $1 billion to pioneer “world models” grounded in physical reality rather than just language.
- Challenging the LLM Hype: LeCun argues that expecting Large Language Models (LLMs) to achieve human-level intelligence is a “delusion,” advocating instead for AI systems equipped with persistent memory, real-world reasoning, and planning capabilities.
- Open-Source and Enterprise-Focused: Valued at $3.5 billion, AMI aims to build open-source models for data-rich industries like manufacturing and robotics, while advocating that the ultimate control of AI belongs to democratic societies, not tech CEOs.
For the past few years, the tech industry has been consumed by a single, prevailing narrative: scale up Large Language Models (LLMs), feed them enough text, and human-level intelligence will inevitably emerge. Yann LeCun, one of the founding fathers of modern artificial intelligence, thinks that idea is entirely backward.
Now, he has the capital to prove it.
LeCun, who departed his role as Meta’s chief AI scientist in November 2025, has unveiled Advanced Machine Intelligence (AMI). Pronounced like the French word for “friend,” the Paris-based startup emerged from stealth on Monday with over $1 billion in funding and a staggering $3.5 billion valuation. Backed by a heavyweight roster of investors—including Bezos Expeditions, Mark Cuban, former Google CEO Eric Schmidt, Xavier Niel, and firms like Cathay Innovation and Greycroft—AMI represents a fundamental shift in how we approach artificial general intelligence.
The Delusion of Language
While AI labs like OpenAI, Anthropic, and even Meta have doubled down on scaling up chatbots like ChatGPT and Claude, LeCun is placing a massive bet against the very technology dominating the zeitgeist.
As a 2018 Turing Award winner, his skepticism carries undeniable weight. LeCun does not dismiss the utility of LLMs entirely—he acknowledges they are highly effective for tasks like code generation—but he views them as the industry’s latest trend rather than a path to true intelligence. In a recent interview, he didn’t mince words: “The idea that you’re going to extend the capabilities of LLMs to the point that they’re going to have human-level intelligence is complete nonsense.”
Instead, LeCun argues that human reasoning is fundamentally grounded in the physical world, not in text. AMI’s mission is to build a new breed of AI systems equipped with persistent memory, the ability to reason and plan, and most importantly, a deep understanding of physical reality.
Building the Physical World in Code
To achieve this, AMI is developing “world models.” Unlike LLMs, which predict the next word in a sequence based on vast libraries of human text, world models predict how the physical environment behaves.
This ambitious vision has immediate, lucrative applications in the enterprise sector. AMI plans to partner with data-rich industries like manufacturing, robotics, and biomedicine. For instance, instead of writing an email, an AMI system could generate a hyper-realistic world model of an aircraft engine. Manufacturers could then interact with this model to optimize fuel efficiency, minimize emissions, and stress-test reliability in a simulated physical space. The startup is already slated to work with major partners like Toyota and Samsung to put this technology to the test.
Life After Meta
LeCun spent years incubating world model research, such as the Joint-Embedding Predictive Architecture (JEPA), inside Meta’s Fundamental AI Research (FAIR) lab, which he founded. However, as Meta pivoted its strategy to catch up in the consumer-facing LLM race, LeCun realized his enterprise-focused research needed a new home.
In November 2025, he pitched his departure to Mark Zuckerberg, explaining that he could build world models “faster, cheaper, and better” outside of Meta by sharing development costs with other enterprises. The split was amicable; Meta and AMI are already discussing potential collaborations, such as integrating AMI’s world models into Meta’s smart glasses.
AMI marks LeCun’s first commercial endeavor post-Meta, and he has assembled a formidable team to execute it. The global operation—with offices in Paris, Montreal, Singapore, and New York—is co-founded by an all-star cast of AI veterans. Alexandre LeBrun, former CEO of Nabla, takes the helm as CEO, while former Google DeepMind researcher Saining Xie steps in as Chief Science Officer. They are joined by former Meta heavyweights including Michael Rabbat, Laurent Solly, and Pascale Fung. LeCun himself will lead the startup while maintaining his professorship at New York University.
Who Controls the Future?
Beyond the technical architecture of AI, LeCun is deeply concerned with its governance. Against a backdrop of increasing geopolitical tension around AI—highlighted by the Pentagon’s recent moves to blacklist Anthropic over the startup’s military use red lines—LeCun is a staunch advocate for open-source technology.
He argues that artificial intelligence is simply too powerful to be gatekept by private corporations. “I don’t think any of us, whether it’s me or Dario, Sam Altman, or Elon Musk, has any legitimacy to decide for society what is a good or bad use of AI,” LeCun stated.
_(cropped).jpg){kind=link}
Having pioneered convolutional neural networks—a technology inspired by human visual processing that is now used both oppressively for state surveillance and defensively in autonomous drones in Ukraine—LeCun understands the dual-use nature of AI intimately. He maintains that in liberal democracies, the democratic process, rather than tech executives, must dictate how technology is deployed.
While AMI plans to release its initial models quickly to its enterprise partners, LeCun’s ultimate goal is anything but modest. The startup is working toward a “universal world model,” a foundational, generally intelligent system capable of powering innovation across every industry on Earth. As the AI industry remains fixated on talking, Yann LeCun is teaching machines how to walk.