Yann LeCun Launches AMI Labs to Build AI World Models

Yann LeCun, the AI expert who led research at Meta for years, has started AMI Labs, a new company in Paris aimed at building AI systems that truly understand the real world. He left his role at Meta in late 2025 to lead this effort as executive chairman, with the startup already valued at $3.5 billion after its launch in October 2025.

Background

LeCun built his name in AI over decades. He won the Turing Award, the top prize in computer science, for his work on deep learning, the tech behind today's AI boom. At Meta, he ran FAIR, the company's AI research lab, where he pushed ideas that shaped modern systems. But he grew frustrated with the heavy focus on large language models, the tech that powers chatbots like those from OpenAI and Google.

Those models excel at text but often make up facts, a problem called hallucinations. LeCun wanted to go further, toward AI that grasps physics, plans actions, and handles real-world tasks without errors. He calls these 'world models,' systems that learn from sights, sounds, and sensor data to predict and interact with the environment.

Paris makes sense as home base. LeCun, born in France, picked the city to tap into Europe's growing AI scene. French President Emmanuel Macron welcomed the move, saying France would support the company. AMI joins other players there like Mistral AI. The startup will have offices in New York, Montreal, and Singapore too, making it global from day one.

LeCun still teaches at New York University, one class a year, and guides students. He plans to stay based in New York but run AMI from Paris headquarters.

Key Details

Alex LeBrun runs AMI as CEO. He co-founded Nabla, a company that makes AI tools for doctors, with offices in Paris and New York. Facebook bought his earlier startup, Wit.ai, back in 2015. LeBrun then worked under LeCun at Meta's FAIR lab. Now at AMI, he brings know-how in turning AI research into real products.

Team and Partnerships

Laurent Solly, who was Meta's vice president for Europe until last December, has joined too. This group mixes top research with business experience. Nabla and AMI struck a deal: Nabla gets early access to AMI's world models, and in return, its board let LeBrun shift roles—he's now chief AI scientist and chairman there.

AMI's name stands for Advanced Machine Intelligence. Its website lays out the plan: build world models for reliable AI. These systems would have memory, reason step by step, plan ahead, and stay safe and controllable. No more guessing like in text-based AI.

The company eyes big fields. Healthcare tops the list, where errors can harm people. Tools could help diagnose patients, plan treatments, and watch vital signs. Other areas include factory controls, robots, wearable tech, and automation.

"Real intelligence does not start in language. It starts in the world." – AMI Labs mission statement

AMI plans to license its tech to companies and share research openly with universities through papers and open-source code. Meta might buy in as a first customer, LeCun has said.

This fits a trend. Other startups chase world models. World Labs, started by Fei-Fei Li, another AI star, made Marble, a tool for 3D worlds that follow real physics. It hit unicorn status fast and seeks more cash at $5 billion valuation. AMI focuses more on basic research for broad uses, not quick products.

What This Means

AMI Labs marks a shift in AI. Most money has gone to language models that chat and generate text or images. LeCun bets on models that handle the messy real world—think robots grabbing objects or doctors using sensor data without mistakes. Success could make AI safer for high-risk jobs.

Healthcare stands to gain most. LeBrun told reporters his main draw was applying world models there. Nabla's tools already aid clinicians; AMI's tech could make them sharper, cutting errors in busy hospitals.

In factories and robotics, reliable AI means fewer accidents and smoother work. Wearables could track health better, predicting issues before they hit. The open approach might speed up progress worldwide, as researchers build on AMI's work.

Europe gets a boost. Paris already draws AI talent with labs from Meta and others. Macron's support signals government backing, maybe with funds or rules that favor local startups. This could challenge U.S. dominance, where most big AI firms sit.

LeCun keeps Meta ties but speaks freely against its language model push. His move shows even top researchers want space for new paths. Investors watch close—$3.5 billion valuation means big bets already placed.

Talent drain from big tech is real. AMI pulls in ex-Meta staff, and more may follow. This spreads skills beyond a few giants, sparking competition.

Challenges loom. World models demand huge data and compute power. Unlike text, real-world training needs cameras, sensors, and tests in the field. Early products might take years.

Still, the startup's launch draws eyes. It promises AI that acts like humans do—observing, learning, planning in the physical space. If AMI delivers, it could change how machines fit into daily life, from operating rooms to assembly lines.