Yann LeCun Launches AMI Labs with $1.03B to Build World Models

Yann LeCun Launches AMI Labs with $1.03 Billion to Build World Models

Yann LeCun, Turing Award winner and former Chief AI Scientist at Meta, has raised $1.03 billion to launch AMI Labs, a startup dedicated to building world models—AI systems designed to understand and interact with the physical world. The funding, valuing the company at $3.5 billion pre-money, marks one of the largest early-stage investments in AI history and signals a decisive shift away from large language models (LLMs) toward perceptual, physics-aware artificial intelligence.

How World Models Differ from LLMs

According to AI Advances, LeCun has long criticized LLMs for their lack of true understanding, calling them "stochastic parrots" that mimic text without reasoning about reality. AMI Labs aims to solve this by developing AI that learns from raw sensory input—video, audio, and sensor data—to build internal models of how objects, forces, and environments behave. This approach, inspired by human and animal cognition, enables machines to predict outcomes, plan actions, and adapt dynamically without relying on massive datasets of human-generated text.

The Role of Self-Supervised Learning

LeCun’s vision, detailed in a series of technical papers and public talks, emphasizes hierarchical predictive learning, where AI constructs a mental model of the world through observation and trial, much like a child learns to grasp objects or anticipate falling. AMI Labs is reportedly developing a new architecture called "Ego-Centric Predictive Learning," which prioritizes first-person perspective data from robots and cameras to train models that understand cause-and-effect in 3D space.

AMI Labs’ Technical Roadmap

Target applications include robotics, autonomous vehicles, healthcare diagnostics, and video understanding systems. Bonega.ai reports that AMI Labs has already partnered with several robotics firms and medical imaging startups to pilot its technology. Unlike current AI systems that require retraining for new tasks, world models aim to generalize across domains—enabling a single AI to navigate a factory floor, interpret an MRI scan, and predict weather patterns using the same underlying framework.

Why This Shift Matters for Real-World AI

Wired highlights that this approach could revolutionize fields where safety and precision are paramount. For example, a robot equipped with a world model could safely assist in surgery by anticipating tissue deformation, or a self-driving car could predict how a pedestrian might step into the road based on subtle body cues—not just pattern recognition.

From Meta to Independence: The Catalyst for Change

LeCun’s departure from Meta in late 2025, after over a decade leading AI research, underscores the urgency he places on this new direction. Sources indicate he left due to internal resistance to moving beyond LLMs at Meta, despite his public advocacy for world models. The funding round, led by a consortium of venture capital firms including Andreessen Horowitz and Sequoia, includes strategic investors from the robotics and healthcare sectors.

While skeptics question whether world models can scale effectively without the data abundance that powers LLMs, LeCun argues that efficiency and generalization will ultimately outperform brute-force training. AMI Labs’ team includes former Meta AI researchers, robotics engineers from Boston Dynamics, and neuroscientists from MIT and the Max Planck Institute.

As the AI industry grapples with the limitations of current generative models, Yann LeCun’s $1.03 billion bet on world models represents not just a new company—but a potential paradigm shift in how machines perceive, learn, and act in the real world. The success of AMI Labs could redefine the future of artificial intelligence, placing physical understanding at its core. Yann LeCun’s world models may be the next frontier.