AMI Labs Founder Yann LeCun Raises $1B for Smarter AI Beyond LLMs

Yann LeCun, one of AI's leading figures, left Meta in 2025 to found Advanced Machine Intelligence Labs (AMI Labs) in Paris. His goal: develop a new type of AI not based on large language models (LLMs) like ChatGPT or Claude. With $1B in seed funding from Nvidia and Jeff Bezos’s investment fund, AMI Labs is building Joint Embedding Predictive Architecture (JEPA). LeCun argues LLMs are fundamentally limited—they can regurgitate patterns but lack true understanding of the physical world. JEPA, by contrast, creates abstract representations that filter out irrelevant information, enabling it to reason about real-world actions without needing to predict every detail. For example, when a pen is dropped, an LLM might try to guess its exact fall direction, while JEPA knows that’s unpredictable and focuses on useful outcomes.

LeCun’s approach directly targets the robotics industry, where billions have been invested in humanoid robots that still struggle with basic household tasks like ironing or dishwashing. 'LLMs are largely hopeless for robotics,' LeCun says. He believes scaling LLMs will never lead to human-level intelligence. Oxford professor Ingmar Posner agrees, noting the next decade will focus on systems that can answer causal questions: 'What causes what? What would happen if I did something else?' AMI Labs’ JEPA represents a radical departure from the current AI paradigm, aiming for flexible, world-aware AI that could finally enable robots to navigate messy, unpredictable physical environments.