Yann LeCun, a prominent artificial intelligence researcher who previously served as chief AI scientist at Meta, departed that position in 2025 and established Advanced Machine Intelligence Labs (AMI Labs) based in Paris. His stated objective is to develop AI systems that move beyond the capabilities of current large language models such as ChatGPT, Claude, and Gemini, which he contends lack the flexibility needed to address real-world challenges, particularly in robotics and household automation.
LeCun asserts that large language models, while proficient at well-defined tasks such as coding and mathematical computation, fundamentally lack genuine understanding and reasoning capabilities. These systems are trained to recognize statistical patterns rather than comprehend physical reality. He illustrates this limitation using a simple example: while a toddler intuitively understands that an upright pen will fall when released, a language model might generate an implausible prediction based on statistical patterns in its training data.
To address these limitations, AMI Labs is developing an alternative approach called Joint Embedding Predictive Architecture (JEPA), designed to create abstractions of the real world that enable AI systems to reason about outcomes and possibilities. Earlier this year, the company announced securing more than $1 billion in funding from investors including Nvidia and a fund managing Amazon founder Jeff Bezos’s private wealth, representing one of Europe’s largest seed-stage funding rounds.
Multiple researchers across the AI field share LeCun’s perspective on the need for alternative approaches. Oxford University’s Ingmar Posner and his team are advancing what they call mechanistic world models, while other organizations including Google’s DeepMind and London-based Wayve are pursuing related research directions. LeCun indicates that AMI Labs will spend the remainder of this year refining its model, with plans to deploy it in industrial settings beginning next year.
Regarding broader implications, LeCun suggests that future AI systems will ultimately serve human purposes under human direction, comparing the relationship to that between leaders and their staff of assistants. He maintains that humans will remain essential for determining what questions to ask and what to create.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI