News summary produced by Claude AI
Yann LeCun, a prominent AI researcher who departed Meta after a decade as chief AI scientist in 2025, has established Advanced Machine Intelligence Labs (AMI Labs) based in Paris. The startup secured over $1 billion in seed funding earlier this year from investors including Nvidia and a fund managing Amazon founder Jeff Bezos’s private wealth, marking one of Europe’s largest early-stage fundraising rounds.
LeCun contends that current large language models such as ChatGPT, Claude, and Gemini, while effective at specific tasks like coding and text generation, lack the flexibility required for real-world applications. He argues these systems cannot develop genuine understanding of physical reality and are consequently unsuitable for training robots to perform household tasks. Rather than accumulating and regurgitating patterns from training data, LeCun asserts that artificial intelligence must reason about actual physical environments to handle unpredictable outcomes.
AMI Labs is developing an alternative approach called Joint Embedding Predictive Architecture (JEPA), designed to create abstractions of the real world enabling machines to evaluate action consequences. This methodology differs fundamentally from language model approaches by filtering irrelevant information and maintaining useful representations of physical reality. LeCun illustrated the limitation of current systems using the example of a falling pen—while a child intuitively understands it will topple, a language model would likely generate an incorrect prediction based on statistical patterns rather than physical reasoning.
LeCun’s perspective aligns with researchers across the AI field exploring alternative architectures called World Models. Ingmar Posner, professor of Applied Artificial Intelligence at Oxford University, advocates for systems capable of causal reasoning and scenario prediction. Multiple organizations, including DeepMind’s Genie model and San Francisco-based World Labs founded by AI pioneer Fei-Fei Li in 2023, are pursuing similar research directions.
LeCun stated that AMI Labs intends to refine its AI model through the remainder of this year, with initial industrial deployment planned for next year. He envisions eventually developing general-purpose intelligence systems requiring minimal training for diverse applications, with humans continuing to determine what questions require answering and what should be created, while AI functions as an assistive tool rather than a replacement for human direction.