Why Yann LeCun’s New Startup Is Betting on Physics‑Based AI Over Language Models

Advanced Machine Intelligence (AMI), a startup led by Turing‑Award winner Yann LeCun, announced a $1.03 billion seed round with a pre‑money valuation of $3.5 billion. The company brings together top researchers such as Xie Saine, former Meta and NYU scientist, and a senior executive team that includes former Meta AI leaders, venture capitalists, and industry veterans.

The funding surge reflects investors’ belief that AMI is tackling the most pressing bottleneck in current AI: the inability of large language models (LLMs) to understand and reason about the physical world. LLMs operate as massive text‑completion engines, predicting the next token based on statistical patterns, which limits their capacity to capture causality, physics, and real‑time interaction.

LeCun’s team proposes a Joint Embedding Predictive Architecture (JEPA) that departs from pixel‑level generation. Instead of reconstructing every pixel in a video frame, JEPA first encodes high‑dimensional sensory data (vision, audio, sensor streams) into compact abstract representations, stripping away irrelevant visual noise. A predictor then forecasts the evolution of underlying physical states rather than raw pixel values, enabling the system to plan actions conditioned on future outcomes.

This approach mirrors human cognition: we ignore inconsequential visual details and focus on actionable information such as object trajectories and traffic signals. By allocating compute resources to causal inference instead of exhaustive visual rendering, JEPA promises more efficient, reliable, and adaptable AI suitable for safety‑critical applications like medical diagnosis, autonomous driving, manufacturing, aerospace, and biopharma.

AMI emphasizes a long‑term research agenda over short‑term product launches. The company plans to invest heavily in algorithmic iteration and large‑scale distributed systems while keeping its core codebase open‑source and publishing results in top academic venues. This open‑research stance, combined with a roster of world‑class investors—including Nvidia, Samsung, Jeff Bezos’s personal fund, and pioneers such as Eric Schmidt and Tim Berners‑Lee—aims to accelerate the transition from text‑centric AI to truly general, physics‑aware artificial intelligence.