Yann LeCun on What Comes After LLMs

Yann LeCun, a pioneer in neural networks, challenges the prevailing reliance on Large Language Models (LLMs) as the sole path to human-level intelligence. He argues that LLMs fail to grasp the physical world because they operate on token-based prediction rather than reasoning through world models. His new venture, AMI, focuses on the Joint Embedding Predictive Architecture (JEPA), which enables systems to predict the consequences of actions in high-dimensional, continuous spaces. LeCun contends that true intelligence requires modeling world dynamics, a capability currently absent from generative architectures. Furthermore, he proposes "Tapestry," an open-source, federated platform designed to ensure global AI sovereignty and cultural representation. By enabling decentralized training and parameter exchange, this approach aims to provide a robust, Linux-like infrastructure that transcends the limitations and proprietary constraints of current industry-standard models.