AI guru LeCun will leave Meta to start a new company focusing on "world models".

According to recent reports in US media, Yann LeCun, a French-American scientist hailed as the "godfather of AI," is about to leave Meta after twelve years and plans to found an artificial intelligence company.

As one of the founders of deep learning, LeCun has long maintained a critical stance on industry trends, repeatedly pointing out that the current industry's pursuit of large language models (LLMs) has fallen into an "irrational obsession."

In his view, chatbots like ChatGPT are not the right path to achieving intelligence surpassing human levels. LeCun proposes that the key to truly propelling artificial intelligence towards a general level lies in building "world models"—systems capable of autonomously learning the operating rules of the real world through multiple modalities such as video, spatial information, and sensor data.

He emphasizes that only by training AI within the cognitive framework of the physical world can the limitations of current AI systems in understanding, reasoning, and generalization be overcome, ultimately achieving the grand goal of Artificial General Intelligence (AGI).

This viewpoint coincides with that of Fei-Fei Li, a former Stanford University professor known as the "Godmother of AI." Li also believes that the next frontier in artificial intelligence lies in breakthroughs in "spatial intelligence." She vividly points out that while current large-scale language models are "eloquent," they lack "experience" and struggle to truly understand the world—they excel at processing textual symbols but cannot construct cognition through interaction with three-dimensional space, the flow of time, and physical laws, unlike humans.

She advocates that AI should shift from "text-driven" to "experience-driven," using multimodal perception—especially visual and spatial understanding—to compensate for the current models' severe deficiencies in basic common sense about the real world. The consensus of these two leading scholars reflects a new paradigm shift emerging in the AI field: from relying on massive amounts of textual data to generate language symbols to building intelligent systems capable of perceiving, simulating, and reasoning about the physical world. With Yang Likun's upcoming entrepreneurial endeavors, this concept may give rise to a new generation of AI architectures with greater cognitive depth and environmental interaction capabilities, opening up a new battlefield for the entire industry that transcends the limitations of language models.