李飞飞创办 World Labs

2024 年 9 月，AI 领域最有分量的名字之一——李飞飞——宣布创办一家新公司 World Labs，押注一个当时还冷门的方向：空间智能。在生成式 AI 正忙着生成图片和视频的当口，李飞飞的选择显得与众不同：她要让 AI 理解的不是二维像素，而是三维世界。

李飞飞的判断有其深厚的背景。作为 ImageNet 的缔造者，她深知视觉智能的核心从来不是「认图」，而是理解空间、物体与物理规律。当大模型学会用文字和图像表达世界时，她认为下一个真正的前沿是让 AI 生成「可以被走进的世界」——有几何、有物理、能交互。

World Labs 的速度很快。成立几个月内完成巨额融资，年底就发布了首个产品：一个能根据一句话生成三维世界、并且实时可交互的模型。演示里，用户可以「进入」AI 生成的世界自由探索，画面遵循基本的几何与物理规则。这不再是论文里的示意图，而是能上手体验的东西。

它的意义不止于炫酷。对影视、游戏、机器人训练来说，「用 AI 快速生成可信的 3D 环境」是实打实的基础设施。对具身智能来说，可交互的生成世界意味着更廉价的训练场。World Labs 把世界模型从营销话术变成了产品，也让「空间智能」成为新一轮投资与竞争的焦点。

回看 World Labs 的成立，它像一条新赛道的起跑枪声。生成式 AI 的上半场比的是谁更能「造像」，而下半场的想象是让 AI「造世界」。李飞飞用她的声望和判断力，为这个想象站台——至于这条路能走多远，2024 年秋天只是第一步。

In September 2024 one of the most significant names in AI—Fei-Fei Li—announced a new company, World Labs, betting on a then-niche direction: spatial intelligence. While generative AI was busy generating images and videos, Li's choice stood apart: she wanted AI to understand not 2D pixels but 3D worlds.

Li's judgment had deep roots. As the creator of ImageNet, she knew that the heart of visual intelligence was never "recognizing images" but understanding space, objects, and physical rules. As large models learned to express the world in text and images, she believed the next real frontier was AI generating "worlds you can walk into"—with geometry, physics, and interactivity.

World Labs moved fast. It closed a huge funding round within months and shipped its first product by year-end: a model that generates a 3D world from a single sentence, real-time and interactive. In demos users could "enter" the AI-generated world and explore freely, with scenes following basic geometric and physical rules. This wasn't a diagram in a paper; it was something you could experience.

Its meaning went beyond coolness. For film, games, and robot training, "quickly generating credible 3D environments with AI" is genuine infrastructure. For embodied AI, interactive generative worlds mean cheaper training grounds. World Labs turned world models from marketing talk into product and made "spatial intelligence" the focus of a new wave of investment and competition.

Looking back at World Labs' founding, it sounds like the starting gun of a new track. The first half of the generative-AI era competed over who could "make images"; the second half's imagination is about AI "making worlds". Fei-Fei Li put her prestige and judgment behind that imagination—and how far the road goes, autumn 2024 was just the first step.