The development of artificial intelligence (AI) is entering a new stage, which may move from "understanding language" to "understanding the world". On November 10th, Li Feifei, a professor at Stanford University and co-founder of World Labs, published an article titled "From Language to the World: Space Intelligence is the Next Frontier of AI", proposing that space intelligence will become the next peak of AI climbing. She pointed out that although AI systems represented by large language models can proficiently generate text and images, they still remain in the "world of language" and are trapped in "talk on paper", lacking a true understanding of real space, physical laws, and causal relationships. And AI with spatial intelligence will break through this bottleneck, perceiving, reasoning, and acting in the real world like humans. From algorithmic intelligence to spatial intelligence, today's large-scale language models perform well in reading, writing, and pattern recognition, but have fundamental limitations in representing and interacting with the physical world. Even the latest multimodal models often perform worse than random models in estimating distance, direction, and size. They cannot navigate or identify shortcuts in space, nor can they stably predict the most basic physical laws. In addition, although AI generated videos are impressive, they often lose coherence within seconds. These limitations make it difficult for AI to truly empower human creative work. Whether it is helping students understand complex molecular chemistry concepts, assisting architects in conceptualizing spatial structures, or building virtual worlds for directors and game designers, current AI is still far from achieving this goal. In Li Feifei's view, spatial intelligence is the scaffold of human cognitive construction, driving imagination, creation, and reasoning, and is the core ability that connects perception and action. This ability enables people to easily estimate the distance between cars when parking and navigate through crowds with ease. Its core lies in integrating imagination, perception, and action, achieving a leap from "knowing" to "understanding". She believes that spatial intelligence represents transcending the boundaries of language, integrating imagination, perception, and action, allowing machines to not only understand what they see, but also how these things are related, what they mean, and why they are important. This type of AI will truly expand human creativity and understanding, from healthcare to artistic creation, from scientific exploration to daily assistance, and will bring unprecedented changes. Building a world model is the implementation path proposed by Li Feifei in his article. To achieve true spatial intelligence, it is necessary to break through the paradigm of existing large-scale language models and shift towards a more fundamental world model. This new model not only understands semantic relationships, but also consistently "imagines" and "reconstructs" the world in terms of geometric, physical, and dynamic rules. It should be able to perceive multimodal inputs, predict scene changes, and interact with the environment. The current research directions include generating 3D models of objects and scenes, physics engines that simulate dynamic processes, and embodied intelligent systems through interactive learning. But the universal world model that can truly integrate all of the above capabilities is still in its infancy. World Labs, founded by Li Feifei, is trying to promote this cutting-edge research. The world model developed by their team can understand complex 3D scenes at the semantic and geometric levels, infer their physical properties and interaction relationships, and generate coherent and explorable virtual spaces, while also adhering to physical and geometric laws during interactions. The Marble platform developed by it has taken the lead in applying these capabilities, allowing creators to quickly create and edit complete virtual worlds without the need for traditional 3D modeling tools. This means that AI is moving from understanding language to understanding space, from generating images to generating the world. Reshaping how humans coexist with the world "has been nearly 500 million years since the first emergence of spatial intelligence in ancient animals in nature. Nowadays, we are fortunate to be the generation of technicians who have the potential to endow machines with the same capabilities. ”Li Feifei wrote in his article. She believes that spatial intelligence will redefine the functionality and behavior of AI. With the help of the world model, machines will shift from passive analysis to active planning and adaptation. For example, warehouse robots can bypass constantly changing inventory planning paths instead of stopping when blocked; Autonomous vehicle can predict the movement of pedestrians, rather than waiting for a clear signal; Even one day, digital assistants will be able to interpret gestures, spatial environments, or shared visual frameworks. Specifically, in the fields of creativity and narrative, spatial intelligence will redefine the way humans create. Directors and game designers are free to construct virtual worlds, architects can visualize structures and proportions in real-time, and individual creators and students can also use virtual reality and augmented reality technology to "create worlds" in three-dimensional space. In the field of robotics, spatial intelligence is the key to achieving embodied intelligence. It enables robots to efficiently train and understand spatial relationships and human intentions in virtual environments, enabling safe collaboration from laboratory assistants to home caregivers. In the future, from nanomedical robots to software detection robots, AI will possess environmental perception and adaptive capabilities. In science, healthcare, and education, space intelligence will become a multiplier for human exploration: AI can simulate experiments, predict results, accelerate drug development and imaging diagnosis; In the classroom, it allows students to "step into" molecules or historical sites, making abstract knowledge tangible and perceptible. Li Feifei emphasized that the goal of space intelligence is not to replace humans, but to enhance human professional judgment, creativity, and empathy, allowing technology to serve humanity more deeply. AI with spatial intelligence will help humans understand diseases, reshape narrative forms, accompany vulnerable groups, accelerate scientific discovery and social care. She believes that space intelligence will not only change what AI can do, but also reshape how humans coexist with the world. From language to space, from text to the world, AI is moving towards a new era where it can truly understand reality. (New Society)