Last January, famous AI researcher Fei-Fei LI took a leave of absence from Stanford to conduct preliminary studies. Almost two years later, the venture World Labs unveiled its first commercial product: the world model Marble. Marble can create 3D Virtual worlds from text, images, video or complex structures. It builds on the previous World Labs prototype that creates 3D scenes from 2D images, but with limitations, such as restricted areas.
So-called marble-like world models are the basis of LI’s vision for the future of AI because these models can communicate with each other and interact with complex environments, but the world itself. World Labs aims to imbue its systems with spatial intelligence, teaching them visual concepts for people to handle appropriately, such as parking a car without folding an object, or pouring a drink without looking.
“Today, leading AI technologies such as large-scale linguistic models (LLMS) have begun to revolutionize the way we acquire and work with abstract information,” said Li 10 Blog Post. “Nevertheless they remain in words in the dark; they are good but they are ignorant, they have knowledge but they protect it.”
The emphasis on Visual and Spain Intelligence creates a “North Star,” said the researcher, who played a role in the release of Imagenet, a database of 15 million images that stimulated the rise of deep learning. LI also directs Stanford’s Institute for Human-Centered AI and serves as an advisor to the United Nations on AI policy.
These days, however, Li is focused on World Labs, which has raised $230 million to pursue its intellectual vision. Its backers include the fine details, Andreessen Horowitz and Nvidia, as well as prominent technical figures such as Geoffrey Hinton, Eric Schmidt, Marc Benioff and resume Hoffman.
Marble has been in beta for several months and is now publicly available. It can create a full 3D world from a single photo or quick text. Users can combine multiple scenarios by uploading several images within a short period of time. According to the world labs, the model can combine images or short videos of real world spaces to produce real world spaces.
The model includes various editing tools that allow users to implement their creations. A feature called chisel allows users to draw 3D structures, while other tools make it possible to expand worlds or create entirely new scenes in the same space. Looking ahead, global labs plan to develop global models with many functional capabilities for both AI agents and Agents ai.
While Li may be the most prominent figure developing global models, it is not the only field. Google Depmind and Nvidia have tested similar technologies with their genes and cosmos, respectively. Yann Lecun, Meta’s CEO Ai Scientist, is reportedly in the early stages of fundraising for his global startup model.
Li said the use of spatial intelligence tools such as marbles “can travel at different times.” The model is already used by filmmakers, game designers and building designers to improve the creative workflow. In the medium term, Li expects such technology to advance robotics, while future applications in science, health care, and education may allow successful experimental masking, drug discovery, and deep learning.
“Spatial Intelligence will revolutionize the way we create and interact with real and virtual worlds – transforming stories, intelligence, robotics, scientific discovery, and beyond,” Li said. “This is the next frontier of AI.”
