Li Fei-Fei’s World Labs unveiled Atlas, a new multi-modal world model featuring novel view prediction. Unlike traditional LLMs or video models, Atlas generates content at any spatial-temporal position from minimal inputs, integrating generation and reconstruction. It drastically reduces 3D reconstruction data needs, enabling complex scene creation from just a few photos. Applications span creative design, architecture, and robotics, improving simulation efficiency and spatial understanding. This breakthrough positions new view prediction as a fundamental primitive for AGI, offering high 3D consistency and scalable spatial context for broader industry use.