AI pioneer Fei-Fei Li argues that 'world models'—AI with a deep understanding of the 3D physical world—are the critical next frontier, addressing the limitations of current language-centric AI.
World Labs, founded by Li, is a company dedicated to building these foundational world models, combining top-tier talent in computer vision, AI, and computer graphics.
This technology aims to reconstruct and generate complete 3D environments from limited 2D inputs, unlocking horizontal applications in robotics, creative industries (gaming, film), design, and virtual world creation.
The speakers contend that spatial intelligence is a more fundamental aspect of cognition than language, and developing it in AI requires a concentrated, industry-grade effort similar to what propelled LLMs.
12 quotes
Concerns Raised
The immense technical difficulty of building AI that understands 3D space, as evidenced by the slow, expensive progress in autonomous driving.
The need for a unique and rare combination of talent spanning both AI/ML and computer graphics.
The significant compute and data resources required, necessitating an 'industry-grade effort' beyond academic capabilities.
Opportunities Identified
Creating a new foundational AI platform for robotics, embodied AI, and physical interaction.
Revolutionizing creative industries like design, architecture, gaming, and film by enabling the generation and manipulation of 3D worlds.
Unlocking the ability to create infinite, interactive digital universes or a 'multiverse' for socialization, training, and entertainment.
Developing a horizontal technology with applications as broad and impactful as LLMs.