Fei-Fei Li, a pioneer in computer vision, discusses her new company, World Labs, which is focused on developing foundation models for spatial intelligence—the ability to understand, reason, and generate 3D worlds.
She argues this is a critical, unsolved problem in AI, on par with language, and will unlock applications in creative industries, robotics, and the metaverse.
Li also reflects on her past work, including the creation of the ImageNet dataset which catalyzed the deep learning revolution, and emphasizes her philosophy of building human-centered AI to solve major societal challenges, particularly in healthcare.
12 quotes
Concerns Raised
The scarcity of high-quality 3D data is a major challenge for training world models, unlike the abundance of text for LLMs.
Productizing 3D AI is difficult because it's a less natural and active form factor for users compared to language.
Robotics is a complex systems integration problem where haptic data is underappreciated and difficult to integrate.
Opportunities Identified
Developing foundational models for spatial intelligence represents a major new frontier for AI.
AI-powered creative tools can superpower 3D artists, designers, and game developers.
AI can be a transformative tool in healthcare, from drug discovery to improving care delivery and aging.
Simulation and synthetic data are underrated but crucial for advancing robotics.