Dr. Li posits that AI is incomplete without spatial intelligence—the ability to understand and generate 3D worlds. Her new company, World Labs, is building foundation models to tackle this problem, which she views as being as critical and difficult as language.
Reflecting on her creation of the ImageNet dataset, Dr. Li underscores how a massive, well-structured dataset was the catalyst for the deep learning revolution. She contrasts this with the current data scarcity for 3D models, identifying it as the primary bottleneck for progress in spatial AI.
The discussion explores the future of robotics, arguing for a diversity of energy-efficient, task-specific form factors over a single humanoid model. Dr. Li also emphasizes the underrated importance of simulation and haptic data for training robots, especially for complex manipulation tasks.
Dr. Li advocates for developing AI as a tool to collaborate with and empower humans, particularly in critical areas like healthcare. She co-founded the Stanford Institute for Human-Centered AI to promote this vision of using technology to solve major societal problems while upholding human values.
Keep pulling the thread on Fei-Fei Li.