In this episode of the Latent Space Podcast, Alessio and Swyx interview Fei-Fei Li and Justin Johnson of World Labs about world models and spatial intelligence. They discuss how Fei-Fei and Justin came together to start World Labs, the evolution of AI since AlexNet, and the balance between open science and commercial pressures in AI research. They explore the role of academia in AI, wacky ideas for hardware and neural network architectures, and their past work on image captioning and dense captioning. The conversation further covers the differences between language and spatial intelligence, the challenges of embedding physics into world models, and the capabilities and use cases of Marble, World Labs' 3D world generation system. They also touch on the potential of Marble for robotic training and the broader applications of spatial intelligence in design and embodied AI. Finally, they discuss the future of modeling, the importance of multimodality, and the need for talent in the field.
Sign in to continue reading, translating and more.
Continue