In this episode of No Priors, Sarah Guo interviews Aditya Ramesh, Tim Brooks, and Bill Peebles, the team behind OpenAI's Sora, a new generative video model. The discussion covers Sora's potential as a pathway to AGI, its ability to model complex environments, and the challenges of achieving realistic video generation. The team shares insights on the model's architecture, including the use of diffusion transformers and latent spacetime patches, and its capacity for world modeling and simulation. They also address safety considerations, the future product roadmap, and the limitations of the current model, such as object permanence and the need for cheaper usage costs. The team expresses excitement about the creative potential of Sora and its broader implications for AI and visual intelligence.
Sign in to continue reading, translating and more.
Continue