34:05How To Build Generative AI Models Like OpenAI's Sora
The talk explains how Sora combines transformers, diffusion, and spatiotemporal patches to generate video with consistent motion and scenes. It surveys Sora’s strengths and flaws, then argues that smaller teams can build useful foundation models by focusing on high-quality data, efficient compute, and task-specific expertise. Examples from YC companies cover video, lip-sync, music, hardware design, software, biology, weather, brain signals, robotics, and CAD. The speakers also discuss synthetic and simulated data, and how founders can compete by specializing rather than matching the largest labs’ resources.