Video generation models as world simulators

OpenAI introduces Sora, a large-scale text-conditional diffusion model capable of generating up to one minute of high-fidelity video content. The model uses transformer architecture on spacetime patches and represents a significant advancement toward building general purpose physical world simulators.