Researchers at OpenAI have trained large-scale generative models on video data to create a minute-long, high-fidelity video. The model, called Sora, uses a transformer architecture that operates on spacetime patches of video and image latent codes. This approach allows the model to generate videos with variable durations, resolutions, and aspect ratios. --- Why it matters: This work matters because it shows promise in building general-purpose simulators of the physical world, which could have significant implications for fields like robotics, gaming, and virtual reality. Source: https://openai.com/index/video-generation-models-as-world-simulators