OpenAI, the creators of ChatGPT and DALL-E, have introduced Sora, a text-to-video diffusion model, marking a significant stride in AI development. Currently available to red teamers for adversarial testing and a select group of visual artists, designers, and filmmakers for feedback, Sora enables the generation of one-minute-long videos with detailed scenes and multiple characters.
While other video-generating models exist, Sora is reputed for producing realistic and complex videos. Meta has a tool for short video clips, and Google is researching its text-to-video model, but OpenAI's Sora stands out in terms of its capabilities.
Safety measures include red-teaming, labeling Sora-created videos per C2PA guidelines, and applying existing safety methods from DALL-E to reject inappropriate prompts. OpenAI emphasizes engagement with policymakers, educators, and artists globally to understand concerns and identify positive use cases for this technology, highlighting the importance of real-world learning in enhancing AI system safety over time.
0 Comments