OpenAI Teases New Tech Breakthrough
OpenAI has revealed a tantalizing glimpse of their latest innovation, Sora. However, the company has made it clear that they currently have no plans to release it to the public. Instead, OpenAI will be sharing the model with third-party safety testers for the first time.
The company is concerned about potential misuses of fake but photorealistic videos and wants to ensure they have all their bases covered before making Sora available to the general public. Aditya Ramesh, a scientist at OpenAI, and the creator of the firm’s text-to-image model DALL-E, emphasized the importance of careful deployment.
While Sora won’t be available to the public anytime soon, OpenAI is already looking ahead to a future product launch. They are sharing the model with a select group of video makers and artists to gather feedback and make Sora as useful as possible to creative professionals. Ramesh also stated that the goal is to provide a preview of what these models will be capable of in the future.
The team behind Sora adapted the technology used in DALL-E 3, OpenAI’s latest text-to-image model, to create the video-based model. Sora combines the diffusion model with a transformer, a type of neural network, making it unique in its approach to processing video data.
Transformers are known for their ability to process and understand long sequences of data, making them a valuable addition to Sora’s design. The researchers developed a method to cut videos into chunks that could be processed by the transformer, allowing Sora to be trained on a wide range of video types.
This innovative approach is seen as a significant leap forward from a technical perspective. Sam Gregory, executive director at Witness, a human rights organization specializing in the use and misuse of video technology, acknowledged the expressive capabilities but also highlighted the potential for misuse. The advancements offer the potential for more people to tell stories using video, but there are also concerns about unethical use.
As we look to the future, Sora has the potential to revolutionize the world of video creation and storytelling. OpenAI’s cautious approach to sharing the model with third-party safety testers demonstrates their commitment to responsible deployment and their proactive efforts to address potential issues before making Sora available to the wider public.
Source link