On February 15, OpenAI launched a new generative AI model called “Sora” that can create videos of up to 1 minute from simple text prompts.announced. However, the release date for OpenAI’s new system, which is attracting a lot of attention, has not yet been decided.
According to OpenAI, the videos generated by Sora feature “extremely detailed scenes, complex camera motions, and multiple emotionally vivid characters,” including a stylish woman walking the streets of Tokyo and a fictional character. A series of movie trailers includingVideo generated by Sorawas tweeted.
OpenAI emphasizes that Sora’s safety measures must be completed before being made public, and AI models will be tested “adversarially” by experts in areas like disinformation, hateful content, and bias. He said that it has been done.
OpenAI also describes Sora’s weaknesses, saying it may not be able to understand physics simulation or cause-and-effect relationships in complex scenes.
OpenAI also notes that Sora may have trouble with spatial detail due to certain prompts. For example, examples are provided where the model confuses left and right or struggles with accurate camera trajectory.
OpenAI, which has been the subject of multiple copyright lawsuits, particularly from authors and the New York Times, said it is working with policymakers, educators and artists to understand their concerns.
OpenAI has not yet responded to Forbes’ inquiries about Sora’s general release date.
OpenAI has been active in developing the AI capabilities it offers, most recently announcing that ChatGPT is testing a new feature that remembers user voices and past chats. Users will be able to control what information chatbots can store about them.
OpenAI’s development and training of AI models, including Sora, has been at the center of multiple legal disputes involving the company. A U.S. District Court judge this week ruled on an unfair competition lawsuit filed by authors including Sarah Silverman and Christopher Golden, alleging that OpenAI did not have permission to use their copyrighted books for commercial purposes. The court ruled in favor of the claim. The authors claim there is evidence of “direct copying” of copyrighted books that OpenAI allegedly used to train its AI language models.
SEE
ALSO
Technology > AI
ChatGPT begins testing new feature that “remembers” users’ voice, tone, and preferences