What might two dogs making radio programs look like? How about a bicycle race with different sea animals as competitors?
With Sora, you may soon be able to find out. The latest AI model can create videos up to a minute long from text prompts (提示词) in seconds. It was built by Open AI, the company behind the popular chatbot Chat GPT. Its CEO Sam Altman announced Sora's creation on February 16.
Sora hasn't been made available to the public yet. Open AI is asking experts to test it for possible risks. The model is also open to a small group of artists. The company hopes their advice will help make Sora better.
Before Sora, the leading text-to-video AI model was one made by Runway. Its system, Gen-2, came out in March 2023. The videos it produced were short and often scary. So most people were surprised by the high quality of Sora's videos.
At present, Sora model isn't perfect, however, as it may not understand cause and effect (因果). "For example, a person might take a bite out of a cookie, but after that, the cookie may not have a bite mark," Open AI says on its website. Sora may also mistake spatial (空间
) details, like mixing up left and right. In a video on the Open AI website, a man is seen running on a treadmill (跑步机) in the opposite direction. At the same time, some people have other worries. They fear that AI-generated content could be misused. The Global Risks Report 2024 listed AI-generated misinformation as the biggest risk facing the world this year. In response, Open AI said that it's also building tools to let users know if a video was produced by Sora. Whatever the worries, with the development of technology, Sora could go much farther than we think.
a. Sora may not understand cause and effect.
b. Sora might take a bite out of a cookie.
c. Sora can mistake spatial details.
d. Sora-generated content could be misused.
