[ti:OpenAI Announces New Video Generator]
[al:Science & Technology]
[ar:VOA]
[dt:2024-02-21]
[by:www.voase.cn]
[00:00.00]The maker of ChatGPT recently announced its next move into generative artificial intelligence.
[00:11.51]San Francisco-based OpenAI's new text-to-video generator, called Sora, is a tool that instantly makes short videos based on written commands, called prompts.
[00:27.30]Sora is not the first of its kind. 
[00:30.60]Google, Meta and Runway ML are among the other companies to have developed similar technology.
[00:40.03]But the high quality of videos displayed by OpenAI - some after CEO Sam Altman asked social media users to send in ideas for written prompts - surprised observers.
[00:57.01]At the same time, the video results led to fears about the possible ethical and societal effects.
[01:06.58]A photographer from New Hampshire posted one suggestion, or prompt, on X.
[01:14.29]The prompt gave details about a kind of food to be cooked, gnocchi, as well as the setting - an old Italian country kitchen.
[01:26.62]The prompt said:
[01:28.80]"A instructional cooking session for homemade gnocchi hosted by a grandmother social media influencer set in a rustic Tuscan country kitchen with cinematic lighting."
[01:42.49]Altman answered a short time later with a realistic video that showed what the prompt described.
[01:51.77]The tool is not yet publicly available. 
[01:56.44]OpenAI has given limited information about how it was built.
[02:02.37]The company also has not stated what imagery and video sources were used to train Sora.
[02:11.41]The New York Times and some writers have taken legal actions against OpenAI for its use of copyrighted works of writing to train ChatGPT. 
[02:25.32]And OpenAI pays a fee to The Associated Press, the source of this report, to license its text news archive.
[02:35.23]OpenAI said in a blog post that it is communicating with artists, policymakers and others before releasing the new tool to the public.
[02:48.24]The company added that it is working with "red teamers" - people who try to find problems and give helpful suggestions - to develop Sora.
[03:00.41]"We are working with red teamers - ... experts in areas like misinformation, hateful content, and bias - who will be adversarially testing the model," the company said.
[03:14.50]"We're also building tools to help detect misleading content such as a detection classifier that can tell when a video was generated by Sora."
[03:27.21]I'm John Russell.