According to the video sharing platform, training AI with the videos breaks its terms of service
Using YouTube videos to train Sora, OpenAI’s artificial intelligence text-to-video generator, would break the terms of service of the platform, according to Neal Mohan, YouTube’s CEO.
This represents the first time the CEO has spoken on the subject
Mohan used his first public comment on the subject to add that he did not have any firsthand knowledge of whether OpenAI had indeed trained its Sora artificial intelligence powered video creator with videos from the YouTube platform. That said, he also added that if it had been the case, it would have been a “clear violation” of the terms of use of the video platform.
“From a creator’s perspective, when a creator uploads their hard work to our platform, they have certain expectations,” explained Mohan last week when interviewed by Bloomberg Originals host Emily Chang.
He explained that among those expectations is that there will be adherence to the terms of service. Those terms do not permit downloading of transcripts or pieces of videos, and the terms of service make that deliberately clear. “Those are the rules of the road in terms of content on our platform.”
Training OpenAI artificial intelligence models has lead to substantial public debate
There has been notable debate regarding the way ChatGPT, DALL-E and other AI products are being trained and what material is being used. Generative AI tools such as Sora function by learning off as much content as they can from all around the web, and then using the data to which they were exposed as the base of the tools used for the creation of new content, such as narrative text, photos, videos, and more.
OpenAI isn’t alone in this. It’s the way the technology functions. Other players in AI, such as Google, must also provide their products with massive amounts of data in order to make them useful. The more data they can use, the better the quality of the results they can provide.
OpenAI Chief Tehcnology Officer Mira Murati was quoted in an article in the Wall Street Journal in March, saying that she didn’t know if Sora was trained with user-generated videos from platforms such as Facebook, Instagram, and YouTube. Microsoft – the tech giant backing that artificial intelligence company – had not issued a response on the subject since the Mohan interview.
