Skip to content
Although there isn’t much clear proof of this, several AI experts, including some people from Alphabet, believe that Sora’s OpenAI’s AI video generator was trained on data scrubbed from video platforms and social media platforms like YouTube and Instagram.
If OpenAI used this data to train Sora, it would not get any clearances or licenses from the platforms, which may cause trouble for Sam Altman’s AI venture.
According to YouTube CEO Neal Mohan, using YouTube videos to train OpenAI’s text-to-video generator would be a major violation of the platform’s terms of service.
Mohan stated that creators expect their content to be protected when uploaded to YouTube. One such expectation is that people on the platform will adhere to the platform’s regulations. They also expect that their content will not be reused by anyone for any commercial purpose without their explicit approval.
In his first public remarks, Mohan clarified that he needed more direct knowledge of whether OpenAI had used YouTube videos to refine Sora. However, he emphasized that such actions would breach YouTube’s terms of service.
The debate over the materials used to train AI models, including those behind popular content creation tools like ChatGPT and DALL-E, has sparked widespread discussion. Generative AI tools like Sora rely on scrubbing the internet to improve their capabilities for generating videos, photos, or even text.
While companies such as OpenAI and Google compete to develop more advanced AI models and be the first to reach AGI, they need extensive pools of content or data to train their models effectively.
In a recent interview, OpenAI’s Chief Technology Officer, Mira Murati, was put on the spot when she was asked if OpenAI scrubbed social media platforms and platforms like YouTube for video data to train Sora. Murati was unable to answer the question and claimed she was uncertain.
Reports indicate that OpenAI has explored training its upcoming large language model, GPT-5, using transcriptions of public YouTube videos. Mohan noted that Google, YouTube’s parent company, respects individual contracts with creators before considering using YouTube videos to train its own AI model, Gemini.
Mohan also highlighted the importance of ensuring that any use of YouTube videos aligns with the creators’ licensing agreements and terms of service. Although some of YouTube’s content may be utilized for training AI models like Gemini, Google, and YouTube, they prioritize compliance with creators’ contracts and terms of service.
