Nvidia downloaded videos en masse from YouTube and other platforms to train its AI models, ignoring copyright issues. The company justifies using other people’s videos to train AI by citing the “spirit of copyright law,” but the platforms disagree.
Image source: convertcrypto.ru
According to internal documents and correspondence obtained by 404 Media, Nvidia collected videos “as long as a person’s life” from various Internet resources. According to a former Nvidia employee who wished to remain anonymous, workers were tasked with downloading videos from Netflix and YouTube and other platforms to train an AI model designed to generate the three-dimensional Omniverse world, self-driving car systems and future “digital human” products. The project is codenamed Cosmos.
The company’s employees used free and open-source software yt-dlp and virtual machines that update IP addresses to avoid blocks from YouTube to bulk download content. The goal was to obtain information equivalent to “80 years of life” every day.
Emails reviewed by 404 Media show that project managers discussed using 20-30 virtual machines in Amazon Web Services to upload such a huge volume of video footage per day. When employees raised questions about the legality of using copyrighted content, management assured them that the decision had been approved at the highest level of the company.
In a message to 404 Media, an Nvidia spokesperson said the company respects the rights of all content creators and believes its models and research efforts “fully comply with the letter and spirit of copyright law.”
«Copyright law protects the end product, but not the facts, ideas, data or information. Anyone can learn facts, ideas, data or information from another source and use them to create their own product. Fair use also protects the ability to use a work for a transformative purpose, such as training an AI model,” the company says.
However, YouTube and Netflix emphasized that such use of their content violates the platforms’ terms of service. A Netflix representative also told 404 Media that they do not have an agreement with Nvidia to provide content, and the platform’s terms of use do not allow web scraping, which is used by server scripts to automatically perform downloads.