Netflix and YouTube say Nvidia cannot collect data in this way.
Nvidia illegally downloads millions of videos to train its neural networks, report sources 404 Media. According to the publication, Nvidia collects videos from YouTube, Netflix and other resources.
In particular, the publication writes, Nvidia employees collected special lists of YouTube channels, videos from which should be taken for training neural networks. Nvidia top managers allegedly sanctioned the collection of data for training by almost any means.
To download videos from YouTube, Nvidia employees may have used several dozen Amazon Web Services virtual machines: their use was discussed in a Slack chat, the messages of which were accessed by journalists. It can be assumed that other videos were collected in a similar way.
According to the publication’s calculations, Nvidia could upload enough videos every day to watch them all in 80 years. By the end of May 2024, the company had allegedly uploaded about 38 million videos, and the data collection process only began at the end of April of that year.
The collected videos are supposed to be used to train computer vision systems and neural networks that generate videos based on descriptions. Nvidia is actively working on improving computer vision for car control systems.
In a comment to Engadget, Nvidia representatives reportedthat its actions “comply with the letter and spirit of copyright law.” YouTube and Netflix believe that the company had no right to collect their content to train its services.
