Another big scandal involving an artificial intelligence (AI) company emerged today and this time it involved NVIDIA. According to a 404 Media report that accessed internal documents, NVIDIA stole 80 years worth of video a day from YouTube, Netflix and various other sources to train their new, yet-to-be-launched Cosmos AI model.
Cosmos was developed so that it can then be used to generate 3D models in NVIDIA Omniverse, teach self-driving systems and also for systems to generate digital humans. Among the reported allegations is that NVIDIA downloads YouTube videos and Netflix using up to 30 virtual machines with frequently changed IP addresses to prevent access being blocked.
The issue of data theft to train AI models is an issue that has gained increasing attention in the last two years. Just a few weeks ago the EleutherAI and Runway AI models were also reported to be trained using YouTube videos without permission.
Google has yet to file a lawsuit against the company that stole the data. At the same time Google itself received criticism for using YouTube data to train Veo, a generative AI model that can generate videos.