Publicly available court documents or credible news reports

Nvidia Scrapes 80 Years of Videos Daily for Unreleased AI Model

Aug 5, 2024, 04:16 PM

Nvidia has reportedly used sources such as Netflix and YouTube to train an unreleased foundational AI model, according to leaked documents and internal communications obtained by 404 Media. The documents reveal that Nvidia scraped a massive amount of video content, described as 'a human lifetime' or 80 years of videos per day, with clearance from the highest levels of the company. Despite concerns from some staff about the legal and ethical implications, Nvidia proceeded with the data scraping. The leaks include emails, Slack chats, and other documents that detail the scale and scope of the operation.