“The data is mostly from YouTube, downloaded via yt-dlp using a rotating set of AWS IP addresses and consisting of millions (maybe even hundreds of millions) of videos.”
Nvidia’s Cosmos team pulled YouTube video with yt-dlp through rotating AWS IP addresses, per 404 Media’s leak. Rotating IPs exist to dodge blocks, so nobody there thought this was allowed. The wish list named cinematic and drone footage, movies, and MKBHD’s tech reviews. The shopping list was curated. The licenses were not.