Nvidia used copyrighted content like YouTube videos to train AI models, claiming it was in compliance with copyright law and had company clearance for content use.
Anonymously, a former Nvidia employee disclosed scraping Netflix and YouTube for AI model training, leading to projects like Cosmos for simulations and self-driving systems.
Emails revealed Cosmos project aimed at a video foundation model merging light transport, physics, intelligence, highlighting Nvidia's downstream applications.
Internal Slack messages showed Nvidia's use of open-source YouTube downloaders and virtual machines to scrape content while avoiding IP blocks.
Collection
[
|
...
]