Nvidia
harmonize toa damning story from 404 Media , back with internal Slack New World chat , emails , and documents obtained by the issue , Nvidia helped itself to “ a human lifetime ocular experience worth of preparation data per Clarence Shepard Day Jr. , ” Ming - Yu Liu , frailty president of Research at Nvidia and a Cosmos labor drawing card , admit in a May email .
Unnamed former Nvidia employees told 404 that they had been ask to kowtow picture content from Netflix , YouTube , and other online rootage so as to receive training datum for use with the party ’s various AI products . Those let in Nvidia ’s Omniverse 3D world source , self - drive railcar systems , and “ digital human . ”
Nvidia
When those employee call for about the legality of the project , internally diagnose Cosmos , they were assured by direction that they had been given clearance by the in high spirits levels of the ship’s company to apply that subject matter .
The project sought to build a foundation modelling , akin toGemini 1.5,GPT-4 , orLlama 3.1 , “ that encapsulate pretending of light transport , physical science , and intelligence in one place to unlock various downstream applications critical to Nvidia . ”
To do this , project Cosmos allegedly used an open - source television downloader and employed motorcar learning to IP hops , thereby avoid YouTube ’s attempts to obstruct it . According to emails viewed by 404 , project manager discuss using as many as 30 virtual machine running on Amazon Web Services to download 80 year ’ worth of full - length and clip - duration videos every twenty-four hour period .
For its part , Nvidia exact no wrongdoing . “ We respect the rights of all content Divine and are convinced that our models and our inquiry attempt are in full compliance with the varsity letter and the spirit of copyright law , ” an Nvidia interpreter told 404 Media via e-mail . “ Copyright law protects exceptional reflexion but not facts , ideas , data point , or information . Anyone is free to watch facts , ideas , data , or information from another source and use it to make their own expressions . Fair use also protects the ability to habituate a workplace for a transformative purpose , such as model breeding . ”
This is far from the first clip that Nvidia ( not to cite a vast majority of the rest of the AI landing field ) has take a “ scratch first and maybe ask forgiveness later ” feeler to its AI breeding efforts . In July , Nvidia wasnamed in another reporton illegal scratching of copyrighted TV alongside Anthropic and Salesforce .
At CES 2024,the company set off an cyberspace firestormwith its equivocal answer as tohow its newfangled generative AI for play locomotive was civilize . In reaction , Nvidia reiterated that its tools were “ commercially secure . ”
