404 Media has uncovered leaked documents revealing a disclosure; Nvidia, the chip giant, in AI technology has been secretly collecting extensive YouTube video data to enhance its AI models. This controversial approach was carried out without the consent of video creators or Google the owner of YouTube and a significant customer of Nvidia. The magnitude of this data collection is staggering as internal communications indicate intentions to create a “video data factory of generating a days worth of training data to a human lifetime visual experience.”
Nvidias data gathering operation utilized multiple “virtual machines” that automatically changed their IP addresses to evade detection by YouTube. The gathered data was employed in training Nvidia AI models, including the Cosmos learning model, an algorithm for self driving cars an AI avatar product called “digital human ” and Omniverse, their 3D world building tool. Despite concerns raised about the legality and ethical implications of these actions management asserted that the initiative had received approval, from authorities.
Misuse of Academic Research Data
In an instance it came to light that Nvidia knowingly utilized the HD VG 130M dataset originally intended for research and comprising 130 million YouTube videos to train its models. This act of repurposing data meant for research purposes, for gain raises ethical concerns. Shayne Longpre, a PhD Candidate at the MIT Media Lab highlighted the divide between profiting from data without consent versus exploring AI capabilities using publicly available content.
Nvidias pivotal role in the AI sector has implications. Numerous AI firms such as OpenAI, Microsoft, Meta and Google depend on Nvidias GPUs to power their resource AI systems. YouTube CEO Neal Mohan has unequivocally stated that unauthorized use of YouTube data violates the platforms terms of service. Despite Nvidia asserting that its AI training practices adhere to copyright laws both, in letter and spirit questions persist regarding the legality and ethics of their actions.
