OpenAI and Google reportedly used transcriptions of YouTube videos to train their AI models
According to The New York Times, the companies may have violated YouTube creators’ copyrights.
OpenAI and Google trained their AI models on text transcribed from YouTube videos, potentially violating creators’ copyrights, according to The New York Times. The report, which describes the lengths OpenAI, Google and Meta have gone to in order to maximize the amount of data they can feed to their AIs, cites numerous people with knowledge of the companies’ practices. It comes just days after YouTube CEO Neal Mohan said in an interview with Bloomberg Originals that OpenAI’s alleged use of YouTube videos to train its new text-to-video generator, Sora, would go against the platform’s policies.
According to the NYT, OpenAI used its Whisper speech recognition tool to transcribe more than one million hours of YouTube videos, which were then used to train GPT-4. The Information previously reported that OpenAI had used YouTube videos and podcasts to train the two AI systems. OpenAI president Greg Brockman was reportedly among the people on this team. Per Google’s rules, “unauthorized scraping or downloading of YouTube content” is not allowed, Matt Bryant, a spokesperson for Google, told NYT, also saying that the company was unaware of any such use by OpenAI.
The report, however, claims there were people at Google who knew but did not take action against OpenAI because Google was using YouTube videos to train its own AI models. Google told NYT it only does so with videos from creators who have agreed to this. Engadget has reached out to Google and OpenAI for comment.
The NYT report also claims Google asked a team to tweak its privacy policy in June 2023 to more broadly cover its use of publicly available content, including Google Docs and Google Sheets, to train its AI models and products. The changes, which Google says were made for clarity's sake, were published in July. Bryant told NYT that this type of data is only used with the permission of users who opt into Google’s experimental features tests, and that the company “did not start training on additional types of data based on this language change.” The change added Bard as an example of what that data might be used for.
Correction, April 6, 2024, 3:45PM ET: This story originally stated that Google updated its privacy policy in June 2022. The policy update was actually made in 2023. We apologize for the error.