FILE – The YouTube app is displayed on an iPad in Baltimore after YouTube blocked access to a video of a protest song in Hong Kong, days after a court approved an injunction banning the song in the city. (AP Photo/Patrick Semansky, File)
Tech giants Apple, Anthropic, Nvidia, and Salesforce have stolen data from tens of thousands of content creators on YouTube to train AI.
A Proof News investigation published in partnership with Wired found that Silicon Valley notables had provided an AI program with subtitles for 173,536 YouTube videos across more than 48,000 channels.
The “YouTube Subtitles” dataset was allegedly stolen from YouTube and its users without permission, a violation of YouTube's guidelines.
“Nobody came to us and said, 'I want to use this,'” David Pakman, host of “The David Pakman Show,” told Proof News.
“This is my livelihood and I put time, resources, money and staff time into creating this content,” Pakman added. “There's no shortage of work.”
“Apple sources data for its AI from multiple companies,” said Marques Brownlee, who goes by the handle MKBHD. Shared on X. “One of them scraped a ton of data and transcripts from YouTube videos, including mine. Apple didn't do the scraping, so technically Apple avoids 'fault' here. But this is going to be a long-term, evolving issue.”
YouTubers weren't the only content used in the dataset: the report also included online learning resources like videos from Khan Academy, Harvard University and MIT, as well as transcripts from news outlets like NPR, BBC, “Late Show with Stephen Colbert” and “Last Week Tonight with John Oliver.”
According to research published by EleutherAI, the “YouTube Captions” dataset is part of a large-scale language modeling collection called Pile that is being used by large companies to train AI.
Click here to access a tool that shows you which YouTube channels and videos were siphoned off into the dataset.
