As Apple works to ready Apple Intelligence for a beta launch later this year, a new report claims that the company used YouTube videos as a source of data when training their AI models.
Apple is just one company thought to have used data collected by a third party when training AI, with Nvidia and Anthropic also among those thought to have used the same information. The dataset, called YouTube Subtitles, was collected by EleutherAI and created by taking the transcripts from videos created by some of the biggest names on the platform including MKBHD and MrBeast.
While the dataset was not created using the actual videos themselves but rather their transcripts, it's still thought that the act is against YouTube's terms of service.
It's all about the subtitles
The Wired report notes that the dataset is part of a compilation the outfit released called the Pile which is accessible and open to anyone on the internet.
An investigation found that subtitles from 173,536 YouTube videos across 48,000 channels were used to train data with Apple one of the companies that benefited. It's thought that Apple used the Pile, to train OpenELM, a model that was announced in April just weeks before Apple announced that Apple Intelligence would launch alongside iOS 18. The offering is made up of multiple new AI-powered features that generate text and images across multiple apps and services.
Understandably, YouTubers are less than happy with the news. “No one came to me and said, ‘We would like to use this,’” said David Pakman, the host of The David Pakman Show. Others suggested that the use of subtitle data in this way was theft, noting that the same technology could well be used to take creators' jobs in the future.
Apple Intelligence will launch later this year, albeit in beta, alongside iOS 18 and software updates for the Mac, iPad, Mac, Apple Watch, Apple TV, and Apple Vision Pro.
