Apple Affirms YouTube's Use Of Training Only For Open Source Projects

Illustration of Apple Intelligence. (photo: x @Fernando5x1)

JAKARTA - Earlier this week, an investigation revealed that Apple, along with other major technology companies, had used YouTube subtitles to train their AI models. These datasets cover more than 170,000 videos of popular creators such as MKBHD and Mr. Beast. Apple uses this data to train their OpenELM open-source model, which was released in April.

However, Apple has confirmed to 9to5Mac that OpenELM is not used to power their AI or machine learning features, including Apple Intelligence. The company states that OpenELM is being developed to contribute to the research community and advance the development of major open-source language models. Apple has described OpenELM as a "exemplary open-source language model" created solely for research purposes.

Apple's clarification shows that YouTube's mitigating dataset is not used to power Apple Intelligence features. Instead, Apple Intelligence models are trained on licensed data and available public data compiled by Apple's web-crawlers.

Additionally, Apple mentions that it has no plans to develop a new version of the OpenELM model. According to Wired, companies like Apple, Anthropic, and NVIDIA use YouTube'sdiadatasets, which are part of a larger collection called "The Pile" from the EleutheRAI nonprofit organization, to train their AI model.