Apple and other tech giants have been using YouTube subtitles to train their AI models, including over 170,000 videos from popular content creators. Apple utilized this dataset to train its OpenELM models, which were released back in April. Despite this, Apple has confirmed that OpenELM does not power any of its AI or machine learning features, including Apple Intelligence.
The purpose of creating the OpenELM model was for Apple to contribute to the research community and advance open-source large language model development. This model was specifically designed for research purposes and is not used to power any of Apple’s AI features. Although the model was published as open-source and is widely available on Apple’s Machine Learning Research website, it is not integrated into Apple Intelligence, meaning it does not rely on the “YouTube Subtitles” dataset.
Apple also stated that they have no intentions of building new versions of the OpenELM model in the future. Companies like Apple, Anthropic, and NVIDIA have all utilized the “YouTube Subtitles” dataset to train their AI models, which is part of a larger collection called “The Pile” from the non-profit EleutherAI. This news comes in light of recent discussions about data collection and model training in the tech industry.