Apple Clarifies Research Use of OpenELM Model Amid YouTube Data Controversy

Apple has revealed to 9to5mac that its OpenELM model does not play a role in any of its AI or machine learning applications, such as Apple Intelligence. This statement follows reports that Apple, along with other tech giants, utilized subtitles from 170,000 YouTube videos to train their AI frameworks. Notable content creators like MKBHD and Mr. Beast were part of this dataset. Purpose of OpenELM Model Apple explains that the OpenELM model is solely for research and contributing to the open-source community. Introduced in April, Apple described OpenELM as an advanced open language model. The company insists it was designed for research in large language models and is not integrated into any of Apple’s proprietary AI functionalities. The technology firm has reiterated that Apple Intelligence models are built using licensed and publicly accessible data obtained through its web-crawler. Importantly, the “YouTube Subtitles” dataset, part of “The Pile” from EleutherAI, is not used in Apple Intelligence. Apple also indicated there are no plans to create new versions of the OpenELM model. Industry-Wide Practices This controversy sheds light on the broader methods adopted across the industry. Companies like Anthropic and NVIDIA have also been reported to use the same YouTube subtitles dataset for their AI training purposes.  By clarifying the role of OpenELM, Apple aims to address ethical considerations around data usage in AI training. They make it clear that OpenELM is a tool intended purely for research, thus distancing their commercial AI features from controversy over using YouTube subtitles for model training. The OpenELM model remains accessible to researchers on Apple’s Machine Learning Research site.