Just recently, WIRED and Proof News carried out an investigation alleging that Apple and other tech behemoths including NVIDIA and Salesforce, had used YouTube subtitles to train their AI systems. Apple and others relied on a non-profit called EleutherAI to provide a dataset that allegedly was trained on more than 170,000 videos from popular creators and media houses.
Now, Apple has confirmed to both 9to5Mac and AppleInsider that the trained dataset using YouTube subtitles wasn’t used in any of its AI applications, including the recently unveiled Apple Intelligence which shapes its upcoming iOS 18, iPadOS 18, and macOS Sequoia.
Apple created the OpenELM model to help researchers and improve open-source large language models. The company has called it a “state-of-the-art open language model”, but wants to make clear it’s just for research and isn’t powering any Apple Intelligence features. It’s freely available on Apple’s Machine Learning Research website for anyone interested in using it.
In fact, Apple has been very ambitious with its AI plans. Last year, it approached several major publishers to negotiate a deal that could help the company train its AI dataset.
As Apple has made it clear that it has no further intentions to build any new versions of the controversial OpenELM model, it’s certain that there won’t be unauthorized use of data without the creator’s knowledge or consent. Meanwhile, we’re keeping an eye on further developments and will keep you posted with whatever happens next. So, stay tuned.