Apple confirms that Apple Intelligence has not been trained with YouTube videos

With the presentation of Apple Intelligence during the last WWDC 2024, Apple claimed that Apple Intelligence models are trained with licensed data, including data selected to improve specific features, as well as publicly available data collected by your web crawler. This claim was questioned earlier this week, and now Apple is providing a defense.

iOS 18 Download

OpenELM is not part of Apple Intelligence, it’s just an open-source product for research

A few days ago, an investigation revealed that certain companies allegedly used YouTube subtitles to train their artificial-intelligence models. Not directly, but through a dataset compiled by a non-profit organization called EleutherAI. With or without knowledge of the fact, the report concluded that, therefore, Anthropic, Nvidia, SalesForce, Apple, and others had used the content of over 170,000 videos by popular creators such as MKBHD and Mr. Beast, generating considerable controversy regarding the ethics and legality of their methods. However, Apple has made an important clarification regarding the use of this data.

Apple confirmed to 9to5Mac that its OpenELM model, although trained with this data, is not used to power any of the features of its artificial-intelligence suite, known as Apple Intelligence. According to the company, OpenELM was created solely for research purposes, to contribute to the scientific community and the development of open-source language models.

The OpenELM model was published as open-source and is still widely available, including on Apple’s Machine Learning research website. This allows researchers from around the world to access and use this model in their own research projects.

Alongside this statement, Apple confirmed that they have no plans to develop new versions of the OpenELM model. As far as we know, this model has already fulfilled its purpose and will become less relevant as the rest of Apple Intelligence products evolve without it.

iOS 18 Download

The whole event is undoubtedly a reminder of the complexity and ethical challenges that companies face in the era of big data and AI. The reliability and performance of an artificial intelligence depend to a large extent on the dataset used for its training. A careful and measured selection, as we see Apple doing, is definitely the way to go to achieve product efficiency and quality.