During yesterdays WWDC keynote Apple announced the creation of their new Foundation Models framework. Using this framework developers will be able to access Apples on-device AI models.

This has two main benefits. Firstly it means that the models can work while the device is offline and secondly it means that developers won’t incur API costs. This second point didn’t get a lot of attention but is a big deal I think.

Also of interest are the paragraphs on how Apple trains it’s models and acquires it’s data.

We believe in training our models using diverse and high-quality data. This includes data that we’ve licensed from publishers, curated from publicly available or open-sourced datasets, and publicly available information crawled by our web-crawler, Applebot. We do not use our users’ private personal data or user interactions when training our foundation models. Additionally, we take steps to apply filters to remove certain categories of personally identifiable information and to exclude profanity and unsafe material.

Further, we continue to follow best practices for ethical web crawling, including following widely-adopted robots.txt protocols to allow web publishers to opt out of their content being used to train Apple’s generative foundation models.

There is a fairly stark contrast between this and OpenAI’s approach.

I’d love to see more attention and discussion around the ethical training of AI.