

That’s why it’s on the OS-level. For example, for text, it seems to work in any text app that uses the standard text input api, which Apple controls.
User activates the “AI overlay” on the OS, not in the app, OS reads selected text from App and sends text suggestions back.
The App is (possibly) unaware that AI has been used / activated, and has not received any user information.
Of course, if you don’t trust the OS, don’t use this. And I’m 100% speculating here based on what we saw for the macOS demo.
It goes a tad bit beyond classical conditioning… LLM’a provides a much better semantic experience than any previous technology, and is great for relating input to meaningful content. Think of it as an improved search engine that gives you more relevant info / actions / tool-suggestions etc based on where and how you are using it.
Here’s a great article that gives some insight into the knowledge features embedded into a larger model: https://transformer-circuits.pub/2024/scaling-monosemanticity/