I think the main issue is that you can't fully control what an llm will output. If Apple products output wrong/silly/dangerous content it is in the news and big

I think the main issue is that you can't fully control what an llm will output. If Apple products output wrong/silly/dangerous content it is in the news and big reputation hit for them, whereas for a small ai startup people don't really care. Apple likes to have full control over the user experience.