[jeromeof]

I honestly don't know why Apple can't do their own ChatGPT. They have the money / resources to then implement something like the ChatGPT 'plugin' architecture so if someone wants to turn on their lights or get specific 'private' information that apple cannot give as 'training' data. i.e. callout to 'private' API's if appropriate to a sensible LLM generated response

[generalizations]

I'm expecting that they will - and considering the raw power of their local hardware, I think they have the best shot of anyone at cornering the AI assistant market.

- private, on-device language model execution (see llama.cpp for feasibility)

- a single, consistent AI available with you wherever you go

- total access to your personal information / documents (knows your birthday, can see your meeting notes)

Because they have the hardware to run it locally, they have three very hard-to-beat advantages:

1) privacy, because the LMs can see all your stuff but none of it goes back to apple. Microsoft can't do this; they get flak every time they try and phone home with telemetry, and they don't control their platform enough to run a massive LM in the background.

2) omnipresence: if you're in the apple ecosystem, you'll always have your iphone with you. That means the LM will have access to location data, maps, chat - everything. And since it never leaves the phone, privacy-oriented people may be ok with it. And that means the LM can be exponentially more useful than just summarizing documents.

3) evaluation costs - they are the only competitor who will not have to pay for a massive datacenter, which means that the LMs can be as powerful as the M2+ hardware they sell. Everyone else will have no alternative to running the LMs on their centralized, expensive hardware.