>>107122746
It still exists, for the simple stuff that can be done on device, in the background, with tiny models.
For everything good, it's unavoidable that Apple call a beefy server in some data center instead.
I don't think edge AI is going to work. Manufacturers, Apple in particular, give you way too little VRAM and disk space to run models that would do a good enough job to justify using them. And that's not even mentioning that AI coprocessors are literally hundreds of times weaker at AI compute than a 5090.
Apple initially claimed they were going to create something called private cloud compute (
https://security.apple.com/blog/private-cloud-compute/) where they were going to host their own hardware (mac pros and mac studios, and probably custom built Apple server hardware) and host their own models on it, but it looks like they didn't manage to make any remotely competitive model run on that architecture, which is why they're making deals with "Open"AI and Google instead now. Sad!
