I agree. An AI assistant with the following properties would make me upgrade my phone: 1) fully local / ultra low latency; 2) privacy guarantees; 3) access / training / fine-tuning over all the data on my phone.
But it would probably require at least a 10x increase in on-device compute/memory resources to achieve that, unless they can pull off an efficiency miracle (1.58bit models?).
Apple has included NPU hardware in its SoCs since the A11 in 2017. It has been used for a variety of local AI in your pocket since then. This "Neural Engine" has only gotten more powerful with each iteration of their SoC in the intervening years.
I guess I disagree. Custom silicon, running a local AI (no network or network agency) in your pocket would be a game changer. Apple has to move fast to catch up on the software, but they own the hardware already, can design the AI-forward silicon.