I'm skeptical of the on-device AI. They crave edge compute but I'm doubtful their chips can handle a 7B param model. Maybe ironically with Microsoft's phi 3 mini 4k you can run this stuff on a cpu but today it's no where near good enough.