Personally I’m not too excited by that future, if AI are made efficient enough to be contained on device & more trustworthy, I’m afraid of a surveillance state where the spying is not done remotely in large datacenters, but rather where data is processed by on-device agents and reported.
A black box in every device, maybe even in the form of hardware; kind of like the on-device scan Apple tried to impose a while back, but far reaching & that can be modulated & updated to be tailored to every region. One country using it to spy on dissidents, another for crime, another to repress LGBT people…
I don’t see AI as it exists as really useful beyond just being a “smarter google” that doesn’t (yet?) get all its results from SEO-optimized garbage. UI & UX is pushed so far nowadays that a “smart assistant” barely saves you a few taps on your smartphone, so I don’t think the smartest AI, whether LLM or otherwise, will be useful for everyday life as a tool for individual users.
AI integrated to robotics as a “servant” is one possible use, but the advancements there are so snail-paced that I don’t think it’s something that will exist even in ~50 years.