Isn't one of the issues with local LLM the huge amount of GPU memory needed? I'd say that we'll go a lot longer than 5 years before phones have 24+GB of VRAM.