What I'm saying is that not all local LLMs are more power efficient, and that we are currently massively over-using the technology for things it isn't all that useful for.