yeah, strategy is everything. this is an arms race. we are building a reusable set of tools and systems that allow us to replace centralized ones and beat them on efficiency and cost. the LLMs will gradually become part of that. little mini-pc boxes are arriving this winter with 128gb that can run full reasoning/visual models. IMO the current projected power apocalypse would be fixed just by decentralizing AI server location. skynet also needs that centralized pattern but it's only a matter of time now before common domestic hardware levels the playing field between us and the big corporates as well. eventually scaling techniques will become better and we will just stack these things on our desk and have our own pocket einstein over ContextVM.