Eventually, when local compute gets good enough, running LLMs on this kind of system will be far more useful than most internet resources could be directly.