What is HN's experience of running local LLMs inside an organization? What hardware? What model? How is it resourced? How is access and usage managed? What question should I be asking about it if you're doing it? What's interesting?