Or perhaps “chat” just isn’t the best use of LLMs. Autocomplete-based integrations, rather than chat, could be better. I know I love and get great utility from GitHub Copilot, but not really any online or offline LLM chatbot.

Also, from a UX perspective, I feel an autocomplete-based UI encourages me to make sure the thing “I’m saying” is actually what I mean to say - ie, to not implicitly trust LLM hallucinations.

LLM-based generative text everywhere, in every app, could be useful, very straightforwardly integrated at an OS level into all text controls, something MS is well-poised to do, and “Copilot” a very good brand for that. (And I would be shocked if Apple and Google aren’t working on the same.)