I'm looking forward to an OpenAI quality embedding that I can use locally. I think eliminating the latency/cost from calling their API will allow for some novel UX patterns.