I do find the text streaming aspect of LLM interfaces to be a charming solution to the latency issue. What's old is new again.