Yes I get that. But just speed the god damn thing up.
The god damn thing has to do 175B multiply and add operations for each one of your words.
Who cares? OP was asking about UX improvements. Maybe I missed the part, but IMHO no one was asking only for UX improvements that come cheap.
They said "just speed the god damn thing up" - this is related to model inference speed. If your answer is one page long (about 2000 tokens), be prepared to wait for a whole minute.
I’d much prefer the type be hidden until it’s ready entirely.