Something that needs to be considered before answering that question is that current predictive text engines are ridiculously stupid compared to what an LLM (or even an "SLM") with access to all of your previous texts could do.
When somebody finally gets a clue and implements that, no typist on Earth will be able to keep up with it.
When somebody finally gets a clue and implements that, no typist on Earth will be able to keep up with it.