I think it is more fundamental that models are trained using existing data. An autosuggestion will suggest things people have written before. Providing a larger number of predictions doesn't change that this is inherently conservative.
Re: "I think it is more fundamental that models are trained using existing data."
Yes, I agree, but this does not have the be the final word.
There is a lot of room for design. For example, an autosuggestion tool can trade off between exploration (randomized recommendations) and exploitation (pure collaborative filtering).
