I think what you are asking is how long until researchers find out a way of expanding the input size of the model? Because there ways you can get it write a book right now. For example, you can tell it to write the first page, then a summary of it. Then feed the summary and let it write another page. And so on.

As to increasing the input size, if no new way comes around, it will get bigger as the transformers get more efficient, which they will. Look up flash transformers for example.