My thinking is writing articles would be easier if there’s someone asking thought provoking questions along the way.
For an article longer than the token window allows, you could use a script to do the same as above, chunk by chunk, but after the first completion, ask GPT-3 to refine the original list of questions.
Isn't this what OpenAI are doing with ChatGPT tho? They're basically getting tons of genuine follow-up questions from users, which they could then use to train the model to be more conversant