There are a wide variety of style guidelines that GPT-3 routinely violates, such as spelling out acronyms on first use, using full names when people are introduced, and so on. This creates a lot of editing work for humans trying to bring GPT-3 output up to professional standards. It would be very helpful if GPT-3 learned to do this better, either by training the model, or by post-processing outputs.
Noted! Thanks for your feedback Fred!