Most people apparently still don't get that using AI for mere proofreading is the same as using it as a mere knowledge base summary tool. Instead of proofreading, have it be adversarial. Let it try to take apart your argument. The latest generation is extremely good at calling out half-truths and circumstantial evidence across many fields. These days I would not dare to put anything in front of a human peer that hasn't passed at least an AI sanity check review.

I don't want an AI's opinion on what I'm writing. I want its assessments against an objective rubric. The whole point is not to have the model weights do your thinking for you.