Subtle problem with AI-written academic papers is that AI isn't calibrated for feedback. AI often just folds, but if you are sending a paper for review, challenges from peer reviewers should result in some attempt to defend your idea (while still giving up on it if really wrong)