I had the same thought while I was reading. Currently, ChatGPT ultimately defers to the user as the one who "knows best".

And it probably has to be that way now, since ChatGPT doesn't seem to possess a truly generalizable logic engine / pattern recognition / predictive system. For instance, last I checked, ChatGPT couldn't solve sudoku puzzles or write poems that follow a particular rhyminc pattern an meter.

But I do hold a cautious optimism that once ChatGPT has true logical abilities, then it might be able to become the thing that the author is hoping for - an entity that can weigh ideas logically and help correct people when they are wrong.