Sometimes it makes up strawmans where it implies you wrote or implied something insanely stupid and then "corrects" this. My interpretation of it is that it has been taught to give nuanced answers and seeing things from every perspective and somehow this goes overboard where it starts nuancing something "just in case" the user held non-nuanced views. Some cases are OK (if it just adds information) but I hate it when it goes "it is not X, it is Y..." where X is some stupid view you never implied and Y is what you actually wrote!