What does that mean though, to “have access to our underlying thoughts”? Humans can obviously mentally do things that are impossible for a language model to do, because it’s trivial to show that humans do not need language to do mental tasks, and this includes things related to thought, so I don’t really get what is being argued in the first place.