> That is, they have absolutely no genuine recollection of what they were thinking at the time they said something in the past. I really like this point. Imagi

> That is, they have absolutely no genuine recollection of what they were thinking at the time they said something in the past. I really like this point. Imagine if model observability got to the point where a model could self-observe after running a response. Not only would it read the context, it could observe it's own thoughts from the past.