I get what the author is trying to say, that in many respects people go about their day in a form of autocomplete mode, that is guided by our assumptions about the world, and to an extent I agree, but that is different from a LLM, that is auto completing based on learned probability from patterns in a data set.

I say that because if you break social norms, say getting up into the face and staring at people in a lift, they will try and perpetuate their autocomplete mode functions for a short while, then break from them. In contrast an LLM will just hallucinate, or to put it more bluntly, ‘make shit up’ when faced with a dilemma, rather than genuinely adapt to novel circumstances.