My own opinion: I equate LLMs to habitual action in humans. Asking what kind of processing it can or cannot do is the wrong question, because it comes down to how it is trained. What's missing is the rational layer that humans have and which they use to enrich their habitual skills and to train them