Well, the LLMs were trained with data that required human effort to write, it's not just random noise. So the result they can give is, indirectly and probabilistically regurgitated, human effort.