Whatever the answer to that conundrum might be, LLMs are trained on these patterns and replicate them pretty faithfully.