Current LLMs have a number of limitations that human reasoning doesn't. Whether these are intrinsic to the technology or can be overcome with larger and better datasets is an open question.