Despite all the anthropomorphizing of LLMs, you must have come across already how each has VERY DISTINCT failure modes?