Except, of course, LLMs are not humans, and they do not learn or "reason" in a way which even remotely resembles humans.

Plus obviously humans can still overfit to a specific style of test.