“Tests conducted at MIT show that when an LLM readily provides an answer, users tend to stop thinking for themselves.”