You cannot think with facts you have to look up
Reasoning happens in working memory, and working memory can only operate on what is already in your head. An insight arrives when two things you know collide — and they cannot collide if one of them is in a browser tab.
This is why the expert reads a paper and immediately notices the flaw, while the novice reads the same paper and notices nothing. The expert is not looking things up faster. They already had the material, so the contradiction was *visible*.
You cannot evaluate what you cannot check
The model will confidently tell you something false. To catch it, you need to know enough to be suspicious. A person with no knowledge cannot supervise a machine that produces plausible text — they can only accept it.
So the more capable AI gets, the *more* valuable it becomes to actually know things — because the job shifts from producing the answer to judging it, and judgment requires knowledge you already hold. Fabricated citations are caught by people who know the literature.
You cannot ask a question you don't know exists
Retrieval systems answer questions. They do not tell you which question to ask, and knowing what to ask is nearly the whole of expertise. The gap between a good question and a bad one is the gap between a useful hour and a wasted one, and it is made of prior knowledge.