The idea of that exercise is to see how "confident" the AI is in giving you a correct-sounding answer for something it can not have any knowledge of because you

The idea of that exercise is to see how "confident" the AI is in giving you a correct-sounding answer for something it can not have any knowledge of because you just made it up - and to realize that it's doing the very same thing with any other topic unless its statistical models happen to nudge it into using memorized facts to increase the likelihood of sounding correct. I think it gives you more insight into what GPT actually is than just seeing it being wrong about something you're knowledgeable in (which shouldn't really be surprising).