Aaronson's test seems to require less "fluid intelligence" to get right. Someone who has read the book and just pattern matches basically can do well on the te

Aaronson's test seems to require less "fluid intelligence" to get right. Someone who has read the book and just pattern matches basically can do well on the test. A lot of tests work this way. Steve's test requires.. more fluid intelligence. Question 2 is simply not something you see on the internet in any common form, but the if you do lots of deduction of the implications -- ya, you get it. I've noticed similar weird fails with GPT-4. It seems like a genius recalling factual information and even (to some degree) translating common word problems into math. It starts failing hard though when you give it things way outside its domain (often children could outperform it since humans seem to not need so many shots of novel knowledge to understand things)