You're right. But if you enter 0.3 on average over 28 questions and the actual number of true answers differ by a lot from 8 then you have learned your general

You're right. But if you enter 0.3 on average over 28 questions and the actual number of true answers differ by a lot from 8 then you have learned your general sense of GPT-4s abilities is uncalibrated.