Some observations:
1) Questions are picked at random, but not random enough. I found it useful to say "Pick very random questions. I'm ready".
2) Occasionally GPT gets into "nazi" mode (e.g. very strict) and treats short answers as incorrect, when they're allowed. For example, who is current vice president. I say "Vance" and it replies "Incorrect, it's J.D. Vance".
3) There's some limit of how many conversations you can have per hour.