I'm so good at AI, that I can beat this.
Ask the same question 4 different times, on 4 different models.
If any of the 4 contradict each other, have a human manually review.
You can even change the question, change the seed, etc...
Its interesting how few people know how to use LLMs effectively. Here is my most recent protip, after getting an answer say:
"That was a 2/10 answer, give me a 10/10 answer"