What prompts did you use for gauging accuracy? I was just struggling with coming up with some.

I would also like to hear how you benchmarked the accuracy.

Comparing output between competing GPT models?