https://bfcmath.github.io/posts/Prompting-bias-in-LLM-Benchmarks/