Benchmarking large language models presents some unusual challenges. For one, the main purpose of many LLMs is to provide compelling text that’s indistinguishable from human writing. And success in ...