It is completely reasonable and often - very - useful to evaluate and interpret instructions with LLMs.

You're stuck on the anthropomorphize semantics, but that wasn't the purpose of the exercise.