AI evaluations: Practical content standards for real LLMs