My questions with all of these tests is how do we verify the llm wasn't trained on the same questions/prompts especially since llms have been out for a while an

My questions with all of these tests is how do we verify the llm wasn't trained on the same questions/prompts especially since llms have been out for a while and they can incrementally update the trainings? Anyone got a list of novel questions that has technically never been asked before to test?