I think this is because testing candidates in human trials is far removed from LLM prompting.