I'm an AI implementation engineer consulting with real business implementing LLMs. In my most recent post, I investigate how prompting with an expert persona ("act like an expert") actually affects factual accuracy of language models when answering multiple-choice questions, and whether the added cost is worth it.