𝗧𝗵𝗲 𝗦𝗶𝗺𝗽𝗹𝗲𝘀𝘁 𝗪𝗮𝘆 𝘁𝗼 𝗜𝗺𝗽𝗿𝗼𝘃𝗲 𝗟𝗟𝗠 𝗔𝗻𝘀𝘄𝗲𝗿𝘀
Most people get average answers from AI.
An LLM predicts the average response found on the internet. It gives you the most common answer. This often lacks depth.
You can fix this with one step.
Force the model to pick a specific person to answer your question. Then, make it answer as that person.
This works because experts leave a large footprint online. Their books, blogs, and lectures are in the training data. The model understands how they think.
Asking "what does Paul Graham think about my startup" works better than asking for a general opinion.
I improved this method through testing. Do not let the model pick a generic role. Force it to choose one specific person or a panel of experts.
Ask the model to explain why these specific people fit your question. This step makes the character simulation stronger.
I turned this method into a reusable Claude Code skill.
You can download the expert.md file and place it in your ~/.claude/skills/ folder. You can also give this format to any LLM you use.
Source: https://dev.to/lukaskunhardt/the-simplest-way-to-improve-llm-answers-43f0
Optional learning community: https://t.me/GyaanSetuAi