Chapter 1.3: Role-based prompting
What is role-based prompting?
A role-based prompt focuses on the persona that the model should assume when answering your question or performing the task you assigned. You ask the model to 'play a role' to make sure you get feedback from a useful point of view.
How to write a good role-based prompt
- Define the role: Give the AI a detailed profile of the expert you want it to impersonate.
- Let the AI confirm the role: Ask the model to confirm the role and have it explain why this expertise is relevant to your question.
- Reference to AI's expertise: State your question or explain the task and refer back to the expert-role that the model should assume.
Example: Feedback on grant proposal
In this example we ask the AI model to give us feedback on an NWO Vidi proposal that we wrote. We want the feedback to be given from the perspective of an experienced reviewer.
Step 1: Define the role
You are a member of the NWO Vidi Assessment Committee.
You are an experienced reviewer, but also a well-established researcher in ecology yourself.
You are known for your honest, thorough feedback.
You pay attention to detail, compliance with funding guidelines, feasability and originality.
Step 2: Let the AI confirm the role
Please confirm that you have assumed the role of a member of the NWO Vidi Assessment Committee and explain why your expertise is relevant to assessing my Vidi proposal.
Step 3: Reference to AI's expertise
As a member of the NWO Vidi Assessment Committee, what feedback would you give me on my proposal for a Vidi grant?
Source
Based on materials from AI report
⬅ Back to Few shot prompting | ➡ Next: Audience