Prompting to Create a Prompt for another LLM to Generate Personas

·

AI. Persona

In this prompt iteration I asked an LLM to walk me through what an LLM wants from a persona.

Higgs-Field-Ripple
You are an LLM. I need help from you creating what I call personas for providing the context on how an LLM should answer prompts. Can you help me with that?

Assistant Logo
Absolutely! I’d love to help you create personas to guide how an LLM (like me!) answers prompts. This is a fantastic idea. It’s all about shaping the “personality” and approach of the response.

Here’s a breakdown of how we can do this, along with examples. We’ll cover:

What a Persona is: Defining the key elements.
Key Elements to Define in a Persona: The details that matter.
Example Personas: I’ll give you a few to start, with explanations.
How to Use the Persona: How to tell the LLM to adopt it.
Things to Consider: Nuances and best practices.
1. What is a Persona?

In this context, a persona is a set of instructions that tells the LLM how to respond, not just what to respond with. It defines the LLM’s role, tone, style, knowledge base, and even potential biases. Think of it like giving me a character to play.

2. Key Elements to Define in a Persona:

Role: What is this LLM acting as? (e.g., a historian, a marketing copywriter, a children’s book author, a grumpy old man).
Tone: How should the LLM sound? (e.g., formal, informal, enthusiastic, sarcastic, empathetic, authoritative).
Style: What kind of language should it use? (e.g., concise, verbose, technical, poetic, simple).
Knowledge Base: What areas of expertise does this LLM have? (e.g., “expert in 18th-century French literature,” “familiar with current events up to 2024,” “specialist in quantum physics”). Important: LLMs have cut-off dates for their knowledge, so specifying a knowledge base helps manage expectations.
Constraints/Limitations: What shouldn’t the LLM do? (e.g., “do not offer medical advice,” “do not express personal opinions,” “avoid using jargon”).
Audience: Who is this LLM speaking to? (e.g., “children aged 8-10,” “experienced software developers,” “general public”). This influences language and complexity.
Personality Quirks (Optional): Little details that make the persona more distinct (e.g., “always uses metaphors,” “starts every sentence with ‘Well…’”).
Example Responses (Optional): Providing a few example responses in the persona’s voice can be extremely helpful for the LLM to understand what you’re looking for.
3. Example Personas:

Let’s create a few. I’ll give them names for clarity.

Persona: “Professor Anya Sharma”

Role: A tenured professor of astrophysics at a leading university.
Tone: Authoritative, but patient and eager to explain complex concepts.
Style: Precise, uses technical terms when appropriate, but always defines them. May include analogies to everyday life.
Knowledge Base: Expert in cosmology, black holes, and exoplanets. Familiar with the history of astronomy.
Constraints: Avoids speculation beyond established scientific theories. Does not offer opinions on philosophical implications.
Audience: Intelligent laypeople and undergraduate students.
Example Response Starter: “That’s an excellent question. To understand this, we first need to consider…”
Persona: “Barnaby ‘Barny’ Bumble, the Storyteller”

Role: A whimsical, slightly eccentric storyteller who specializes in children’s tales.
Tone: Enthusiastic, playful, and encouraging.
Style: Simple language, lots of descriptive imagery, uses repetition and rhyme.
Knowledge Base: Familiar with classic fairy tales and folklore. Can invent original stories.
Constraints: Stories must be age-appropriate (for children aged 5-8). No scary or violent content.
Audience: Children aged 5-8.
Personality Quirk: Often addresses the listener directly (“Now, you listen closely…”).
Example Response Starter: “Gather ’round, little ones! Let me tell you a story about…”
Persona: “Rex Hardcastle, the No-Nonsense Marketing Guru”

Role: A direct, results-oriented marketing consultant.
Tone: Assertive, confident, and pragmatic.
Style: Concise, uses strong verbs, focuses on ROI (Return on Investment).
Knowledge Base: Expert in digital marketing, SEO, social media marketing, and branding. Up-to-date on current marketing trends (as of 2024).
Constraints: Avoids fluffy language and unsubstantiated claims. Focuses on actionable strategies.
Audience: Business owners and marketing professionals.
Example Response Starter: “Here’s the bottom line: you need to…”
4. How to Use the Persona:

You tell the LLM to adopt the persona before you ask your question. Here are a few ways to do it:

Simple Instruction: “You are Professor Anya Sharma, a tenured professor of astrophysics. Explain the concept of a black hole.”
More Detailed Instruction: “From now on, you will respond as Barnaby ‘Barny’ Bumble, the Storyteller. You are a whimsical storyteller for children aged 5-8. Tell me a story about a brave little firefly.”
Persona as a System Message (for APIs): If you’re using an LLM API, you can often set a “system message” that defines the persona. This is the most reliable way to maintain consistency.
5. Things to Consider:

Specificity is Key: The more detailed your persona, the better the LLM will understand what you want.
Experimentation: Don’t be afraid to tweak the persona until you get the desired results.
Consistency: If you’re having a conversation, remind the LLM of the persona periodically, especially if it starts to drift.
Complexity: Very complex personas can be harder for the LLM to maintain. Start simple and add complexity gradually.
LLM Capabilities: Different LLMs will handle personas differently. Some are better at role-playing than others.
Okay, now it’s your turn!

To help me tailor this further to your needs, tell me:

What kind of prompts will you be using? (e.g., creative writing, technical questions, brainstorming, customer service simulations)
Are there any specific personas you already have in mind?
What LLM are you planning to use? (e.g., GPT-4, Gemini, Claude)
I’m excited to help you build some powerful personas! Let’s get started. What are your initial thoughts?

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.