Define the AI Role or Persona

In this lesson we will discover one of the most powerful techniques within the art of prompting: role assignment. You will learn how to completely transform the way a language model responds simply by telling it "who" it should be when responding to you.

When you define a role (for example, “act as a historian”, “as a lawyer”, or even “as a robot from the future”) you are not only asking for information, you are giving it a voice, a tone, a point of view, and a mindset from which to build your response. This makes the interaction more natural, more precise, and much more aligned with the context you need.

You'll see how the role you choose affects the tone (whether it sounds formal, empathetic, humorous, or authoritarian), the style (whether you use technical, simple, or narrative language) and the type of knowledge which prioritizes the model. And, furthermore, you will understand why this technique is not limited to professions or experts, but can also be applied to fictional characters, abstract concepts, or even personified objects.

In summary, in this lesson you will learn how to give "life" to AI, so that it stops sounding generic and starts speaking with purpose, coherence, and personality.

“Act like a [Profession/Character/Expert]…”: The Magical Instruction

• What is Role Prompting?
  • It is a powerful technique where you explicitly instruct the LLM to adopt the identity, perspective, and communication style of a profession, a fictional character, a historical figure, or a specific type of expert.
  • The key phrase is usually: “Act like a…” either “Take on the role of a…” either “Imagine you are a…”.
• Why does it work?
  • LLMs have been trained with texts that include dialogues, character descriptions, professional manuals, articles written by experts, etc. They have “learned” the language patterns associated with different roles.
  • By assigning a role, you are activating those specific patterns, guiding the model to generate a response that aligns with the expected characteristics of that person or profession.
  • It is a very effective way to provide a very rich implicit context with a single sentence.
• Examples of Role Assignment:
  • Professional:
    • “Act like a lawyer specializing in intellectual property and explains the risks of using unlicensed images.”
    • “Act like a senior software engineer and review this Python code snippet for potential efficiency improvements: [code].”
    • “Act like a digital marketing consultant and suggests three strategies to increase engagement on Instagram for a small coffee shop.”
  • Creative/Fictional Character:
    • “Act like Sherlock Holmes and analyze this brief description of a mystery: [description].”
    • “Act like a 17th-century pirate and tell how you found hidden treasure.”
    • “Act like a robot of the future describing what daily life is like in the year 2242.”
  • Expert in a Specific Subject (even if it is not a “profession”):
    • “Act like a expert in the history of the Roman Empire and describes the life of a legionary.”
    • “Act like a organic gardening enthusiast and give me some advice on how to start a garden on my balcony.”
  • Personified Object or Concept:
    • “Act like planet Earth and write a letter to humanity.”
    • “Act like the concept of 'creativity'‘ and explains how to inspire people.”

Impact on Tone, Style, and Knowledge of the Response

Assigning a role has a transformative effect on multiple aspects of LLM output:

• Tone:
  • Definition: The general attitude or feeling conveyed by the text (formal, informal, serious, humorous, empathetic, critical, etc.).
  • Impact of the Role:
    • “Act like a stern judge“: Formal, authoritarian, perhaps critical tone.
    • “Act like a stand-up comedian“: Humorous, irreverent, colloquial tone.
    • “Act like a empathetic counselor“: Understanding, gentle, encouraging tone.
    • “Act like a 5-year-old boy“: Simple, innocent, curious tone.
• Style:
  • Definition: The way in which language is used, including word choice (vocabulary), sentence structure, sentence length, and the use of rhetorical figures.
  • Impact of the Role:
    • “Act like a 19th-century academic“Formal style, elaborate vocabulary, complex sentences.
    • “Act like a clickbait news headline writer“: Concise, impactful style, use of superlatives or intriguing questions.
    • “Act like a romantic poet“: Lyrical style, use of metaphors, evocative language.
• Knowledge (or Knowledge Prioritization):
  • Definition: Although the LLM has access to a vast knowledge base, the role helps to filter and prioritize the most relevant information for that person or perspective.
  • Impact of the Role:
    • Question: “Explain general relativity.”
      • Without a role: I could give a standard explanation.
      • “Act like a A theoretical physicist explaining it to another physicist“: He will use technical terminology, equations, and delve into the nuances.
      • “Act like a high school science teacher explaining it to his students“"It will use analogies, simpler language, and focus on key concepts without so much math.".
      • “Act like a science fiction writer seeking inspiration“: It could focus on the strangest or most speculative implications of the theory.
    • The role doesn't give him "new" knowledge, but it does indicate which part of your existing knowledge is most relevant and how to present it.
• Format and Structure (sometimes):
  • Certain roles involve specific communication formats.
  • “Act like a journalist writing a news story“: It will probably use the inverted pyramid structure.
  • “Act like a executive assistant drafting the minutes of a meeting“: It will use a minutes format with points, attendees, decisions.

Tips for Using Role Assignment Effectively:

• Be Specific About the Role: “Act like a chef” is good, but “Act like an Italian chef specializing in fresh pasta from northern Italy” is much better if you're looking for that level of detail.
• Combine with Other Instructions: Role-playing is a tool; it doesn't replace the need to give a clear task. “Act like a historian. Write a 500-word summary about the French Revolution for university students.”
• Experiment: Try different roles for the same task and see how the responses change. Sometimes, an unexpected role can lead to very creative results.
• Don't assume the role knows everything or is perfect: It's still a simulation. The AI could make mistakes or not perfectly capture all the nuances of the role, especially with very complex characters or very specific niches.
• Consider the “Depth” of the Role: Sometimes you just need a light touch (“write this in a friendly tone”), other times you’ll want a deeper dive into the person.

Defining a role for AI is one of the most impactful and easily implemented techniques for improving the quality and appropriateness of its responses. It allows you to shape the AI's "voice," ensuring that its responses are not only informative but also appropriate in tone and style for the intended purpose and audience. It's a fundamental tool in any prompt engineer's arsenal.