A Quick Guide

When working with language models, a well-crafted prompt can be the difference between vague output and valuable insights. This guide offers practical prompting patterns, customization instructions, and examples tailored for reasoning models. 

General Prompting Patterns

These patterns help guide LLMs more effectively, whether you’re seeking creative ideas or structured output.

1. Persona Pattern 
   Intent: Assign a specific role or identity to the LLM to shape its responses.
   Example:

   From now on, act as a senior academic journal reviewer in the field of computer science. 
   When I share my research abstract or methodology, provide feedback as a reviewer would, 
   focusing on clarity, novelty, and methodological rigor.

2. Question Refinement Pattern
   Intent: Help users improve their questions by suggesting better versions. 
   Example:

   Whenever I ask a question related to designing a research experiment or selecting a methodology, 
   suggest a better version of the question that includes relevant variables, constraints, 
   or assumptions. Then ask me if I’d like to proceed with your refined version.

   Another Example:

   Whenever I share a new research idea, ask me 10 detailed questions that will help refine and 
   clarify the idea. These questions should cover aspects such as the research problem, objectives, 
   methodology, data sources, expected outcomes, and potential limitations. After I answer them, 
   use my responses to suggest a more focused and well-structured version of my original research idea.

3. Alternative Approaches Pattern
   Intent: Encourage the LLM to suggest multiple ways to solve a problem. 
   Example: 

   Whenever I ask how to analyze my research data, suggest at least two alternative statistical or 
   computational methods, compare their strengths and weaknesses in terms of accuracy, 
   interpretability, and suitability for small sample sizes, and then ask me which one I’d like to 
   explore further.

4. Fact Check List Pattern
   Intent: Ensure the LLM outputs a list of critical facts that should be verified. 
   Example:

   When you generate a literature review summary or suggest related work, include a list of key 
   factual claims or references that should be verified for accuracy and relevance to my research 
   topic in machine learning. 

5. Flipped Interaction Pattern
   Intent: Flipped Interaction Pattern for improving initial prompt draft. 
   Example:

   I would like to create a prompt suitable for reasoning LLM. Topic is suggestions about effective 
   prompt engineering techniques for Intent Classification within Customer Service questions. Give me 
   a simple list of 10 questions that will help to get enough context to build the right prompt.

Prompting Instructions

Use instructions to shape style, clarity, and structure in responses.

1. Avoiding Double Dashes Instruction 
   Intent: Get rid of double dashes in the output. 
   Example:

   LLMs should avoid using em dashes (—) when generating text. Apply these guidelines instead:
   Use commas for parenthetical or interruptive phrases.
   Use periods to separate independent clauses for clarity and emphasis.
   Use colons to introduce lists or explanations.
   Use ellipses (...) sparingly, and only in narration, to indicate a soft dramatic pause—not to 
   imply incomplete thoughts.
   
   Revisions:
   “He arrived—unexpectedly—at the meeting.” → “He arrived, unexpectedly, at the meeting.”
   “The storm passed—they were relieved.” → “The storm passed. They were relieved.”
   “She packed three things—books, snacks, headphones.” 
   → “She packed three things: books, snacks, headphones.”
   “I was about to reply—” → “I was about to reply...”
   “If only they had listened—” → “If only they had listened.”
   "Our neighbor—Mrs. Patel—moved away last month.” 
   → "Our neighbor, Mrs. Patel, moved away last month."
   "He told me to head to—” → "He told me to head to..."
   "She opened the gift—and smiled."
   → "She opened the gift. She smiled."
   → "She opened the gift and smiled."
   → "She opened the gift...and smiled." (Use sparingly)

2. Writing Styles Instruction 
   Intent: Make the text look less like AI generated. 
   Example:

   Use simple language:
   Write clearly using short, plain sentences.
   Good example: "I need help with this issue."
   
   Avoid AI-sounding phrases:
   Skip clichés like “dive into” or “unleash your potential.”
   Avoid: "Let's dive into this game-changing solution."
   Good example: "Here's how it works."
   
   Be direct and concise:
   Get to the point. Cut unnecessary words.
   Good example: "We should meet tomorrow."
   
   Keep a natural tone:
   Write like you talk. It’s fine to start with “and” or “but.”
   Good example: "And that’s why it matters."
   
   Skip marketing language:
   Avoid hype or exaggerated claims.
   Avoid: "This revolutionary product will transform your life."
   Good example: "This product can help you."
   
   Be honest and real:
   Don’t force friendliness or enthusiasm.
   Good example: "I don’t think that’s the best idea."
   
   Simplify grammar:
   Perfect grammar isn’t required. Use your natural style.
   Good example: "I guess we can try that."
   
   Cut the fluff:
   Avoid unnecessary adjectives and adverbs.
   Good example: "We finished the task."
   
   Focus on clarity:
   Make your message easy to understand.
   Good example: "Please send the file by Monday."