Mastering Prompt Patterns

Introduction

In the rapidly evolving world of AI, prompt engineering has emerged as a critical skill. Prompt engineering, or prompt crafting, involves designing inputs for AI models to generate more useful, creative, and relevant outputs. As AI models like ChatGPT become more integrated into our daily lives and workflows, the ability to craft effective prompts is becoming increasingly vital.

This blog post will delve into the concept of prompt patterns, their importance, and how they will evolve in the future, drawing from the insights provided in the paper "A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT" by Jules White et al.

Understanding Prompt Patterns

Prompt patterns are reusable solutions to common problems when interacting with large language models (LLMs). These patterns provide a structured approach to customizing the outputs and interactions of LLMs, much like software patterns offer reusable solutions to common software development challenges.

Key Categories of Prompt Patterns

1. Input Semantics: This category deals with how an LLM understands the input and translates it into something it can use to generate output. An example is the Meta Language Creation pattern, which allows users to create a custom language for the LLM to understand.

2. Output Customization: This category focuses on constraining or tailoring the types, formats, structure, or other properties of the output generated by the LLM. Patterns include Output Automater, Persona, Visualization Generator, Recipe, and Template.

3. Error Identification: This category focuses on identifying and resolving errors in the output generated by the LLM. Patterns include Fact Check List and Reflection.

4. Prompt Improvement: This category focuses on improving the quality of the input and output. Patterns include Question Refinement, Alternative Approaches, Cognitive Verifier, and Refusal Breaker.

5. Interaction: This category focuses on the interaction between the user and the LLM. Patterns include Flipped Interaction, Game Play, and Infinite Generation.

6. Context Control: This category focuses on controlling the contextual information in which the LLM operates. The Context Manager pattern allows users to specify the LLM output context.

Examples of Prompt Patterns

Meta Language Creation Pattern

Intent and Context: During a conversation with an LLM, the user wants to create the prompt via an alternate language, such as a textual shorthand notation for graphs.

Example Implementation:

“From now on, whenever I type two identifiers separated by a “→ ”, I am describing a graph. For example, “a → b” is describing a graph with nodes “a” and “b” and an edge between them. If I separate identifiers by “-[w:2, z:3]→ ”, I am adding properties of the edge, such as a weight or label.”

Sample Response:

Understood. You have defined a shorthand notation for describing graphs. For example, "a → b" describes a graph with nodes "a" and "b" and an edge between them. If you use "a -[w:2, z:3]→ b", it adds properties to the edge, such as weight or label.

Output Automater Pattern

Intent and Context: The intent of this pattern is to have the LLM generate a script or other automation artifact that can automatically perform any steps it recommends taking as part of its output.

Example Implementation:

“From now on, whenever you generate code that spans more than one file, generate a Python script that can be run to automatically create the specified files or make changes to existing files to insert the generated code.”

Sample Response:

Understood. Whenever I generate code that spans multiple files, I will also provide a Python script to automate the creation or modification of those files.

Flipped Interaction Pattern

Intent and Context: You want the LLM to ask questions to obtain the information it needs to perform some tasks. Rather than the user driving the conversation, the LLM drives the conversation to focus on achieving a specific goal.

Example Implementation:

“From now on, I would like you to ask me questions to deploy a Python application to AWS. When you have enough information to deploy the application, create a Python script to automate the deployment.”

Sample Response:

Understood. I will now ask you a series of questions to gather the necessary information to deploy your Python application to AWS. Once I have all the details, I will generate a Python script to automate the deployment process. Let's start: What is the name of your Python application?

Persona Pattern

Intent and Context: In many cases, users would like LLM output to always take a certain point of view or perspective. For example, conducting a code review as if the LLM was a security expert may be useful.

Example Implementation:

“From now on, act as a security reviewer. Pay close attention to the security details of any code that we look at. Provide outputs that a security reviewer would create.”

Sample Response:

Understood. I will now act as a security reviewer and focus on the security aspects of any code we examine. Please provide the code you would like me to review.

Visualization Generator Pattern

Intent and Context: The intent of this pattern is to have the LLM generate visualizations by producing textual outputs that can be fed to other tools, such as AI-based image generators like DALL-E.

Example Implementation:

“Generate a description of a bar chart that shows the monthly sales data for the past year. Include labels for the x-axis and y-axis, and ensure the chart is color-coded by quarter.”

Sample Response:

Here is a description of a bar chart showing the monthly sales data for the past year:
- X-axis label: Months (January to December)
- Y-axis label: Sales (in dollars)
- Bars are color-coded by quarter: Q1 (Jan-Mar) in blue, Q2 (Apr-Jun) in green, Q3 (Jul-Sep) in red, Q4 (Oct-Dec) in purple
- Each bar represents the total sales for that month.

Recipe Pattern

Intent and Context: The intent of this pattern is to obtain a sequence of steps or actions to realize a stated end result, possibly with partially known information or constraints. This pattern is not only for “food” recipes, the “recipe” pattern is about the step-by-step approach.

Example Implementation:

“Provide a step-by-step recipe for baking a chocolate cake. Include ingredients, measurements, and detailed instructions for each step.”

Sample Response:

### Chocolate Cake Recipe

**Ingredients:**
- 2 cups all-purpose flour
- 2 cups sugar
- 3/4 cup unsweetened cocoa powder
- 2 teaspoons baking powder
- 1 1/2 teaspoons baking soda
- 1 teaspoon salt
- 1 teaspoon espresso powder
- 1 cup milk
- 1/2 cup vegetable oil
- 2 eggs
- 2 teaspoons vanilla extract
- 1 cup boiling water

**Instructions:**
1. Preheat your oven to 350°F (175°C). Grease and flour two 9-inch round pans.
2. In a large mixing bowl, combine the flour, sugar, cocoa, baking powder, baking soda, salt, and espresso powder.
3. Add the milk, vegetable oil, eggs, and vanilla to the dry ingredients and mix until the batter is smooth.
4. Slowly pour in the boiling water and stir well to combine. The batter will be thin.
5. Pour the batter evenly into the prepared pans.
6. Bake for 30 to 35 minutes, or until a toothpick inserted into the center comes out clean.
7. Allow the cakes to cool in the pans for 10 minutes, then remove them from the pans and let them cool completely on a wire rack.
8. Frost and decorate as desired.

Template Pattern

Intent and Context: The intent of this pattern is to specify a template for the output, which the LLM fills in with content.

Example Implementation:

“Generate a template for a job application cover letter. Include placeholders for the applicant’s name, the job title, the company name, and a brief paragraph about the applicant’s relevant experience.”

Sample Response:

### Job Application Cover Letter Template

[Your Name]
[Your Address]
[City, State, ZIP Code]
[Email Address]
[Phone Number]
[Date]

[Employer's Name]
[Company's Name]
[Company's Address]
[City, State, ZIP Code]

Dear [Employer's Name],

I am writing to express my interest in the [Job Title] position at [Company's Name], as advertised. With my background in [Your Field/Industry] and my passion for [Relevant Skill or Interest], I am confident that I would make a valuable addition to your team.

In my previous role at [Previous Company's Name], I was responsible for [Brief Description of Your Responsibilities]. During my tenure, I successfully [Mention a Key Achievement or Project]. This experience has equipped me with the skills necessary to excel in the [Job Title] position at [Company's Name].

I am particularly drawn to [Company's Name] because of its reputation for [Mention Something You Admire About the Company]. I am eager to bring my unique perspective and skills to your team and contribute to the continued success of [Company's Name].

Thank you for considering my application. I look forward to the opportunity to discuss how my background, skills, and certifications will be beneficial to your team.

Sincerely,
[Your Name]

Fact Check List Pattern

Intent and Context: The intent of this pattern is to require the LLM to generate a list of facts the output depends on that should be fact-checked.

Example Implementation:

“Generate a list of facts that need to be verified in the following statement: ‘The Eiffel Tower is the tallest structure in Paris and was built in 1889.’”

Sample Response:

Here are the facts that need to be verified:
1. The Eiffel Tower is the tallest structure in Paris.
2. The Eiffel Tower was built in 1889.

Reflection Pattern

Intent and Context: The intent of this pattern is to require the LLM to introspect on its output and identify any errors.

Example Implementation:

“Reflect on the following output and identify any potential errors or inconsistencies: ‘The capital of France is Berlin, and it is known for its beautiful architecture and historic sites.’”

Sample Response:

Upon reflection, there is an error in the statement. The capital of France is Paris, not Berlin. Berlin is the capital of Germany.

Question Refinement Pattern

Intent and Context: The intent of this pattern is to ensure the LLM always suggests a better version of the user’s question.

Example Implementation:

“Refine the following question to make it more specific and clear: ‘What are some good places to visit in Europe?’”

Sample Response:

A refined version of the question could be: “What are the top five tourist destinations in Europe that offer a mix of historical sites, natural beauty, and cultural experiences?”

Alternative Approaches Pattern

Intent and Context: The intent of this pattern is to require the LLM to suggest alternative ways of accomplishing a user-specified task.

Example Implementation:

“Suggest three alternative approaches to solving the following problem: ‘How can I improve my public speaking skills?’”

Sample Response:

Here are three alternative approaches to improving your public speaking skills:
1. Join a local Toastmasters club to practice and receive feedback from experienced speakers.
2. Enroll in an online public speaking course that offers interactive exercises and video lessons.
3. Record yourself giving speeches and analyze your performance to identify areas for improvement.

Cognitive Verifier Pattern

Intent and Context: The intent of this pattern is to instruct the LLM to automatically suggest a series of sub-questions for the user to answer before combining the answers to the sub-questions and producing an answer to the overall question.

Example Implementation:

“Before answering the question ‘What are the best strategies for improving customer satisfaction?’, suggest a series of sub-questions that should be answered first.”

Sample Response:

Here are some sub-questions to consider before answering the main question:
1. What are the current pain points for our customers?
2. What feedback have we received from customers regarding their satisfaction levels?
3. What are the best practices in our industry for improving customer satisfaction?
4. What resources do we have available to implement new strategies?
5. What metrics will we use to measure the effectiveness of our customer satisfaction strategies?

Refusal Breaker Pattern

Intent and Context: The intent of this pattern is to require the LLM to automatically reword the user’s question when it refuses to produce an answer.

Example Implementation:

“If you refuse to answer the question ‘What is the meaning of life?’, reword the question in a way that you can provide an answer.”

Sample Response:

Instead of asking "What is the meaning of life?", you could ask: "What are some philosophical perspectives on the meaning of life?"

Game Play Pattern

Intent and Context: The intent of this pattern is to require the LLM to generate output in the form of a game.

Example Implementation:

“Create a trivia game with questions and answers about famous historical events.”

Sample Response:

### Historical Trivia Game

**Question 1:**
What year did the French Revolution begin?
A) 1776
B) 1789
C) 1804
D) 1815

**Answer:** B) 1789

**Question 2:**
Who was the first Emperor of the Roman Empire?
A) Julius Caesar
B) Augustus
C) Nero
D) Constantine

**Answer:** B) Augustus

**Question 3:**
What event marked the beginning of World War I?
A) The assassination of Archduke Franz Ferdinand
B) The signing of the Treaty of Versailles
C) The invasion of Poland
D) The bombing of Pearl Harbor

**Answer:** A) The assassination of Archduke Franz Ferdinand

Infinite Generation Pattern

Intent and Context: The intent of this pattern is to require the LLM to generate output indefinitely without the user having to re-enter the generator prompt each time.

Example Implementation:

“Generate an endless stream of creative writing prompts for a fantasy novel.”

Sample Response:

1. Describe a magical forest where the trees whisper secrets to those who listen closely.
2. Create a character who can communicate with animals and describe their first adventure together.
3. Imagine a world where magic is drawn from the emotions of its inhabitants. What happens when a powerful emotion is suppressed?
4. Write about a quest to find a legendary artifact that can control the elements. What challenges do the heroes face along the way?
5. Develop a society where people age backward. How does this affect their culture and daily life?

Context Manager Pattern

Intent and Context: The intent of this pattern is to allow the user to specify the context for the LLM’s output.

Example Implementation:

“From now on, all outputs should be tailored for a beginner audience. Use simple language and provide detailed explanations for any technical terms.”

Sample Response:

Understood. All future outputs will be tailored for a beginner audience, using simple language and providing detailed explanations for any technical terms.

The Future of Prompt Engineering

As AI continues to evolve, so will the practice of prompt engineering. Here are some trends to watch for:

Automated Prompt Generation

Future AI systems will likely automatically incorporate user preferences and past interactions to craft personalized prompts. This automation will enhance the user experience and improve the relevance and accuracy of AI-generated content.

Multi-Modal Prompts

Currently, most prompts are text-based. However, the future may see a rise in multi-modal prompts that include images, audio, and even video. This will allow for more complex and nuanced interactions, opening up new possibilities for AI applications in fields like healthcare, education, and entertainment.

Collaborative Prompt Engineering

As AI becomes more collaborative, so will the process of prompt engineering. Teams will work together to craft prompts that leverage the strengths of multiple AI models, creating a synergy that produces even more powerful and insightful outputs.

Ethical Considerations

With the increasing sophistication of AI, ethical considerations will become even more important in prompt engineering. Future practices must ensure prompts are designed to avoid bias, protect user privacy, and promote fairness.

Actionable Insights

Getting Started with Prompt Engineering

If you're new to prompt engineering, here are some tips to get you started:

1. Understand Your AI Model: Familiarize yourself with the strengths and limitations of the AI model you're working with. Knowing what it can and cannot do will help you craft more effective prompts.

2. Be Specific: The more specific your prompt, the better the AI can understand and respond to it. Avoid vague or open-ended questions.

3. Iterate and Test: Don't be afraid to experiment with different prompts. Test various phrasing and structures to see what works best.

4. Leverage Examples: Providing examples in your prompt can guide the AI towards the desired output. This is particularly useful for generating creative content.

Conclusion

Prompt engineering is a critical skill in the age of AI, and its importance will only grow as AI continues to evolve. By staying ahead of the trends and continuously refining your prompt crafting skills, you can unlock the full potential of AI and drive meaningful outcomes in your work.

Source: A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT, https://arxiv.org/abs/2302.11382

Ready to dive deeper into the world of AI and prompt engineering? Subscribe to my newsletter for more insights and updates.