A Prompt Pattern Catalog

Check out our Prompt Research Page

For now, I am cataloging Peer Reviewed evidence-based prompting techniques only

Link to Prompt Catalog: Original Research (not mine) – Companion KB Annotation (Mine) – Annotation

As we step further into the age of artificial intelligence, the ability to communicate effectively with AI systems like ChatGPT has become increasingly important. Prompt engineering is not just about asking questions; it’s about asking the right questions and structuring those questions in a way that leverages the full capabilities of large language models (LLMs).

Researchers at Vanderbilt University have developed a catalog of prompt patterns that offer structured approaches to prompt engineering. These patterns are designed to solve common problems encountered when interacting with LLMs and enhance the quality of the outputs.

Key Highlights from the Catalog:
  • Meta Language Creation: Tailoring the language model to understand and generate responses in a custom language defined by the user.
  • Output Customization: Directing ChatGPT to produce outputs in specific formats or styles.
  • Error Identification: Using prompts that help identify inaccuracies in the generated responses.
  • Prompt Improvement: Refining prompts to elicit more accurate or relevant information.

Practical Applications

The patterns outlined in the catalog are not just theoretical; they have practical applications in software development, education, and beyond. For instance, software developers can use these patterns to automate coding tasks, while educators might use them to generate custom learning materials.

Getting Started with Prompt Engineering

For those new to this field, the key to mastery is practice. Begin by experimenting with different prompt patterns and observing how they influence ChatGPT’s responses. Over time, this practice will yield a deeper understanding of how to effectively communicate with AI models.

Detailed Overview of Prompt Patterns

Prompt Patterns are predefined methods or strategies used to improve interactions with language models like ChatGPT. They are designed to address specific challenges and enhance the quality of the AI’s responses.

1. Input Semantics: Meta Language Creation

  • Purpose: This pattern helps you define a custom language or set of terms that ChatGPT will understand and use.
  • Example: If you’re working with specific technical data, you might establish that “X -> Y” always means a specific type of data flow or relationship in your prompts.

2. Output Customization: Persona

  • Purpose: This pattern allows you to assign a role or persona to ChatGPT, shaping its responses based on that character.
  • Example: You might instruct ChatGPT to answer questions as if it were a historical figure, providing insights or responses that reflect that figure’s perspective and knowledge.

3. Error Identification: Fact Check List

  • Purpose: Helps ensure the accuracy of the information provided by having ChatGPT list the facts or data points it used, which you can then verify.
  • Example: After generating a summary of recent scientific findings, ChatGPT could list the key facts and figures it referenced, allowing you to double-check their accuracy.

4. Prompt Improvement: Question Refinement

  • Purpose: This pattern refines the questions you pose to ChatGPT to help elicit more precise or relevant answers.
  • Example: If you ask a vague question about climate change, ChatGPT could refine it to inquire about specific impacts of climate change in a defined region or time period.

5. Interaction: Flipped Interaction

  • Purpose: Switches the roles so that instead of you asking questions, ChatGPT does, guiding the conversation based on your responses.
  • Example: In a flipped interaction, if you were trying to solve a programming problem, ChatGPT might ask you questions to narrow down the error in your code.

Implementing These Patterns

Practical Application

  • For Educators: Use the Persona or Flipped Interaction patterns to create interactive learning experiences where students engage with AI to explore historical scenarios or solve complex problems through guided inquiry.
  • For Developers: The Meta Language Creation and Output Automater patterns can be instrumental in developing applications that interact with AI, helping to streamline coding tasks or automate repetitive software testing processes.
  • For Content Creators: Utilizing the Fact Check List pattern ensures the credibility of content generated by AI, crucial for maintaining trust and authority in published materials.

Conclusion

Integrating these prompt patterns into your work with AI can lead to more efficient, accurate, and engaging interactions. They provide a structured way to harness the capabilities of large language models, making them more applicable and useful across various fields and professions.

For a deeper dive into each pattern and more examples, the original research paper is a valuable resource.

This enriched overview not only aims to inform but also to inspire practical application, bridging the gap between abstract research and everyday utility.

Link to Prompt Catalog: Original Research (not mine) – Companion KB Annotation (Mine) – Annotation


Home » prompt » A Prompt Pattern Catalog