Prompt Engineering for Technical Writing 101
Chapter 2: The 5 Step Prompt Engineering Framework

Step 0: Starting Point

Complete & Continue Next Lesson Learn More

Outcomes

  • Analyze the different classifications of prompting methods and their impact on the quality and accuracy of AI-generated responses.

In This Lesson

  • Outcomes

  • Step 0: Starting Point.

  • Reference on Types and Classifications of Prompts.

  • References

In the following few lessons, we will review how best to utilize ChatGPT in your technical writing workflow by working through an example. I have created this “Technical Writing Framework” based on the community, documentation, samples, and what I have learned from creating my own technical content and learning materials.

5-Step Prompt Crafting Framework

Note: These prompts are a little wordy for descriptive purposes in this demo.  You can definitely convert this to shorter sentences or point form when creating your prompts.

Step 0: Starting Point.

Let’s take this sample just as it is in a new ChatgGPT prompt. We will refine the content as we go. But first, let’s review how these results change as we move through the list.

Prompt:
I need a content summary and outline about learning Power BI.

Response:

This was an excellent initial response. The goal is now to engineer the prompting to be closer to the desired output.

Reference on Types and Classifications of Prompts.

I am going to put the following tables here as a point of discussion and an example of the different types of prompts we can use depending on the task at hand. Also I use the references at the bottom of the lesson consistantly.

Below is a table that categorizes prompting methods and describes each classification. This will help you understand how these classifications guide the types of prompting used to interact with language models.

Classification
Description
Logical and Sequential Processing
It aims to direct the model's output more precisely toward specific goals or emotional responses.
Contextual Understanding and Memory
It involves guiding the model through a logical sequence of steps to improve reasoning and problem-solving abilities.
Specificity and Targeting
Aims to direct the model's output more precisely towards specific goals or emotional responses.
Data Manipulation
Concerns the use of example-based methods to teach the model how to approach new tasks or concepts without prior direct experience.
Iterative Refinement
Involves refining and improving the model's responses through repetition and iterative feedback, enhancing the depth and accuracy of its reasoning.

Each classification represents a strategic approach to shaping the model's interactions and outputs, ensuring more effective and targeted results in various applications.

This expanded version below reviews the Types of Prompting methods by classification. It provides a more comprehensive overview of the different types of prompt engineering techniques.

This table provides a structured overview, showcasing the diversity in prompting methods used to steer or improve the outputs of language models. Each type of prompting can be selected based on the specific needs of the task.

Classification
Type of Prompting
Description
Sample
Logical and Sequential Processing
Chain-of-Thought Prompting
Breaks down complex reasoning tasks into step-by-step logical chains, improving problem-solving and human-like reasoning.
"Explain how a bill becomes law in the United States, step by step."
Logical and Sequential Processing
Step-Back Prompting
Provides underlying principles or context before asking the model to solve a problem, ensuring robust understanding.
"Before solving this math problem, consider the fundamental properties of integers."
Contextual Understanding and Memory
Retrieval Augmented Prompting
Provides relevant information or context to the model for better understanding and coherence.
"Given recent economic data, analyze the likely impact on monetary policy."
Contextual Understanding and Memory
Conversational Prompting
Allows the model to reference past interactions for seamless conversational experiences.
"Recall our last conversation about climate change and continue discussing its impacts on agriculture."
Specificity and Targeting
Emotional Prompting
Uses persuasive and emotional language to stimulate deeper focus and commitment from the model.
"Write a persuasive speech about the importance of recycling."
Specificity and Targeting
Directional-Stimulus Prompting
Includes hints, keywords, or desired outputs to guide the model towards specific goals.
"Create a list of bullet points focusing on the benefits of renewable energy."
Data Manipulation
Few-Shot Prompting
Provides a few examples or instances to the model for learning a new task or concept.
"Here are three examples of email greetings; now write a professional email opening for a job application."
Data Manipulation
Zero-Shot Prompting
Tests the model's ability to perform a task without any examples or training data.
"What are some potential benefits of space exploration?"
Data Manipulation
One-Shot Prompting
Provides a single example or instance to the model for learning a new task or concept.
"Given this example of a customer complaint email, draft a response."
Iterative Refinement
Self-Refine Prompting
Prompts the model to solve a problem, critique its solution, and refine the solution iteratively.
"Solve this algebra problem, then review and correct any errors in your solution."
Iterative Refinement
Tree-of-Thought Prompting
Generates multiple "possible next steps" and explores them through tree search or beam search.
"Identify several strategies to reduce urban pollution and evaluate each for effectiveness."
Iterative Refinement
Maieutic Prompting
Encourages the model to explain its reasoning, then prompts for further explanations iteratively.
"Discuss the causes of the French Revolution, and elaborate further on each point."

Used ChatGPT to brainstorm and format the list into a table.

 

These prompting techniques enhance language models' precision, coherence, and reasoning capabilities for various tasks and applications.

References

I must say, these are the best references I have found for this topic. In-depth examples and explanations.