ChatGPT Prompts Book - Precision Prompts, Priming, Training & AI Writing Techniques for Mortals : Crafting Precision Prompts and Exploring AI Writing with ChatGPT, 2024, by Oliver Theobold, Packt Publishing
Prompts are the input you give that allow an AI tool to produce output. The output quality from AI models is highly impacted by how you craft your prompt. Prompts might be questions to be answered, partial input to be completed, or instructions, simple or complex. Learn about the CLEAR framework in the box below.
As a student, employee/employer or citizen, remember that YOU are legally and ethically responsible for your decisions, including responsibility for input you provide to AI models and how you use any content you create with AI assistance. Here are some things to consider—
Are you authorized to use AI in the case you are considering?
Evaluate suitability- should AI be involved in this process?
What is your plan for transparency? Ethical AI use includes disclosing to clients, team members, instructors, etc. when AI tools have contributed to your work product,
Concise prompts are focused, specific, and clear.
For example,
"What are the physical health benefits of living with a dog?"
rather than
"Please tell me about the benefits a person might experience from having pets."
Both of these prompts would work, but a more focused, specific, and clear prompt will provide a more specific answer.
You might iterate this prompt by adding, "List the physical health benefits of living with a dog. Provide an explanation for each point and include evidence from experts."
At the top of this box, use the next tab in the sequence to move to the L in the CLEAR Framework.
Logical prompts provide a coherent structure, such as a natural progress, and make relationships between concepts clear. Including the logic in the prompt reduces errors and results in better output.
For example,
"List and describes the steps in effectively searching a library database."
rather than
"How do I research a topic."
You might iterate this prompt by adding, "You are a college librarian. List and describe the steps in effectively searching a library database as if you are teaching a first year student."
At the top of this box, use the next tab in the sequence to move to the E in the CLEAR Framework.
Explicit prompts give the output specifications, providing context and instructions for desired format, content, or scope.
For example,
"Provide a very simple recipe for chocolate cake with no dairy."
rather than
"How to make a chocolate cake without dairy."
You might iterate this prompt to something like, "Provide a very simple recipe for dairy-free chocolate cake and frosting. This is for a first-time baker who needs a lot detailed explanation. Include an equipment list and timing chart."
These first three considerations (concise, logical, explicit) should produce a reasonable effective prompt. Next, we focus improving, both our approach, and our output. At the top of this box, use the next tab in the sequence to move to the A in the CLEAR Framework.
Those first three steps should produce a reasonable effective prompt. Next, we focus improving both our approach and our output. Even if the initial output was ok for our purpose, it likely can improve if we are ADAPTIVE .
We have iterated in each step, but now it is time to collect clues from the output and experiment with changing the prompt— be flexible and customize the prompt to get better output.
For example.
Initial prompt— "How should a student use AI?"
Iterated prompt—"List specific uses of AI for college students."
Adapted prompt—"You are a success coach for first year college students. List and describe ways students can appropriately apply AI tools to enhance learning and recall of course materials in the sciences."
At the top of this box, use the next tab in the sequence to move to the R in the CLEAR Framework.
After crafting a Concise, Logical, and explicit prompt and being adaptive, the final step in the CLEAR Framework calls for us to be Reflective, evaluating our prompts and the output thoughtfully and critically.
We must be strategic as we review our prompt and output carefully. Sometimes output contains misinformation, so we may need to fact-check. Sometimes we need to go back to adapting our prompt because we want our output to be more specific or more broad, focused in a slightly different way, or intended for a different audience.
Consider prompt engineering a cycle- craft the prompt, test the prompt, analyze the output, improve the prompt, re-test the prompt, analyze the output, and so on.
The final tab provides a citation for the article that offered this framework for an information literacy oriented approach to prompt engineering.
Lo, L. S. (2023). The CLEAR path: A framework for enhancing information literacy through prompt engineering. The Journal of Academic Librarianship, 49(4), 102720–.
The following provides an overview of some specific types of prompts.
Direct prompting, or Zero-shot— a prompt which doesn't include an example. It might be phrased as a question.
One-shot prompting (prompt with an example)—a prompt which includes a clear, descriptive example of what you would like your output to be. Prompts can also be few-shot or multi-shot, providing more than one example.
Chain of thought, tree of thought prompting—a next-step iterative technique that directs the model to break down the logical connections leading to its output, allowing you to redirect, simplify, or add more complexity. Can be combined with few-shot prompting to improve results on complex tasks.
Metaprompting—a prompt that directs the AI model to help you craft a better prompt. Assign a training simulator persona to help guide you to create better input and thus better output.
Effective prompt crafting is a moving target due to ongoing developments and shifts in AI. A more "enduring and adaptive skill" is problem formulation (Acar, 2023). Solving problems seems to get all the attention but our ability to formulate problems is fundamental to solving problems and is key to designing a effective prompt and developing this underlying skill will make it more adaptable.
According to Oguz Acar, problem formulation has 4 key elements—
Problem diagnosis is about identifying the core problem for AI to solve. In other words, it concerns identifying the main objective you want generative AI to accomplish. This can be very simple but if your AI output is off-focus, start here as you analyze how you can change your input to get the output you want. More complex problem diagnosis might include process that digs into the "why" behind the problem.
Problem decomposition refers to breaking down complex or multifaceted problems into smaller, single-focus sub-problems. If you enter a complex prompt your AI response is likely to be too broad and generic to be useful. Think carefully through all the component parts that you can address them individually in your input.
Problem reframing involves changing the perspective from which a problem is viewed. This allows alternative interpretations. Reframe problems by taking on a different perspective or exploring analogies or parallel examples. If your AI output is missing important information, consider reframing your problem so your input provides this extra context.
Problem constraint focuses on determining the boundaries of a problem by defining input, process, and output restrictions of the solution. Some constraints are determined by the task at hand, but your creative goals can also be used to determine what constraints are needed.
Acar, O. A. (2023, June 6). AI prompt engineering isn’t the future. Harvard Business Review. https://hbr.org/2023/06/ai-prompt-engineering-isnt-the-future
Olympic College |
Privacy Policy
1600 Chester Ave. Bremerton WA 98312 | Information Desk: 360.475.7252
