Article
Prompt Engineering: How to Write Killer Generative Prompts for LLMs
May 25, 2024
15 Minute Read
Introduction
In today’s rapidly evolving educational landscape, the demand for innovative learning design isn’t something that’s “nice to have.” It’s a necessity. You have to keep pace with emerging technologies and methodologies. While it can feel like an endless barrage of waves cascading over your industry, if you don’t learn to surf, it’s lights out for your organization.
Enter prompt engineering—a concept that is reshaping the way we approach learning content creation by leveraging the power of generative AI bots such as ChatGPT. While the debate rages on about the proper use of this tech in the educational sphere, that’s the focus of this blog.
Instead, I will walk you through what prompt engineering is, how it works, and how to master it when your organization decides to embrace AI for content generation. Because believe me, that’s not a question of “if,” but a question of “when.”
Prompt engineering involves crafting precise, targeted instructions or prompts to guide these AI models in generating tailored learning content. By providing clear and strategic prompts, educators and instructional designers can harness the capabilities of AI to create personalized, high-quality learning experiences at scale.
In this article, ChatGPT and I are writing to you from a place of experience when it comes to prompt engineering. The findings, recommendations, and guidelines we’ll lay out for you are learnings I gained after completing content development for three online courses. To build these courses, my learning design team used ChatGPT to develop over 250 different course content deliverables, including talking head video scripts, case study video scripts, custom readings, formative and summative quiz banks, hands-on projects, learning objectives, and course outlines.
Introduction
Prompt engineering involves crafting precise, targeted instructions or prompts to guide these AI models in generating tailored learning content. By providing clear and strategic prompts, educators and instructional designers can harness the capabilities of AI to create personalized, high-quality learning experiences at scale.
In this article, ChatGPT and I are writing to you from a place of experience when it comes to prompt engineering. The findings, recommendations, and guidelines we’ll lay out for you are learnings I gained after completing content development for three online courses. To build these courses, my learning design team used ChatGPT to develop over 250 different course content deliverables, including talking head video scripts, case study video scripts, custom readings, formative and summative quiz banks, hands-on projects, learning objectives, and course outlines.
Prompt engineering involves crafting precise, targeted instructions or prompts to guide these AI models in generating tailored learning content. By providing clear and strategic prompts, educators and instructional designers can harness the capabilities of AI to create personalized, high-quality learning experiences at scale.
In this article, ChatGPT and I are writing to you from a place of experience when it comes to prompt engineering. The findings, recommendations, and guidelines we’ll lay out for you are learnings I gained after completing content development for three online courses. To build these courses, my learning design team used ChatGPT to develop over 250 different course content deliverables, including talking head video scripts, case study video scripts, custom readings, formative and summative quiz banks, hands-on projects, learning objectives, and course outlines.
Understanding ChatGPT: The Engine Behind Prompt Engineering
At the heart of prompt engineering lies ChatGPT, a cutting-edge AI model that has revolutionized the way we approach educational content generation. Developed by OpenAI, ChatGPT is part of a new generation of large language models that excel at understanding and generating human-like text based on the input provided to them.
Overview of ChatGPT
ChatGPT is built upon a deep learning architecture known as the transformer model, which allows it to understand and generate text with remarkable accuracy and coherence. With its vast repository of knowledge and linguistic capabilities, ChatGPT has become a powerful tool for educators and learning professionals seeking to create diverse and engaging learning content. But it’s this vast repository that creates obstacles when we utilize ChatGPT for content generation. More on that to come in the following sections.
How ChatGPT Functions
ChatGPT operates by analyzing and processing input prompts provided by users, which can range from simple queries – What’s the current population of Mesa, Arizona? – to more complex instructions – I need you to write a script for an instructional video about the integumentary system. Once presented with a prompt, ChatGPT employs its intricate neural network architecture to generate contextually relevant and coherent text, drawing upon its understanding of language and vast training data.
One of the main issues or problems I’ve seen in the implementation of utilizing ChatGPT for educational content generation is the users limited understanding of how to “talk” to the bot. Because we’ve all been told this is a “large language model” that understands natural language, we assume – erroneously – that we are actually talking to a person when we give the bot commands and ask it to create something. That’s not the case. While a faculty member or instructional designer may think the instructions they are giving to the bot are simple, they aren’t. In the sections to come, I’ll explain why it’s important to follow these prompt engineering steps, even for generation tasks that we humans think of as “simple.”
The CRAFT Principle: Crafting Effective Prompts for Optimal Results
Prompt engineering is not merely about providing instructions to an AI model like ChatGPT; it’s about crafting prompts with precision and purpose. Enter the CRAFT principle,a foundational framework for designing prompts that yield optimal results in content generation.
What is CRAFT?
The CRAFT principle outlines five key elements that are essential for crafting (pun intended) effective prompts:
Context | Role | Action | Format | Target Audience
Each component plays a crucial role in shaping the direction and quality of the generated content, ensuring that it aligns with the intended objectives and audience needs.
Craft
The prompt should provide relevant context to ChatGPT, setting the stage for the content generation process and helping the AI model understand the broader context of the task at hand.
This part of the prompt-engineering process is more important than you might think. Remember, you’re having a conversation with a bot whose reference library is the entire World Wide Web! If you ask it to write a multiple choice question about pharmacology, it won’t just reference publisher textbooks, focused on the specific industry you are writing for. It will process a literal world of pharmacology information to give you what it deems the most comprehensive generated content. That’s not what you want. You have to tell the bot specifically who you are and what you’re trying to achieve.
Here is an example of a context prompt; in fact, it’s the one I used to help me write this article:
“CONTEXT:
I’m a Director of learning design in the United States who needs to write a blog article about prompt engineering.
This blog article will be read by all stakeholders at my company and will be published on our company website, where future customers and competitors will read it.
The blog needs to establish my company as industry leaders in learning design and the utilization of AI generative bots.
Please confirm that you understand CONTEXT by responding “read” and we will proceed to the next step.”
*Crafty Note: Notice that this initial context prompt is part of an instruction set with steps. It is extremely important that you break your prompts into steps rather than writing them all out in one big, long prompt. More about why in the following sections.
Role
The prompt should define the role or perspective from which the content should be generated, guiding ChatGPT in adopting the appropriate voice or stance for the intended audience.
Remember that big, vast ocean of information the bot is pulling from? Well, “who” you want the bot to be is swimming in that same water. The role – or what kind of writer you want the bot to be – differs from user to user and industry to industry. Here is the role I gave ChatGPT to write this article:
“ROLE:
You are an expert blog writer who is very well versed in prompt crafting, the CRAFT prompt principle, and using ChatGPT for generating learning content for universities, corporations, and workforce teams.
When you write, you follow these core principles:
-You use an informative, professional tone but also implement plain-language
-You are confident and engaging
Please confirm that you understand ROLE by responding “read” and we will proceed to the next step.”
Action
The prompt should specify the action or task that ChatGPT is expected to perform, outlining the desired outcome or objective of the content generation process.
The Action Step of the prompt is the meat of your instruction set. This is where you tell the bot what you want it to do. While it might seem straightforward and basic, this is the part where you have to remind yourself that you are talking to a bot and not a person. Experimenting with this step in the process is where I learned about ChatGPT’s inability to hold priorities. More on that in the following sections. For now, just know that this is the reason I chose to have the bot write an outline for me rather than writing the article in one go:
“ACTION:
Based on the CONTEXT and ROLE I’ve provided, I want you to write an outline for me of the blog article. Here are the topics I want you to include in the outline:
Using prompts to generate content using ChatGPT – an overview of what ChatGPT is and how large language models work.
Explanation of the CRAFT principle for prompt engineering
Limitations of ChatGPT when it comes to prioritizing directions given in prompts
Best practices for prompt engineering using CRAFT
Please confirm that you understand ACTION by responding “read” and we will proceed to the next step.”
Format
The prompt should tell ChatGPT how you want it to format whatever it’s generating. Is it writing a script for you? Tell it to specify which character is talking. Is it writing exam questions that need general or specific feedback? Give it an example to work off of. Here’s what I used for this article:
“FORMAT:
When you write the outline, please use headers and subheaders so that I know what content goes into each paragraph of the final blog.
Please confirm that you understand FORMAT by responding “read” and we will proceed to the next step.”
Target Audience
The prompt should consider the characteristics and preferences of the target audience, tailoring the content generation process to meet the audience’s unique needs and expectations. This may seem like it’s already covered in context, and sometimes it is, but not in the example I gave you. For my article writing prompt steps, the target audience is actually me and the bot, not my company stakeholders, potential clients, or our competitors.
“TARGET AUDIENCE:
This outline will be used by me, and I will use it in prompts that I will give you to help me write the final blog.
Please confirm that you understand TARGET AUDIENCE by responding “read” and we will proceed to the next step.”
Importance of Adhering to the CRAFT Principle
Adhering to the CRAFT principle is paramount for maximizing the effectiveness of prompts in guiding ChatGPT. By incorporating Context, Role, Action, Format, and Target Audience into the prompt design, educators, instructional designers, and instructors can:
- Generate high-quality and relevant content in the first round of prompting by providing context and defining roles;
- Ensure clarity and alignment with specific objectives;
- Save time on reformatting tasks;
- Tailor content generation to the preferences and characteristics of the target audience, maximizing engagement and effectiveness.
Recognizing Limitations: Addressing ChatGPT’s Constraints in Prompt Response
While ChatGPT represents a significant advancement in AI technology and content generation, it is not without its limitations. Understanding and mitigating these constraints is crucial for maximizing its effectiveness in prompt-driven content generation.
One of the primary limitations of ChatGPT lies in its inability to prioritize directions given in prompts. Unlike humans, who can intuitively understand the most important aspects of a task or instruction, ChatGPT relies solely on the input provided to it. This can lead to challenges in discerning the key objectives or requirements of a prompt, resulting in content that may not fully align with the intended goals.
Here’s an example:
Imagine that you’re a doctor at a local hospital who’s responsible for mentoring 1st-year residents. You need to write a 50-question exam bank, and the questions need to follow these specific guidelines:
- They need to test the information presented in four different training modules
- They need to test the test-taker’s ability to recognize different symptoms
- They need to align to a specific enabling outcome
- They need to be written in a specific format
You’ve heard your colleagues and friends talking about how quickly ChatGPT can do tasks like this, so you figure you’ll give it a try.
So you sit down with the ChatGPT bot (and because you didn’t read this article) you think: “Ok, let me just put these 4 guidelines into the bot, upload the training modules I want it to reference, and think, BOOM! I can generate this exam bank in under a minute!”
But that’s not how it works. Remember, you aren’t writing an email of instructions to a human; you’re basically engineering, or programming, the bot to give you the desired output. If you don’t break this task down using the CRAFT prompt-creation workflow, the bot’s brain will break.
That means the resulting exam bank that the bot gives you won’t be what you want. There will be errors in the generation because of the bot’s inability to prioritize your instructions. For example, while it might be able to align question 3 to your specific enabling outcome, maybe it can’t find content that it feels aligns with it in your training modules. What will it do? To give you what it considers the best results, it will go out to that big vast ocean that is the internet and find content that “does” align with your outcome, and it will write the question about that.
I know what you’re thinking…what!? Are you telling me the bot will disregard certain instructions I give it and follow others based on what it thinks is appropriate? The answer is yes. Remember… it’s not a person who can understand the priorities listed as equally important. It’s a super fast processor.
That’s why you need to guide it, very specifically, using CRAFT and creating as many prompts as you need to so that it can complete one priority at a time. So, instead of writing one prompt that lists the four criteria you want it to follow, you need to create multiple prompts for the context, the role, and several prompts for the action step to allow it to focus on one priority at a time.
For example, you might review the content in your training modules and figure out which content aligns with the enabling outcome you are assessing. That section of the training module then becomes a database that you give to the bot to use as you walk it through multiple action prompts.
Best Practices for Prompt Engineering Using CRAFT
Crafting prompts that adhere to the CRAFT principle is essential for guiding ChatGPT effectively in content generation.
Detailed Guidelines for Crafting Prompts
- Even if you think you’re giving the bot a “simple” task, force yourself to write out the CRAFT steps. You’ll find the more detail you give the bot, the better your initial generations will be.
- If you run into the “priority confusion” obstacle, consider the use of databases or datasets, which are additional prompts used in the action step to lead the bot through the actions you need it to take. I use these in my prompts to point the bot exactly at the content I want it to reference for given tasks.
- Remember that the bot is not actually all that intelligent. Yes, it can process information and generate content at incredible speeds, but that doesn’t mean it will process it the way you want or generate what you want. When I prompt-engineer, I think of the bot like a very mathematically talented first-grader. If I want it to write an essay, I make it start with an outline. If I need it to generate tons of questions, I break it down into datasets, telling it which questions come from where.
- Keep the length limitation in mind. At the time of this article’s writing, ChatGPT still had length restrictions. As of now, the bot does not understand prompts that tell it how long something should be. It doesn’t understand word count, characters, or tokens.
Go For It
Prompt engineering, powered by AI generative bots like ChatGPT, offers a gateway to unprecedented levels of efficiency and innovation. Now that you have a good foundation to start from, go for it! Embrace the change and try out your new prompt-engineering skills. Navigating the ever-changing landscape of education and technology is a challenge, but embracing these new methodologies and tools will put you ahead of the curve.
If you stick to the CRAFT principle and implement best practices in prompt construction, you can harness the full potential of AI generative bots. Take the leap into the waters of change and watch your efficiency and bot-generated quality soar!
About the Author
Authors of our articles and whitepapers are proven experts in their field, and thought-leaders in their industries. At LearningMate, we pride ourselves on finding talented professionals and enabling our team to reach new heights in their fields of expertise.
Related Insights
From industry research to thought-provoking articles and podcasts, we bring you insights into our fast-evolving world of education technology. Come learn and evolve with us.
High-quality, university-approved content is expensive and time-consuming, and once paired with a delivery platform, it becomes even more expensive and time-consuming to move and re-purpose. See how we’re rethinking content architecture.
Read More
The business of academic publishing has changed tremendously in the last 20 years. In this episode, we welcome industry veteran Shannon Gattens to provide insight into how publishers are working to keep up with the way education is consumed by learners.
Read More
84% of teachers, administrators, and students think that shifting to digital can solve the challenges of education, with 82% agreeing that digital is the future.
Read MoreLet's have a conversation...
Have Questions? Curious to learn how we can help? Send us a note to start a conversation.