Curriculum Research Findings
Prompt Engineering Best Practices
- Use clear language: Get rid of unnecessary information, jargon, confusing phrases, or mistakes.
- Define the purpose of the prompt using concepts like:
- Tone: How do you want the output to sound? Funny? Professional?
- Format: How do you want the output structured? Bullet list? Paragraph? An essay?
- Audience: Who is this for? Do you want something for children? Beginners? Experts?
- Include important context: If there are specific limitations or requirements, make them clear in your prompt.
- Provide examples
‼️ ChatGPT will confidently share incorrect information, so it’s important to have an SME double-check the accuracy of the content.
✅ DO use ChatGPT for creating first drafts of content.
❌ If you ask for a word count, grade level, etc, ChatGPT may undershoot what you’re asking for, so double-check.
💭 It can be hard to get ChatGPT to copy a specific tone. Using examples rather than adjectives in prompts can sometimes be more useful.
Unfortunately, ChatGPT-generated articles often sounded like ChatGPT.
✅ Can use ChatGPT to give an initial check content or do quick revisions.
“Are there any inaccuracies in this paragraph: [TEXT]
💭 It can be difficult to write a prompt that exactly communicates who your audience is, and sometimes it’s easier to take ChatGPT’s first few attempts and have a human consolidate them.
💭 It seems to have an easier time with well-defined technical concepts like programming languages. It can be more difficult to create a syllabus for open-ended conceptual topics that require different contexts.
Content Item Outlines
✅ ChatGPT is generally useful for creating a quick outline for just about any level of content: lesson or article.
🧐 Generally, it is helpful to provide information about the learner. For Codecademy, this leads to prompts like:
Create an outline for a __ on using for _. The learner is [AUDIENCE]. (For more advanced content, add in the following) The learner has already learned , , and __.
For example, a prompt to write a course on using Golang to build APIs might look like:
Write an outline for an introductory course on using Golang to build WebApis. The target audience is new to coding but has already taken a course on how to use basic Go Syntax.
Or a Tech and Dev Example might be:
Write an outline for an article on using Go to build webapis. The target audience is a manager who wants to know whether Go is the right choice for their team. The manager does not have a lot of technical knowledge.
💭 Remember, these are starting points and it’s necessary to still check the structure of existing content on the subject.
✅ ChatGPT is pretty good at coming up with ideas for quizzes. It’s good at creating multiple-choice questions and coming up with extra distractor options for questions.
🧐 Find some content that encapsulates a learning standard, then consider a prompt like:
“Write 5 multiple-choice questions for the following: [LEARNING STANDARD] Provide a right answer and three wrong answers. Provide an explanation for why each wrong answer is wrong. Have the first answer be the correct answer.”
✅ Ask for a larger amount of assessments than you need to increase the chance you will have enough usable assessments.
❌ So far we have been unable to get ChatGPT to generate fill-in-the-blank questions.
💭 Provide the context! Train ChatGPT with assessment rubric(s) and any additional information to get the best results.
❌ When creating articles, people were often able to distinguish what was written by ChatGPT easily. They described the content as vague and bland.
✅ Some of the best results were achieved by writing parts of an article or lesson, then asking ChatGPT to complete it.
🧐 For example, I would:
Write several exercises of a lesson
Insert that content into ChatGPT
Ask ChatGPT to write a similar section on the content yet to be written
🧐 Giving an explicit outline for lessons can also be useful:
Write a lesson about how Generative AI works. It should have 9 sections. Section 1 is an introduction to generative AI. Section 2 is about training data in generative AI. … Section 9 is a review of the lesson. Each section should be less than 250 words. This lesson is for complete beginners who know nothing about AI.
💭 The exercises looked so similar that it was clearly formulaic, so further edits are needed so that the content is actually engaging or substantive to the reader.
Introduction and Review Sections:
✅ ChatGPT is generally well suited to summarize text. The resulting content may need tweaks but often provides a useful start to an introduction or review (e.g. in a Lesson or Article).
🧐 Use prompts like:
“Write an introduction to the following text: [EXAMPLE]”
“Write a summary of the following text: [EXAMPLE]”
“Write the key takeaways from the following text: [EXAMPLE]”
“Write a conclusion to the following text: [EXAMPLE]”
✅ ChatGPT is pretty great at generating different kinds of code snippets.
💭 Run the code and ensure it actually works (ChatGPT is known for sometimes producing buggy code). Simple code snippets generally work well.
❓ Keep in mind ChatGPT can also explain how code snippets work. This can be useful in writing lessons or project content.