Introducing: ChatGPT Edu-Mega-Prompts

How to combine the power of AI + learning science to improve your efficiency & effectiveness as an educator

🥡 Key Take-Aways This Week

The Bad News: Most AI technologies that have been built specifically for educators in the last few years and months imitate and threaten to spread the use of broken instructional practices (i.e. content + quiz).

The Good News: Armed with prompts which are carefully crafted to ask the right thing in the right way, educators can use AI like GPT3 to improve the effectiveness of their instructional practices.

In this week’s newsletter, I’ll introduce the concept of ChatGPT Edu-Mega-Prompts powered by DOMS™️ and share an example so you can try it for yourself.

Let’s go! 🤖

____

The world has been building AI tools tool for educators for over 30 years now. Over the last couple of weeks, I’ve been researching a rapidly growing list of these tools.

One thing I noticed very quickly is that the vast majority fall into two categories: content creation tools & quiz tools.

Yippity: one of a large number of new AI tools for educators

Around ~80% of the new “AI For Education” tools that have emerged in the last few weeks and months are AI-powered quiz generators.

A new generation of tools like Yippity, Quiz Gecko, Questgen & Kwizie use a combination of natural language processing (NLP) and machine learning (ML) to rapidly generate questions and answers based on information inputted from the user, e.g. text, uploaded document, URL.

There is no question that these tools make the process of creating a quiz faster. There’s also no doubt that it’s pretty exciting to paste in a lecture or a URL and generate a bespoke set of questions and answers in <2 seconds.

But there is danger in this delight.

Research has consistently shown that quizzes have limited pedagogical value. Here’s a snapshot of what we know:

• Quizzes do not effectively address misconceptions or alternative conceptual frameworks that students may have about a topic (Kuhn, 2002; Windschitl, 2002).

• Quizzes primarily focus on testing immediate recall rather than establishing long term understanding (Roediger & Karpicke, 2006).

• Quizzes are often used as an assessment of learning rather than a tool for learning, and thus may not effectively support the learning process (Black & Wiliam, 1998).

• Quizzes often only assess surface level understanding, rather than deeper conceptual understanding (Biggs, Kember, & Leung, 2001).

In short: with the arrival of a new generation of GPT3-powered tools like quiz generators (and, along with them, flash card generators and content creation tools) we witness a too-familiar trend of new technology being used to imitate, accelerate and proliferate established bad practices, rather than improving on them.

But it doesn’t have to be this way.

The Power of AI + Learning Science: A Case Study

So, what might a better use of AI look like in education? Let’s explore this question through the example of an instructional strategy known as Undoing.

One of the biggest risks of quizzes is that they do not effectively challenge commonly mistaken foundational ideas and assumptions, which is crucial for meaningful recall and the processing of accurate foundational information (Kuhn, 2002; Windschitl).

One well-known instructional strategy for addressing this risk is Undoing.

Undoing is an instructional strategy which surfaces, challenges and reframes commonly mistaken foundational ideas and assumptions - a crucial step in ensuring that learners can meaningfully recall and process accurate foundational information.

So, I decided to try to teach ChatGPT what an Undoing activity is, then ask it to generate some activities for me.

Here’s how it went.

The Undoing Edu-Mega-Prompt

ChatGPT’s outputs are only as powerful as the quality of its inputs. So, my first job was to make sure my prompt was optimised for the best quality output, both in terms of its content and its structure.

Through a process of experimentation and a heap of learning from masters in the prompt-writing space like Rob Lennon (I heartily recommend following Rob if you aren’t already) I landed on a prompt structure which generated some consistently reliable and impressive results.

The prompt had seven key characteristics:

• A defined role for AI: the more context ChatGPT has about the role it’s playing, the better. At the top of your prompt, tell it who it is and what its role is to help it to understand what information matters most.

• Information on the target learners’ demographics & ZPD, e.g. age, subject area, level of understanding & ability: by giving ChatGPT information about your learners, it is better able to generate outputs which are relevant to them.

• Domain expertise, i.e. instructional strategy info: by sharing your domain expertise with ChatGPY - in this case what Undoing is & how it differs from quizzes - you get on the same page and deter ChatGPT from defaulting to the most common rather than most effective approaches. E.g. without the Undoing intel, ChatGPT will default to basic quiz similar to a tool like Yippity.

• A title & topic: Chat GPT likes structure. By adding a title and topic, we draw ChatGPT’s focus to what sorts of information matters most.

• An example: examples are a great way of creating a shared understanding of what good looks like both in terms of structure and content.

• A clearly specified task: provide clear instructions on what you want ChatGPT to do and how.

• A clearly defined output: tell ChatGPT both what you want and how you want it to present it.

The results were pretty impressive: in <2 seconds ChatGPT generated a robust Undoing activity, explanatory + discrepancy feedback for each potential response and a suggested resource.

As is always the case, ChatGPT is your assistant. If you’re not happy with the result, you can edit and refine it using your expertise, either alone or through further conversation with ChatGPT.

For example, once the first response is generated, you can ask ChatGPT to make the activity more or less complex, to change the scenario and/or suggest more or different resources - the options are endless.

Interestingly, once I’d entered the prompt and generated the first response above, I was able to build on my and ChatGPT’s shared understanding of Undoing activities to generate an infinite number of these activities on a variety of topics aimed at variety of learner types. Here’s the follow-up prompt I used:

Using the information above, generate an Undoing activity which helps [learner type, e.g. university level learners, five year olds with limited grasp of English] to correctly understand the foundational knowledge relating to [concept or skill, e.g. Algebraic theory, creativity]. The activity should include some common misconceptions about Algebraic theory and a clear, short and simple explanation of the conceptual difference between the correct answer and the misconceptions.

Key Take Aways

Armed with an an understanding of a) how to talk ChatGPT’s language and b) the science of learning, AI can be leveraged to not only improve our efficiency as educators but also - and critically - our effectiveness.

Unlike a basic quiz, the Undoing activity that I co-created with ChatGPT:

• Effectively addresses misconceptions & alternative conceptual frameworks that students may have about foundational concepts.

• Builds a strong foundation for establishing long term, meaningful understanding by introducing context and rationale.

• Quizzes are often used as an assessment of learning rather than Operates as a tool for learning, rather than merely an assessment.

• Supports deeper conceptual understanding, rather than surface level remembering.

If you do one thing to future-proof your role and optimise your effectiveness, deep dive the science of learning.

Want to dive and in learn more? Great! You can:

• Have a go! I’ve pasted the Undoing edu-mega-prompt below so you can dive right in and try it for yourself right now.

• Sign up for my AI for Educators course, where I’ll share a collection of DOMS™️ Edu-Mega-Prompts to 100X your efficiency & effectiveness.

• Subscribe to my Learning Science Digest to access easy to apply summaries of the latest learning science research so you can build your own evidence-based Edu-Mega-Prompts.

• Join me on my Learning Science Bootcamp: a 4-week cohort baed course to hone your learning science skills.

• Access free guides and other courses on my website.

Happy designing,
Phil 👋

📚 References:

• Biggs, J., Kember, D., & Leung, D. Y. P. (2001). The revised two-factor Study Process Questionnaire: R-SPQ-2F. British Journal of Educational Psychology, 71(1), 133-149.

• Black, P., & Wiliam, D. (1998). Inside the black box: Raising standards through classroom assessment. Phi Delta Kappa International.

• Kuhn, D. (2002). The skills of argument. Cambridge University Press.

• Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249-255.

• Windschitl, M. (2002). When “why” doesn’t always lead to “how”: The implications of alternative frameworks for teaching and learning. Journal of Research in Science Teaching, 39(7), 671-701.

---

The DOMS™️ Edu-Mega-Prompt: Undoing Activities

Copy and paste the Undoing Edu-Mega-Prompt + follow-up prompt below to generate robust Undoing activities for you and your learners.

You are an expert teacher of [enter subject, topic, concept, skill etc]. You teach students with [enter learner demographics, psychographics & ZPD information, e.g. age, what they know already, levels of confidence and motivation].

Your aim is to ensure that your students have a reliable and shared understanding of the foundational concept: [enter subject, topic, concept, skill etc]

Without the ability to recall & process accurate foundational information, the learning process is futile. You want to support your learners to recall and process accurate foundational information without using quizzes which are proven to only enable short term recall rather than long term memorisation.

Instead, you want to use two instructional strategies proven to increased the effective memorisation of foundational knowledge. The first strategy you will use is known as Undoing.

Topic: Undoing Instructional Strategy

Research shows that learners learn better & more quickly when the learning experience intentionally challenges commonly mistaken foundational ideas and assumptions and encourages new ways of thinking. The highest impact learning experiences are designed to identify and correct foundational misconceptions before memorisation occurs.

Undoing activities are particularly helpful when used at the start of the learning process by identifying and then correcting gaps in knowledge, misconceptions and misplaced reasoning. Undoing activities often ask learners to predict the outcome of a scenario.

An example of a well formed undoing activity goes as follows:

Sharna is watching a candle burn. Once it’s burned down, she wonders what happened to the wax. She has 4 ideas - which do you agree with most, and why?
A. XXXX - common misconception
B. XXXX - a common misconception
C. XXXX - the correct answer
D. XXXX - a common misconception

Your task: Using the information above, generate an undoing activity which helps the learners described above to correctly understand core foundational knowledge relating to [enter topic, concept, skill etc].

Write feedback for each potential response. The feedback should be succinct and explain the difference between the student's response and the correct answer. It should replace the misconception with an alternative conception, explaining what the alternative, correct way of thinking about the scenario is and how it differs from the student's misconception.

Finally, suggest a resource for the defined learner type to look at which will help them to learn the correct version of the foundational knowledge.

Once you’ve generated your first activity, in the same ChatGPT conversation use this follow-up prompt to generate more!

Using the information shared above, generate an Undoing activity which helps [learner type, e.g. university level learners, five year olds] to correctly understand the foundational knowledge relating to [concept or skill, e.g. Algebraic theory]. The activity should include some common misconceptions about Algebraic theory and a clear, short and simple explanation of the conceptual difference between the correct answer and the misconceptions.