WN #12

If You Can Teach, You Can Prompt

Author

Kyler McClary
December 20, 2023

If You Can Teach, You Can Prompt

When generative AI burst onto the scene a year ago, one of the buzzy terms to emerge along with it was “prompt engineering,” the art of carefully crafting inputs to optimize the responses generated by an AI language model. And while I initially thought that prompt engineering was a skill reserved for the CompSci majors and coders of the world, I quickly learned that not only can anyone be an adequate prompt engineer with just a little bit of practice, but teachers are particularly well-equipped to write good prompts. That’s because, as it turns out, working with AI can be very similar to working with a student. 

Don’t just take it from me, take it from OpenAI itself. The creators of ChatGPT recently released their own prompt engineering guide based around six key strategies, all of which should sound very familiar to teachers:

  1. Write Clear Instructions

    OpenAI suggests being detailed and specific, including about how the model should respond, the steps it should take, and what the output should look like. Try to clearly indicate distinct parts of your instructions, and (surprise, surprise!) adding an example or two can be really beneficial.

  2. Provide Reference Text

    If you assign students questions with no reference text, you’re leaving it open for them to pull from sources from anywhere. Just as with students, it can be helpful to narrow an AI model’s focus by giving it a reference text to work from, as well as have it answer with citations from that text.

  3. Split Complex Tasks into Simpler Subtasks

    Chunking! Good for students, good for AI models. OpenAI suggests breaking things up into more manageable pieces. This includes long dialogues with lots of text. An AI model’s memory can only process and hold onto so much before it eventually starts forgetting things from earlier as it makes room for new information. Condensing prior content through summarization/review can help consolidate it into a base on which to build going forward. But we already knew that, right?

  4. Give the Model Time to "Think"

    Fastest isn’t always best. Have the model show its work or explain its reasoning process before rushing to a conclusion. You can also ask it to take time to double-check its work. It can often catch its own errors and omissions if you give it time to do so.

  5. Use External Tools

    Teachers have lots of external tools at their disposal that they can integrate into their existing systems to maximize efficiency, productivity, and desired outcomes. Knowing what’s available and why, when, and how to use it can help improve your experience. The same holds true with AI models, which have different tools that can be integrated to help perform specific tasks like retrieving information from a document, writing code, or generating images.

  6. Test Changes Systematically

    There will be plenty of experimentation and evaluation happening as we start becoming more and more familiar with AI models. When we try new things, collecting and analyzing data is the best way to make sure our changes are for the better.

When it comes to AI, prompt engineering isn’t everything. You may find that a 5-word prompt gives you just what you were looking for, or that you are able to get a desirable output through iterative dialogue and addressing needs as they arise, rather than crafting a single, complex prompt, or even that your precisely crafted prompt that works for you may not be ideal for your colleague (or another AI model). But there are little things that do help models perform better, and, conveniently for us, a lot of those are the same little things that help students perform better. Teachers already have a leg up in this whole prompt engineering thing. You’ve actually been building these skills for years.

If you want to read OpenAI’s full prompt engineering guide, complete with explanations, examples, and some technical jargon, you can find it here. Note that this is not meant to be an exhaustive list, you may find other helpful strategies elsewhere, or even discover some of your own!

California Will Upskill its Workforce for AI

California governor Gavin Newsom wants to provide generative AI education and training for the state’s workforce and lead the way in helping employees leverage the benefits of GenAI while also understanding and mitigating the risks, according to a recent report. California Congressman Eric Swalwell also called for widespread AI training earlier this year. This should further signal to schools the importance of preparing students for the next generation of work.

Benefits vs. Risks in Schools

On that note, here is a collection of short essays addressing the question of how schools can harness the power of AI while mitigating its risks. The essays were submitted as part of the Fordham Institute’s annual competition to “generate substantive conversation around key issues in education reform.”

From the Classroom…

The most popular links in my newsletters are ones where educators share how they have been using AI in their schools and classrooms. This week, high school English teacher Mary Martin shares 5 ways she’s been using it in hers.

📌 Gemini is Here

Google released its hyped new AI model, Gemini (at least in part). You can try out a version of Gemini by using Google’s free AI chatbot, Bard. You can also use Gemini in Google’s new NotebookLM tool which is now available to users in the US. NotebookLM allows you to upload your own documents and materials and will help you with the process of reading, summarizing, note-taking, and transforming notes into structured writing (read about it here).

📌 A Chatbot for Student Journalists

Murrow is a free AI journalistic writing tutor for middle and high school students named after the acclaimed American journalist, Edward Murrow. It is designed to help students with all parts of the journalistic process, including adherence to journalistic standards and ethics, without actually writing the stories for them.

📌 Updates from Meta and OpenAI

Meta has a new image generator that is free to use as long as you have a Meta account (Facebook/Instagram). And good news if you have been waiting to sign up for a ChatGPT Plus account, as they re-enabled new subscriptions after a brief pause. As a reminder, ChatGPT Plus uses an upgraded model from the free version, can read files, view and generate images, connect to web search and other external plugins, and allows for the creation of custom GPTs.

Creative Ways to Engage with Key Concepts

I love using AI to generate creative little ways to engage with key concepts. There are infinite possibilities, but here are a couple of fun examples:

“Create an acronym for ________ in the form of a poem to help students better understand the concept. Each letter should get its own line so that the word is spelled out vertically on the left side in bold.”

Here’s another:

“Create 3 haikus, each about a different Enlightenment thinker, without giving away who is who. Make each distinctive enough so that students can guess who each poem represents.”

Adding Personalized Reflection Questions

Using AI to help provide feedback on student work? Try adding this to the end of your prompt/instructions to leverage personalized opportunities for reflection:

"End feedback by asking questions or providing prompts that lead students to think critically about their own work."

For simplicity's sake, you can also skip the rest of the feedback and just have it ask the submission-specific reflection questions. Have students respond to the questions, then revise their submission based on their responses.

That’s all for this week! If you appreciate the content in this newsletter, consider subscribing for free, or sharing with people in your network who may find value in it. If you are looking for more, feel free to check out the archive to catch up on any editions you may have missed.