Teachers' Cheat Sheet: Create Better Quizzes Faster with Chain of Density Prompts

Introduction

Teachers and lecturers, we all know making quality quizzes takes a lot of time. Between coming up with questions, writing them clearly, and ensuring they test the right material - it can feel like a never-ending task.

When asking a Large Language Model to generate a quiz from a text, the results can be disappointing. That's why I want to share how I've been using chain of density prompting to generate quiz questions much faster. This technique has been applied earlier to generate better summaries (see Forbes 2023 article), but it can be applied to quizzes as well.

Iterative Prompting

Chain of density prompting is a technique that utilizes large language models to refine prompts through multiple iterations. With each iteration, you provide more context and guidance to help the AI homing in on the type of output you need. This allows you to leverage the AI's capabilities while maintaining human oversight over the final product.

Here's an example of how I used it to create a quiz on parts of a plant:

Step 1: Initial prompt:

"Generate 10 multiple choice questions about parts of a plant"

Step 2: Feedback on output and refine prompt:

"The questions lacked detail and variety. Refine the prompt to be more specific."

Step 3: More detailed prompt:

"Generate 10 multiple choice questions about specific parts of a plant. Include questions about roots, stems, leaves, flowers, and fruits. Make sure each question has 4 answer options with only one being correct."

Step 4: Review new output and iterate:

"Great, the questions now have more depth. A few could be rewritten for clarity. Generate 5 additional questions based on this prompt."

After a few rounds, I ended up with 15 strong quiz questions covering the target material in good variety. The whole process took me around 30 minutes versus the hours it normally would!

I've found chain of density prompting hugely helpful for generating quizzes, assignments, and other assessments efficiently. Give it a try - your students will thank you for the engaging material, and you'll thank your past self for saving so much time! Let me know if you have any other questions.

Iterations in Single Prompt

Here is a single prompt with in built iteration, as tested on ChatGPT3.5. 

The output must be edited but then can be read into a platform like HelpTeaching.com to generate online or on paper quizzes.

Article:<past text or link to pdf or webpage> 

Tone is formal. Temperature is precise. 

You are an expert in learning theory, specialising in retrieval practice. You are also an experienced high school teacher.

Create a set of multiple-choice and fill-in-the blank questions for high school students with 4 answer options each, covering the whole content of the the Article above. You will generate increasingly precise questions from the above Article and more challenging answer options.

Always list the question and then answer option a), b), c) and d).  Always list the correct answer option with “Answer :” followed by the letter only.

Make sure that for each question the answer option is different from the previous question. For questions 3, 6, 9, 12, 15 and 18 include an answer option “None of these answers are correct”.  

Mix up the correct answer options so that each is different from the previous question. 

Repeat the following 4 steps 5 times.

Step 1: Identify 1 to 5 informative entities (";" delimited) from the Article which are missing from the previously generated questions. 

Step 2: Add 1 to 5 new, denser questions and answer options to the previous set of 10 questions, which cover every entity and detail from the previous question plus the Missing Entities.

Step 3: Now for each of the new set of questions and answer options, generate a total of 4 challenging answer options each of no more than 15 words. Including in these 4 options an option which is very close in meaning to the correct answer.  

Step 4: Combine these sets into 1 final set, renumbering them starting with number 1. In the final set, list the question and then answer options a), b), c) and d)

A Missing Entity is:

- Relevant: to the main question.

- Specific: descriptive yet concise (5 words or fewer)

- Novel: not in the previous question or answer option.

- Faithful: present in the Article.

- Anywhere: located anywhere in the Article.

Guidelines:

- The first set of questions should be long (1-2 sentences, about 30 words each) yet highly non-specific containing little information beyond the entities marked as missing. Use overly verbose language and fillers (e.g. "this question covers") to reach about 30 words.

- Make every word count: rewrite all previous questions to improve flow and make space for additional entities.

- Make space with fusion, compression and removal of uninformative phrases like "this questions covers".

- The questions and answer options should become highly dense and concise yet self-contained, e.g. easily understood without the Article.

- Missing entities can appear anywhere in the new questions

- Never drop the entities from the previous questions. If space can not be made, increase the number of questions.

Enjoy!

Check out my handy guide on Gumroad.