Image Prompt: A robot helping a math teacher with curriculum review
I've had ChatGPT Plus for about 3 months thus far, and I've had a couple spurts of inspiration come and go where I feel like its availability to my curiosity in the moment has been well worth the $60 I've spent on it ($20/month).
- I've gotten some useful R tips and code.
- I've summarized some articles I never got around to reading.
- I've loosely planned multi-stop vacations.
- I've had it explain complex topics to me like I was a 5th grader and create funny analogies.
- I've used it to outline and expand on (and shoot down) some dumb theories I've had.
- I've used it to do some arduous Excel data cleaning and formatting pretty easily.
- Those alone would be worth the $60 thus far on ChatGPT Plus.
But by far I've spent the most time and seen the most value from it helping me with being a math teacher.
Besides thinking of different ways to teach content, coming up with many, many different analogies and real-life examples for the content and skills is a big part of the job for math teachers. Often, the students don't fully understand a topic or skill --EVEN IF they know how to do the math correctly -- until they can relate it to something else outside of classroom - sports, TV/movies, things they've heard about around the dinner table, etc.
They can calculate slope but understand ramps.
They can calculate the roots of a quadratic but understand the flight of a baseball.
They can calculate exponential growth but understand how a virus spreads.
Coming up with many different examples has been very helpful thus far.
The biggest way GPT-4 and various plugins have helped is with lesson plan and curriculum review. Every year I try to do a little revise and review, but mostly this means big picture stuff (curriculum plan, etc.) during the summer and small detail stuff (actual problems in the lesson) during the year, and often the day of or night before. It's that darn procrastination I tell you.
Now with GPT-4 and the Wolfram plugin, I can get many different variations of a math problem instantly. The formatting will make it hard to easily incorporate them into my slides and worksheets but I'll figure that out.
With GPT-4 and the Wolfram plugin, and the prompt below that has been personalized to match the current lesson plan format, I can get lesson plans made for various topics on the spot.
Design an 80-minute Algebra 2 Honors lesson plan on the learning target. The plan should be divided into 5-minute segments and include activities that can be completed on a single sheet of paper. The plan should guide an experienced teacher in introducing the learning goal gradually, and include a list of 5 common student misconceptions. The plan should begin with a review of 3 prerequisite skills, and end with a list of 3 skills that could be learned next. The lesson should start with a 5-minute activity that reviews prerequisite skills, then balance instruction and practice, with practice time being three times instruction time. The lesson should end with a 5-minute summary and a 5-minute formative assessment. The plan should include 3 example problems in each section that align with the learning goal, and 3 assessment problems that could appear on a summative assessment. The total time should be 80 minutes. This lesson’s learning target is: XXX
With GPT-4, I've been able to easily and quickly (albeit roughly) think up a year's worth of curriculum planning for a new subject I'll be teaching in the fall, Statistics & Probability. Obviously I'll lean on previous curricula and established AP/IB topics and pacing, but GPT has been able to explain things in more detail that I'm not familiar with and even is helping me create a week-by-week breakdown. This could have all happened without GPT but it has been much easier and quicker with the help of AI.
More recently, with GPT-4 and the Wolfram + Link Reader plugins, I've been surprised and excited that I could review, summarize, and suggest improvements for my classroom presentation slides! This is a big deal, because even though it won't do a fantastic job, a good enough job is better than anything I could do in the same amount of time. To do so, this was the process I followed:
1. Go to Google Drive and make each presentation shareable with a link
2. Copy the link
3. Paste the link at the end of a customized prompt that reviewed, summarized, and suggested improvements (below)
4. Copy and paste the answers from GPT-4 in a Google Doc
The prompt I used was based on chain-of-thought, self-critique, and expert planning and can be found below. It was good, not great, and can be improved:
Let's go through the presentation linked below together.
First, start by summarizing the key points from the presentation. After that, proceed through the presentation and summarize the main learnings from it. Finally, provide an enticing overall summary, make it compelling for the reader, and suggest ways to improve the presentation.
Second, critique your own summary. Does it accurately reflect the main points of the presentation? Is it compelling enough to make someone want to read the presentation? What improvements can be made to the summary?
Third, imagine how experts in math pedagogy and teaching, like Jo Boaler and Sal Khan, would summarize and critique this linked presentation. What key points would they focus on? What aspects of the presentation might they suggest to improve for better understanding?
Finally, synthesize all of those points above into one coherent and beautiful answer. Then provide your 3 biggest improvements you would make to either the flow of information or to the presentation to improve the student's understanding of the topic and include 3 examples of each of the improvements.
The linked presentation is: XXX
Again, even though these summaries and answers weren't perfect, they were good. And I can work with good.
I took my Google Doc of all the answers and cleaned it up, removing the repeated AI responses, until I had a Doc just of topics covered and improvements suggested per lesson. Most of the improvements were centered around more visual aids and graphics, more interactive tools, and activities that would increase student participation.
Then I used this prompt below and customized it with each lesson's topics and improvements for an expanded list of improvements for the topics that GPT-4 pulled out of the presentation links I fed it. The prompt is:
You are an AI program that is designed to create the world’s most fun, interesting, and informative math lessons for high schoolers. The lessons are not in the exact style of but are similar to Richard Feynman’s famous physics lectures. Above all, these lessons should be relatable to the students' lives so that the students care more about them.
Imagine a lesson plan that covers these topics:
XXX
and give me at least 20 specific examples or math problems that would make this the world’s best math lesson and are similar to these:
XXX
A lot of the improvements suggested won't be used, but some will. And that's some more than I probably would have updated it with. So, all in all, it was a nice way to get an outsider's perspective on the topics covered and to generate some additional ideas for ways to communicate the learnings and skills.
I plan to review the answers, revise them, and to keep experimenting with different ways of getting better.
GPT-4 is not perfect. But neither am I.