Welcome to Teacher’s AIed: the newsletter about AI in the K12 Classroom.

How AI will affect K12 Classrooms is complex. Each week, we curate knowledge for educators about the strengths, weaknesses, opportunities, and threats of AI and K12 education.

This week’s edition is a second installment on leveraging the opportunities of AI to support lesson planning.

When I taught elementary school, we had a unique learning block called “Power Hour.” During this hour-long block, all the students in a given grade would switch classes to learn literacy skills alongside their peers with a similar literacy level. Within each differentiated class, we broke students into smaller groups so that they were learning discrete skills alongside their peers.

Each quarter, we would re-assess students and re-create groups. We took an entire teacher workday to group students by classroom and then by discrete skills within the classroom. I spent hours creating bespoke lesson plans for my small groups and analyzing their data each week.

With the affordances of AI, the hundreds of hours we spent each year to differentiate and plan lessons for our small groups would be cut to a few hours each quarter.

Today, we will focus on strategies for Tier 2 teachers, namely those who are AI-competent and want to leverage AI tools to take their differentiation to the next level.

Too Long Didn’t Read; TLDR

For those of you who are AI-proficient, we will walk through the data security steps you should take to protect your student's personally identifiable information, then explore a surprisingly simple process to create small groups from assessment data, understand the unique needs of each small group, and finally create a week’s worth of differentiated literacy lessons for one small group in a fifth-grade classroom.

Getting Started

Before we jump in, I want to establish a few assumptions.

First, for this demonstration, I assume that this classroom uses the Fountas and Pinnell reading system to differentiate students’ reading levels. I am not advocating for this reading system as the best, but it is the one I am most familiar with. Using this system, we can expect that most students will either “meet expectations” or “approach expectations,” a smaller percentage of students will “exceed expectations,” and an even smaller percentage “does not meet expectations.”

Since we are approaching the middle of the year, my students’ reading levels are based on the “1 Interval of the Year” or the second testing cycle of the year. As you’ll see below, most of my students are on track, above expectations, or nearly on track, with a T, S, or U reading level. I have a handful of students who need explicit interventions with R-level reading.

With these expectations in mind, meet my class!

If you are pop culture savvy, you should recognize at least a handful of students on my roster. I purposely chose these names to make a quick point about data security.

After using an anonymization model, Aloy Huntress is 1, and Walter White is 30.

Take a look at the new anonymization data below.

Feeding the LLM

Now, it is time to group my students. I used Advanced Data Analysis, formerly called Code Interpreter, from OpenAI, the creators of ChatGPT. Advanced Data Analysis is a mode in ChatGPT4 that allows users to upload data files, like Excel workbooks. Advanced Data Analysis will then analyze the data set and generate insights from user prompts. To use Advanced Data Analysis, you will need to purchase a subscription to ChatGPT4.

In my prompt, I asked,

I was particularly impressed by the fact that Advanced Data Analysis took my input and rewrote the steps in more straightforward English.

It then grouped my students by their reading levels.

After grouping my students by reading level, I moved the inputs to ChatGPT (with a simple copy/paste function) to ask it to provide observations and recommendations for my reading groups. The results were impressive. Check out its summary for my groups.

The level of detail it gives me on my students is impressive. Writing a quick summary of my students and their needs would have taken me anywhere from an hour to an hour and a half. This information was useful when justifying my groupings to my administrators or talking to a parent about their student’s reading needs.

I then asked ChatGPT to give me five learning objectives for each reading group. It gave me thoughtful and varied objectives. When I was teaching, I would take hours to scope out the quarterly learning objectives for each group. I used these objectives to map students’ progress toward their next reading target and my weekly lesson plan.

Creating specific learning objectives or targets for each reading group would take me about an hour each quarter. These objectives grounded the small group lessons I did with each group. It only took ChatGPT a minute to create these specific learning targets for each reading group.

Now, I can take these objectives and build out straightforward weekly lessons.

I asked ChatGPT to help me create a lesson plan based on one of the objectives it provided me for one week and include some sample paragraphs for me to use when teaching.

And here are ChatGPT’s small group lesson plans:

These lessons follow a specific learning trajectory: I do, we do, you do. I purposefully asked for this model of lesson to see if ChatGPT could replicate it well enough. This model was the one I often iterated with my small groups when teaching them new reading skills.

I would typically reserve Sunday afternoons to create my small group lesson plans. I would begin around noon and finish around 4 or 5 PM. A task that would have taken me 4-5 hours each week to complete, ChatGPT was able to accomplish in about 15 minutes.

While there are still some areas of improvement, which I will discuss in the next section, gaining 3 hours and 45 minutes each week translates to saving over 131 hours each school year or five and a half full DAYS.

Room for Improvement

One difficulty I ran into when writing this piece was that I expected the AI plugins for ChatGPT to be able to read PDFs accurately. They were unable to read a PDF with tables. Thus, I couldn’t get highly accurate reflections on student learning objectives based on the reading behaviors checklist from Fountas and Pinnell. I will spend more time exploring PDF readers to find one better able to read and analyze complex structures in PDFs. Even if I am unsuccessful in finding a proper PDF reader, I anticipate that in the next 3-6 months, the PDF readers will be advanced enough to read the tables and special formatting on PDFs like the one I used as a reference.

Using ChatGPT to optimize my small group instruction would have been game-changing in my classroom. Instead of spending dozens of hours reorganizing my students into new groups, summarizing the needs of each group, organizing these needs into objectives, then creating small group lesson plans for each group, learning how to prompt ChatGPT to do these tasks for me took about 3 hours. Once I was proficient in prompting to get my desired results, creating a series of bespoke lesson plans for each group took about 15 minutes, which would have saved me nearly five and a half full days each school year.

Have you considered using an LLM for your small-group instruction? Have you successfully taught using an LLM for your small groups? Also, do you know of an AI-PDF reader that can read complicated text structures? Please leave a comment below! We’d love to hear from you.