May 16th, 2024

​Integrating Generative AI into Personalized, Competency-Based Learning Settings

​Now that frontier models of generative AI are no longer stuck behind a paywall,  how do we effectively integrate these tools into the flow of our teaching and student learning? While I’d be skeptical of anyone claiming to have a concrete answer on this, it is possible that we already have some promising ground to build upon.

This week I launched two new LLM tools in class to complement and enhance the personalized, competency-based learning initiatives that I’ve been developing with my students over the course of the past 12 years. Both tools are offering promising returns, and neither would have been imaginable to me a year ago. At the same time, I’m not sure if I’d be as open to experimentation and early implementation if personalized and competency-based learning hadn’t already pushed me to reimagine my role in the classroom and the opportunities it afforded student learning.

Finally upgrading to ChatGPT4 last month, I excitedly began to explore how I could build my own GPTs to support my students. I was impressed, but frustrated. No matter what I built, I was going to face an accessibility problem as the full benefits of the tools and prompts I was experimenting with would be stuck behind a paywall that many of my students would be stuck behind. However, with ChatGPT’s recent announcement that the free model would be upgraded to 4o, this is no longer the case, or at least it won’t be for much longer.

With this universal access in mind, I identified two immediate problems I am having. First, my students need more one-on-one writing instruction, and there’s only so much of me that I can make available in class and office hours. I like grading some of my student writing, but not all of it, and not all the time. Every new assignment I give means more work for me to do outside of teaching hours. I could suggest students to just ask an LLM for help, but as Ethan Mollick has warned, that often leads students down the road of temptation, where they can ‘offer evidence’ of achieving outcomes without having to go through the necessary steps to properly develop the skills to achieve those outcomes on their own. I remind my students often that this is a sure path towards being made obsolete in this coming age.
​
The second problem relates to a complicated oral commentary that my IB Language and Literature students have to perform. For this, they are not allowed to “rehearse” with me. They already performed a mock, and they received feedback on this, but this is a completely new task with new texts to comment on, and outside of an initial outline that I am allowed to give feedback on, they are on their own.
​
​







​





Two New Tools
For the first problem, I built a GPT called “Virtual Mr. Pultz,” a sarcastic and helpful writing coach with a Captain Nemo scowl embedded on a bust of a cartoonish bald ginger in a tweed jacket and tie with question marks floating in the background. It's the WWE heel character that I never got to opportunity to play. Virtual Mr. Pultz starts by reminding students that “Test day is a rest day for the well prepared. How can I help you prepare?”

He then asks for students to share the assignment and rubric and prompts students to share what they think they did well and what they think they are struggling with. It then asks to see their work and gives them feedback according to the rubric. It uses the coded language that I use when grading to identify weaknesses so that students can access the resources on our LMS. Virtual Mr. Pultz is able to quite reliably identify those errors and offers a quick explanation with examples without correcting the initial mistake, though occasionally it does. It prompts students to do their own revisions and will then give further feedback. It also reminds them to review the resources I’ve provided on our LMS and to explore the opportunities to master those skills on Khan Academy, routines that I had already built into class independent of this new tool.

Once Virtual Mr. Pultz helps students identify their target areas for improvement, it prompts them to create a specific plan to make the needed improvement. It has been instructed to be encouraging to genuine questions and issues, but if the student responds lazily or disagreeably, Virtual Mr. Pultz employs the same rule posted in the real Mr. Pultz’s classroom: “Warning: Whining and general disagreeableness will be met with strong sarcasm.” Virtual Mr. Pultz has so far communicated to said laziness that “Test day is a rest day for the well prepared, but perhaps you prefer suffering.” Upon further whingeing, it quipped, “Suck it up, buttercup.”

Here is a link to the prompt I used to create it.

The second LLM tool I offered is a prompt that makes use of an existing GPT: IB Lang/Lit. While I’m still exploring its capabilities, I have had some strong early returns on students practicing their Individual oral commentary. For this assessment, students need to identify a global issue, which is defined as an issue of significance that is experienced both locally and globally. The students need to speak for 10 minutes on this global issue and how it is seen in a literary text and the larger body of work of the author and a non-literary text and its larger body of work. The prompt has the AI assessor wait until the student says “thank you” before speaking. At that point, it generates five total questions, one-at-a-time, for students to respond to. I’ve attached the prompt here for you to explore.

After this, it assesses students on the rubric that is provided, tells students what they did well, and gives them suggestions for improvement. The grading is a work in progress, but the question generation and feedback features are already working well.

In the case of both LLM tools, there will be a lot of trial and error and tweaking along the way. But in sharing these tools with my students with proper guidelines and expectations, as imperfect as they may be, an entrepreneurial mindset is modeled, one that opens many of them to exciting possibilities of what they might build on their own, leveraging this technology to support their own learning and to further explore their own passions.

Bigger Picture
While it might look like these opportunities aim at putting teachers out of business or passing the buck to a machine, I want to be clear that my aim is not to replace teachers or what teachers do. I am looking instead to enhance effective practice.

With that in mind, over the next five weeks, I’d like to openly reflect on a 12-year journey that I’ve taken with personalized learning in three very different school cultures that was further bolstered by a competency-based approach introduced to me five years ago. I am emphasizing personalized, competency-based learning because I believe it is the best holistic approach to serve our students and their modern sensibilities. It has the potential to bridge traditional practices with the skills needed to be competitive in the modern workplace while utilizing our existing technological capabilities.
​
This article is the introduction to the series. Each week I aim to reflect on different parts of the experience, highlighting unresolved issues and imagining how LLMs might assist this work in the future. I'll also offer a brief update on how the new tools are working and other initiatives I'm experimenting with. Listed below are the weekly focuses along with a brief description. I welcome feedback on these reflections, and I hope that it generates useful discussion.
Next week’s article will focus on the evolving role of the teacher. It will be released next Thursday.

Evolving Role of the Teacher in a Personalized Classroom and How GenAI might Support
In a classroom driven by personalized learning, the role of the teacher shifts to varying degrees to that of facilitator, curator, mentor, and coach. That doesn’t mean the sage has to completely give up the stage, but that stage certainly needs to be shared. This section will explore this evolving role and how AI tools can support this transition.
Taking the Leap of Faith into Personalized Learning
I was lucky enough to have an early experience where the school I worked at took a collective leap of faith, reducing classroom contact time with students and finding alternative curricular paths to achieve outcomes. I’ll share what that felt like, warts and all, and how it evolved over time. This section will provide practical first steps for educators looking to make personalized learning work and how incorporating LLMs can support.
Assessing and Grading and Leveraging GenAI to Support
With personalized learning, traditional grading systems are often inadequate. This part will discuss more dynamic assessment and grading practices and how LLMs can help. It will span grade-heavy systems to more progressive mastery systems.
Adapting Personalized, Competency-Based Learning Across Different Systems
This section will draw on my experiences working within various educational systems—from a school that collectively embraced this approach to one with more of a suitcase curriculum, to a much more rigid state system. I'll share insights on how personalized learning can be adapted and implemented in different institutional settings, each with its own challenges and opportunities.
Giving Students the Space and Structure to Surprise You
Students often exceed our expectations when given the tools and opportunities to direct their own learning. This concluding section will highlight some inspiring examples of student achievements that were made possible through personalized learning. It will also share some early successes that students are having while collaborating with LLMs.

---