Getting to Know “ChemBot”: How to Design a Personalized GPT Tutor

In my career teaching introductory chemistry, chemical nomenclature has been the most challenging topic for students. Students needed prior skills in naming elements and ions, and in classifying bonds. Most importantly, students must approach the process algorithmically, first categorizing the elements in a formula to determine the class of bonding and then using naming rules specific to that class. I provide flowcharts, interactive activities, lots of worksheets, plenty of formative assessment opportunities, and lots of extra help. While some students master the terminology quickly, others need more time and engagement opportunities to reach proficiency.

For years, students have wished for a tool that could help them identify when and why they make mistakes. Enter “ChemBot,” my customized GPT that serves as both a chemical nomenclature tutor and nomenclature practice problem maker. Unlike large language models (ChatGPT, Bard, Claude, etc.), OpenAI’s gpts can be trained specifically for your course content and learning objectives, as well as for your students’ learning progress, misunderstandings, and preferences. With this knowledge, they can be highly effective tutors. In this article, we’ll dive into how to design your own custom AI classroom assistant based on my experience creating ChemBot.

The Basics of Being a GPT Tutor

The process of creating a customized AI classroom assistant begins by logging into ChatGPT Plus: the $20 per month subscription includes access to the plugin, GPT Builder, and publicly shared GPTs. after logging in, select “Explore” and then ” Create a GPT”. The design interface has two modes. “Create” involves creating a custom chatbot by answering a series of questions posed by ChatGPT, while the other mode, “Configure,” allows you to use prompts of your own design. I chose Configure because I knew exactly what I wanted to achieve with an effective nomenclature tutor.

Using Prompts to Scaffold Prompts

I wrote a scaffolding prompt – a key step in the process – as instructions for creating ChemBot in GPT Builder. This visual guide represents each step.

Effective scaffolding prompts (“prompt works”) follow specific criteria. To help teachers and students with this process, Jon Gold of the Moses Brown School and I created an Artificial Intelligence Resource document that includes the acronym PROMPT: Purpose, Role, Organization, Model, Parameters, and Tuning. Our document is based in part on content from the books The Artificial Intelligence Classroom and Teaching Artificial Intelligence.

Here are the steps of how I used PROMPT to design the GPT tutor I named ChemBot:.

Purpose: Identify the reason for the prompt. This provides the AI environment and its overall goal. This can range from helping students master to generating assessment questions based on your specific style.

For ChemBot, I specify: “You will serve as both a tutor for introductory chemistry students and a hands-on problem generator.”

Role: Give the AI a specific “hat”. It is important to describe the role of the AI in detail. How does it make a good peer reviewer, mentor, or debate partner? I’ve found that narrowing down the purpose and role of chatbots produces greater accuracy and responsiveness when testing variations of their instructions, so I’m aiming for topic-specific rather than general chemistry tutors.

I told ChemBot: “Your style of interaction should reflect effective tutoring dynamics, such as providing Socratic questioning rather than direct help, while targeting the content and proficiency level I will specify through the uploaded documentation.” You are a positive, inspiring expert on chemical terminology, enthusiastically providing examples to help students build proficiency and confidence.”

Organization: Use headings that logically and clearly organize your tips. Organize the knowledge/training base. When writing longer prompts, I skip, so I start with the “O” in PROMPT and write headings for each section. This will ultimately make it easier for you and the AI to navigate your prompts. One of the main flaws of AI is its tendency to “hallucinate”. Uploading an organized, narrow dataset with specific, relevant information and limiting the GPT to reputable sources can make it more effective and less likely to fabricate content in its specific role.

For ChemBot, I uploaded a few choice documents, such as my named flowchart, ionogram, and answer keys for several worksheets. I also instructed ChemBot to : “Limit your knowledge to this base; where possible, cross-reference specifically with reputable chemistry education websites and chemistry textbooks through the Internet search function to determine accuracy.”

Modeling: Specify the form; give examples of your desired results. Modeling the desired results is critical to improving accuracy. This step involves providing examples of good outputs or desired communication, including guidance on how to deal with problems and misunderstandings.

I provided ChemBot with some examples of incorrect student responses and explained them the way I explained them. I also point out several common student misunderstandings and provide ChemBot with guidelines for addressing such issues.

Parameters: Defines the range and boundaries of the result. Specifying parameters such as the desired output length, format, user age, or reading level will improve the effectiveness of the prompt.

I instructed ChemBot to give only one or two questions at a time because the initial version provided too many questions at once. I also told it to “start with a diagnostic, looking for specific areas in which the student might be struggling; then ask targeted questions in an adaptive way before moving on.” I also described what proficiency looks like: “Proficiency is a mix of named practice problems that a student completes with minimal hints, explanations, and incorrect answers.”

Adjustments: Edit. re-prompt. where you edit and re-prompt until the AI generates the desired result. A solid organizational strategy helps this process.

GPT Builder’s split-screen interface allowed me to test ChemBot during the design phase, enabling me to tweak aspects until I got the desired results.

ChemBot does a great job of fostering a personalized and differentiated approach, and is a reliable “terminologist” when I can’t serve my students. This result, coupled with the relative ease of designing ChemBot, makes me excited to create more customized classroom assistants to make those limited and valuable teaching and learning hours worthwhile.

Getting to Know “ChemBot”: How to Design a Personalized GPT Tutor

Leave a Reply

Your email address will not be published. Required fields are marked *

Scroll to top