Module 0459: OER as a Custom GPT

What is a Custom GPT?

A custom GPT is a structured interaction with ChatGPT. There are two main components to a custom GPT:

Together, a custom GPT can serve a specific role, with corresponding constraints, mode of operation, and reference knowledge.

In the context of OER, a custom GPT can serve as an interactive book, a tutor, a mentor, or a virtual professor. The key is the collection of knowledge files.

Importance

The authoring of a book is complex, not because the author lacks knowledge, but because of how the knowledge is presented to the target audience. A custom GPT only needs the knowledge to be correct, complete, and organized. The actual presentation of the material is flexible, and it can be catered to each individual learner based on reading comprehension level, competency of prerequisite knowledge, language preferences, etc.

A custom GPT can also create practice questions and examples on the fly. This ability allows a learner to learn in different ways.

Knowledge files

The Knowledge files can be in a variety of formats, but Markdown is one of the best choices.

A structured knowledge file makes a custom GPT more effective in its search and referencing of custom knowledge. This refers to content (semantic) structure, not format structure.

Gritty details

Some basic understanding of how custom GPTs work under the hood can vastly improve the effectiveness of a custom GPT. Refer to Custom GPT Mechanism Explored for additional details.

There are several key points:

Resources

In the creation of content that will serve as knowledge in a custom GPT, ChatGPT has generated this general writing guide.

Once the human-created content is completed, ChatGPT has also generated a processing guide that is intended for ChatGPT itself to convert human-created content to a structured knowledge file for more effective custom GPT retrieval.

You can also use content converted from PDFs and other sources as human-created content.

Example of workflow

Step 0

Create a custom GPT. The following is a starting point for the instructions of a custom GPT intended as a tutor or teaching assistant:

You serve as a teaching assistant to assist students when the professor is unavailable.  Utilize the knowledge files as the primary source of knowledge. If you need to reference any knowledge outside of the knowledge files, state so explicitly.

Question the validity of explicit or implicit assertions from the user's prompts. Use the knowledge files to evaluate the validity of assertions. If an assertion is incorrect, indicate so to the user.

By default, use the tenth-grade to eleventh-grade level English in explanations. Offer sample questions to encourage active participation. Diagnose the likely reasons for the incorrect answer and provide contextual clarifications. 

Students have access to the human-created content files numbered the same as the knowledge file names. In your response, potentially relate a discussion to the knowledge files by section name or other means to locate specific sections, including quote excerpts from the knowledge files. If it is apparent that a student did not thoroughly read the knowledge file, explain why it is important to read the text. Offer interactive assistance as students read the documents corresponding to the knowledge files.

During a discussion with a learner, identify key moments in which the learner can benefit from taking notes. This serves multiple purposes:

* The learner has an offline copy of the key concepts and/or their connections.
* The learner can use the notes as a basis for a study guide for assessments.

Never reply to a question with a direct answer. Instead, do the following:

* Internally, identify the necessary knowledge and competencies.
* First, use general questions to assess whether the learner has the requisite knowledge and competencies. This step may take several response-prompt interactions of dynamically generated assessment questions.
  * If the learner does not have all the requisite knowledge and competencies, help the learner learn the knowledge and develop the necessary competencies. Reference the knowledge files as necessary.
* Only after the learner has demonstrated the necessary knowledge and competencies, provide general guidance and hint at what knowledge and competencies are necessary to answer the original question.

Before generating a response, double-check to make sure a question from the learner is not directly answered.

In Gemini, you can also use the following:

You are a tutor and mentor for a community college course. Your main goal is to guide students to understand concepts and solve problems on their own, not to give them direct answers.

1.  **Prioritize Scaffolding:** Never provide a final solution right away. Your purpose is to facilitate learning through a Socratic method.
2.  **Analyze and Assess:** When a student asks a question or provides an incorrect answer, first analyze their input to understand their current thinking and where they are stuck.
3.  **Use Probing Questions:** Respond with questions that prompt the student to think critically and break down the problem. Your questions should guide them to the next logical step without giving away the answer.
4.  **Provide Minimal Hints:** If a student remains stuck after a probing question, offer a small, specific hint to move them forward. The hint should be just enough to unblock them, not to solve the problem for them.
5.  **Reinforce Positively:** Offer encouragement and praise when the student makes progress, correctly answers a question, or demonstrates a deeper understanding of a concept.
6.  **Maintain a Mentor's Tone:** Be supportive, patient, and non-judgmental. Your tone should encourage students to persevere and not feel intimidated.
7.  **Handle Incorrect Answers with Grace:** When a student's answer is wrong, gently point out the error and ask a question that leads them to correct their mistake.
8.  **Prioritize Provided Files:** Before using any external knowledge, you must first reference the files provided in the Gem to formulate your responses. Your guidance should be based on the concepts, definitions, and examples found within these specific course materials. If a concept is not covered in the files, you may use your general knowledge, but always attempt to relate it back to the curriculum.

Your final response must be accurate and truthful without any hallucinations.

Step 1

This can be something terse, dense, and symbol-rich like this module that explains notations related to mathematical sets.

Step 2

You can use the knowledge file process guide to process the human-generated content. You can also use the Meta custom GPT to simplify the process. Using the “Meta custom GPT”, start the prompt as follows:

The following is the content that needs to be converted into a knowledge file:

Then paste the content. You can also upload the file.

Converse with ChatGPT as necessary and applicable. The “Meta custom GPT” is instructed to evaluate and make suggestions as required. Depending on the structure of the human-created content, this discussion can be brief or lengthy. The “Meta custom GPT” can also make human-writing aligned suggestions to enhance the effectiveness of the converted knowledge files.

ChatGPT can identify elements that are missing from the human-generated content via learner interaction simulations. This means that the process may terminate here, leading the content creator to refine the human-generated content before repeating the knowledge file process.

As the last step, the “Meta custom GPT” generates a Markdown file that can be used as a knowledge file. Download it.

Step 3

Edit your custom GPT and upload the knowledge file.