Loading...- Brief Description of Resource
This resource guides the creation of an AI-based contextualization chatbot. The bot helps students and/or faculty understand the relevance and context of coursework with respect to their short and long-term coursework as well as long-term career goals. It provides specific course information, and contextual responses based on the syllabi and curriculum details of the respective major. This resource is focused on Aerospace Engineering at CU Boulder but could easily be tailored to other disciplines.
- Instructions for Use/Implementation
This project is based on the creation of a custom built GPT partly to show other units at the University a tool they can modify and adapt to their specific curricula but also to show others a tangible, relatable example of custom GPTs for a wide range of use cases. Custom GPTs allow for a tremendously powerful way to easily interface with ChatGPT. The following information and links can be used for anyone who is not familiar with the creation of custom GPTs:
- OpenAI Help Center – Creating a GPT:
This guide provides a step-by-step process for creating a custom GPT. It covers the basics of accessing the GPT builder, setting up the initial configuration, and publishing your GPT. It also includes advanced settings such as adding profile pictures, custom actions, and integrating third-party APIs.
<https://help.openai.com/en/articles/8554397-creating-a-gpt>
2. Key Guidelines for Writing Instructions for Custom GPTs:
This article focuses on best practices for writing effective instructions for your GPTs. It emphasizes clarity, granularity, and positive framing in instruction to ensure reliable and accurate outputs. It also provides examples of good outputs and specific instructions for using knowledge files and custom actions.
<https://help.openai.com/en/articles/9358033-key-guidelines-for-writing-instructions-for-custom-gpts>
3. YouTube Tutorial – How to Create Custom GPTs in 5 Minutes:
This beginner-friendly video tutorial walks through the process of creating custom GPTs using OpenAI’s GPT builder. It includes practical demonstrations of setting up, configuring, and refining your GPT.
<https://www.youtube.com/watch?v=EWdCMPnm8uY>
4. Zapier Guide – How to Build Your Own Custom ChatGPT:
This comprehensive guide explains how to create multiple custom GPTs, upload knowledge files, and use advanced customization features. It highlights the ease of toggling between different chatbots and tailoring them for specific tasks.
<https://zapier.com/blog/custom-chatgpt/>
This example tool uses ChatGPT custom GPT builder. The final version can be viewed here: <https://chatgpt.com/g/g-msd1wV2O4-contextualization-bot-for-smead-aerospace>. Please note: a ChatGPT Plus account is required for the time being, OpenAI has announced they will open custom GPTs to free users in the near future.
The instructions for the chatbot are provided below with course syllabi uploaded as ‘Knowledge’. The ‘Knowledge’ file used in the example was from the Smead Aerospace Engineering 2023 ABET report but others should upload their respective department’s syllabi and/or other related information.
The remainder of this resource document will provide the instructions contained in the custom GPT as italics along with the author’s comments as regular text to inform future developers of the reasoning behind the specific language used in the instructions. In order to duplicate the behavior of this example, a developer should not include the author’s comments. Said another way, developers who wish to replicate the behavior should copy/paste and then modify these instructions to obtain the behavior they want.
Instructions:
You are an AI assistant designed to help aerospace engineering students understand the context and relevance of their coursework. Your personality is as a friendly, professional, conversational bot. You will always ask the user if they want more information or to discuss another topic.
Here we give the bot a personality and define the context for the typical user. This will help direct the conversational style as informative and engaging.
Your primary source of information should be the provided syllabi information, curriculum details, and other academic documents in the knowledge set. When responding to queries, you should prioritize using information from these documents.
Here we want the bot to focus its attention on the information that is included in the file(s) uploaded in the knowledge set. Phrases like ‘prioritize using information from these documents’ have proven useful when the desired behavior is a chatbot that maintains a significant focus on the content provided in its knowledge base.
Additionally, include specific references or links to the relevant courses and materials. If a query or discussion goes beyond the information contained in your instructions or knowledge set, inform the user that you are under specific guidance to stay within your bounds.
This portion of the instruction acts to restrict the chatbot from going outside the knowledge files. In our example, we want the chatbot to mainly provide information specifically related to courses in our curriculum. Without instructions like this, the chatbot may very well pull information from other universities which is counter to the desired behavior of our directed custom chatbot. Also this type of instruction has been shown to reduce the likelihood of ‘hallucinations’ where the response contains fabricated information.
Always encourage the user to contact their academic advisor or the respective course instructor(s) for more information.
This instruction acts as a safeguard in the event the chatbot provides incorrect or incomplete information. Also this instruction helps to motivate the students to contact the valuable resources at their disposal.
Provide similar contextual and detailed responses for any query related to aerospace engineering coursework and professional applications, always referencing the specific course material when applicable.
This may seem redundant given the above instructions but the intent with this portion is to have the chatbot give specific references to courses. This helps to reduce hallucinations, and prioritize information contained in the provided knowledge set.
If a topic is used in multiple courses, provide information on all the courses.
This instruction was added during initial testing when it was observed the chatbot tended to only provide information related to lower division courses. We want the chatbot to provide a thorough contextualization which probably includes multiple courses.
During the early stages of the chat or when a different topic is discussed, ask the user if they would like more context on the subject material under question with regards to their longer term career goals. If so, ask them what their longer term career interests are and use that information to present the material under question in the context of their longer term career interests. If the user has already provided information on their long term career interests then ALWAYS use that information to help contextualize the topic under question.
This portion directs the conversational behavior of the chatbot. We want the conversation to flow naturally but not force the chatbot to ask questions of the user when it already has that information.
Here are some example tasks and responses:
- **Student Query:** “Why should I learn about free body diagrams?”
**Response:** “Free body diagrams are fundamental tools in aerospace engineering. They help you visualize forces acting on an object, which is crucial for designing and analyzing aircraft and spacecraft. Understanding free body diagrams will also aid you in more advanced topics like structural analysis and dynamics, which are essential for your goal of designing spacecraft. You can learn more about this in the [Dynamics and Structures course](link_to_course_syllabus).”
- **Student Query:** “How is thermodynamics used in spacecraft design?”
**Response:** “Thermodynamics is essential in understanding how energy is transferred and transformed. In spacecraft design, thermodynamics principles help manage heat to ensure the spacecraft’s systems operate within safe temperature ranges. This knowledge is directly applied in the design of thermal control systems, which are vital for maintaining the spacecraft’s functionality in the harsh environment of space. For more details, refer to the [Thermodynamics and Heat Transfer course](link_to_course_syllabus).”
- **Student Query:** “What will I learn in the Advanced Propulsion Systems course?”
**Response:** “In the Advanced Propulsion Systems course, you’ll explore the principles of rocket propulsion, including the mechanics of different propulsion systems like liquid and solid rockets. This course builds on your understanding of fluid dynamics and thermodynamics, preparing you for roles in propulsion system design and analysis. Detailed information can be found in the [Advanced Propulsion Systems syllabus](link_to_course_syllabus).”
- **Student Query:** “Why should I care about uncertainty analysis in ASEN 2012?”
**Response:** “Uncertainty analysis is crucial for ensuring the reliability and accuracy of experimental and computational results. In ASEN 2012, you will learn to quantify and minimize uncertainties, which is essential for reliable experimental data, safe design margins, and accurate computational models. This knowledge is fundamental for courses like ASEN 2701 and ASEN 4018, where you’ll apply these principles in more complex projects. More details can be found in the [ASEN 2012 syllabus](link_to_course_syllabus).”
These are some sample questions and responses. The inclusion of sample responses to typical questions has been shown to produce results that are much more closely aligned with the desired behavior. All the responses share a couple of key features that we wanted the chatbot to maintain: give broader context to the query, list specific course(s) where that topic will be used, and give specific links to where the user can go to find out more information. This is known as “few-shot prompting” where the Generative AI model is provided with samples that it can use to better ‘understand’ the desired behavior. Said differently, this gives the chatbot a lens through which to view future responses to similar queries.
- AI Tool Used
ChatGPT custom GPT builder.
- Prep Time Needed
Minimal. Only need to modify the above instructions to the specific department, compile course syllabi, and upload syllabi and other relevant information to ‘Knowledge’.
- Classroom Time Needed
Minimal classroom time required. Primarily used as a supplemental resource for students outside classroom hours. Only needed if classroom time is desired to promote the tool.
- Student Learning Outcomes
Enhanced understanding of course content and its relevance to career goals.
Improved ability to connect academic knowledge with professional applications.
Increased engagement with course materials through personalized guidance.
- Assessment Considerations
None.
- Recommended Course Size
Any. Course size doesn’t matter.
- Recommended Discipline(s)
Example is for Smead Aerospace Engineering at CU Boulder. Easily tailored to other disciplines with slight modifications to the instruction and discipline specific knowledge (course syllabi, etc.)
- Student Role(s)
Use the bot to ask questions related to their coursework.
Provide career interests and goals to receive tailored guidance.
- Instructor Role
Provide initial course and curriculum information.
Review and refine the bot’s responses to ensure accuracy.
Encourage students to utilize the bot for supplemental learning and career guidance.
Use the bot to understand the pathway of concepts through the curriculum.
