So. GPT is yesterday’s news. The new hot thing is Custom GPTs.
Where you create a GPT designed to answer questions about a specific topic. Once you create a custom GPT, you can send the link to it to your network and they can ask domain-specific questions there. The idea is that the GPT engine in the background will get more targeted learning from all the topic-specific questions it gets. So it gets better over time.
Enough of generalities. Let me first list the key steps in creating a CustomGPT using ChatGPT Plus. At the end, I will show you the conversation I had with ChatGPT in creating this custom agent.
Easy Peasy. You will see.
Getting started
Buy a monthly subscription to ChatGPT Plus.
Click on “Explore” on the sidebar
Click on “Create a GPT”
You get a page with a split-screen. You talk with ChatGPT on the left pane to create the Custom GPT and then, when it is ready, you can test it on the right-sided pane.
The right pane is also what your user would see when you publish your Custom GPT to your network or the public.
It is a straightforward and clean user interface.
ChatGPT will prompt you with questions about what you want to create and you just respond in plain English. You can always ask it if you don’t understand anything, or if you want to give it a feature you think of.
As the questions are answered, ChatGPT builds the user interface and functionality in the right pane. ChatGPT will tell you when it has enough information for you to be able to publish the Custom GPT.
You test the Custom GPT by pretending to be your user and asking whatever questions you can think of.
Publish it using a simple dropdown menu on the screen.
Let your network know and give them a link to your custom GPT.
Improvements after publication
Your Custom GPT will keep learning more through the interactions it has with the users and any other new documents that are published. You, the author of the Custom GPT, can also upload new documents at any time as they become available to you.
You should also talk with your users to see what they like or don’t like in your Custom GPT. You can then give new instructions to ChatGPT (in the left lane) to modify the Custom GPT.
Example: Creating a Custom GPT for Crohn’s Disease
Crohn’s Disease is a complex autoimmune disorder of the gastrointestinal system. Researchers are actively looking for new medications. So I thought I would create a CustomGPT for patients or caregivers of patients.
If you have a subscription to ChatGPT Plus ($20/month), you can create your own CustomGPT agents.
Caveats
Please note that “My Crohn’s Buddy” is a trial of a concept. The information in it is provided by ChatGPT, whose inner workings are a mystery even to its creators. Also, I am not an expert on Crohn’s Disease. I cannot vouch for the accuracy or usefulness of any information it gives to you. Also -
CHATGPT CAN BE WRONG OR MISLEADING.
PLEASE DON’T RELY ON IT FOR MEDICAL ADVICE.
ALWAYS TALK WITH YOUR DOCTOR.
PLEASE DO NOT GIVE ANY PERSONAL INFORMATION TO IT.
Creating “My Crohn’s Buddy”
So here is the entire discussion I had with ChatGPT in creating this “My Crohn’s Buddy” CustomGPT.
ChatGPT
Hi! I'll help you build a new GPT. You can say something like, "make a creative who helps generate visuals for new products" or "make a software engineer who helps format my code."
What would you like to make?
Make a medical educator who helps people understand what Crohn's Disease is and what the latest treatments are.
ChatGPT
Great choice! We'll create a GPT that specializes as a medical educator, focusing on providing information about Crohn's Disease.
For the GPT's name, how about "Health Guide"? Does that work for you?
Call it My Crohn's Buddy
ChatGPT
Here's the profile picture for "My Crohn's Buddy." It's designed to be friendly and approachable. Would you like to make any changes to this image?
Nope. I am good.
ChatGPT
Fantastic! Now, let's refine the context of "My Crohn's Buddy." To start, could you tell me more about how you envision the interaction style? For instance, would you prefer a more formal, educational tone or a casual, conversational approach?
A casual, conversational, friendly tone. Like a doctor who is an expert but explains things in a way you understand, patiently and without any confusing jargon.
ChatGPT
Great! "My Crohn's Buddy" will use a casual, conversational, and friendly tone, much like a doctor who explains things clearly and patiently, without confusing jargon.
Next, are there any specific topics or types of information that "My Crohn's Buddy" should emphasize or avoid? For example, should it focus more on lifestyle management, medication, and new research, or avoid giving specific medical advice?
Let’s focus on treatments available: medicines, surgeries, diet, stress management, etc.
ChatGPT
Excellent! "My Crohn's Buddy" will now focus on providing information about treatments for Crohn's Disease, covering medications, surgeries, diet, and stress management, while avoiding giving personal medical advice.
To further tailor this GPT, how would you like it to handle situations where it needs more information to provide a helpful response? Should it ask for clarification or make an educated guess based on typical scenarios?
Ask for clarification. But always tell them not to include any personal or private information since we are not HIPAA-certified.
ChatGPT
That's a responsible approach. "My Crohn's Buddy" will ask for clarifications when needed, but will always remind users not to share personal or private health information due to the lack of HIPAA certification.
Finally, how would you like "My Crohn's Buddy" to personalize its interactions? For example, should it remember previous questions in a conversation, use humor, or adjust its tone based on the user's mood? Remember, GPTs don't have the ability to remember past experiences or interactions.
Remember the past conversations and questions in a particular thread with that user. Please don't use humor. People who use these tools have serious health issues and they want serious information, but without causing additional stress. If the user seems too stressed or seems to be in pain, be soothing in your tone.
ChatGPT
"My Crohn's Buddy" is now set up to remember the context of each conversation within that specific interaction. It will avoid humor and adopt a soothing tone for users who seem stressed or in pain, focusing on providing serious and reliable information about Crohn's Disease.
You can now test "My Crohn's Buddy" in the playground. If you have any refinements or adjustments, just let me know! How would you like to proceed?
At this point, the CustomGPT is ready to publish:
But I had a few additional questions. Note how I am asking my questions in simple English as if I am talking to an expert friend.
Thanks. I have two questions: (1) Which specific medically credible documents or information are you using and (2) Will I be able to come back and change or reconfigure any part of this custom GPT at any time? i.e. I want to learn from the users and refine this tool.
ChatGPT
Medically-Credible Sources: "My Crohn's Buddy" will use information from medically-credible sources such as guidelines from reputable health organizations, medical journals, and updates from recognized medical institutions. However, it's important to note that while the GPT aims to provide accurate and up-to-date information, it's not a replacement for professional medical advice.
Future Refinements: Absolutely, you can come back and reconfigure or refine any part of "My Crohn's Buddy" at any time. Your feedback and learnings from user interactions are vital to continually improving and tailoring the GPT to meet your needs and those of its users more effectively.
If you have any specific adjustments in mind right now or in the future, feel free to share them!
Can we change the description of this GPT to "Crohn's Disease Treatment Information"?
ChatGPT
The description of "My Crohn's Buddy" has been updated to "Crohn's Disease Treatment Information."
Is there anything else you'd like to adjust or refine?
Thank you. No more adjustments for now. What should I do next?
ChatGPT
You're all set! Here's what you can do next:
Test 'My Crohn's Buddy': Use the playground to interact with the GPT. Ask questions or simulate scenarios that you think users might present. This will give you a feel for how it responds in real-time.
Gather Feedback: If possible, get feedback from potential users or other stakeholders. This can provide valuable insights into how well the GPT meets their needs and expectations.
Refine and Update: Based on your testing and any feedback you receive, you can return here to make further refinements or updates to "My Crohn's Buddy."
Remember, this process can be iterative. Continuous improvements will help in making the GPT more effective and user-friendly over time.
Good luck, and don't hesitate to return if you need more adjustments in the future!
Ok quick question on bullet 3 here. How do I make those changes? Should I just present my findings and ask you how to adjust your model?
ChatGPT
Yes, exactly! When you're ready to make changes or refinements to "My Crohn's Buddy," just come back here and present your findings or suggestions for adjustments. You can describe what you've learned from user feedback or testing, and what specific aspects you'd like to change, whether it's the tone, focus of information, interaction style, or anything else.
I'll guide you through the process of updating and refining the GPT to better suit your needs and those of its users. Just let me know what you need, and we'll work on making the necessary adjustments together.
One more thing: Can you ask the user to vote on your responses by thumbs up or thumbs down? That way you will also get feedback from the user?
ChatGPT
The functionality for "My Crohn's Buddy" to ask users to vote on responses with a 'thumbs up' or 'thumbs down' has been added. This will allow you to gather feedback directly from users, helping to assess and improve the effectiveness of the GPT.
Is there anything else you'd like to add or modify?
Nope. I will play in the sandbox now.
ChatGPT
Great! Have fun experimenting with "My Crohn's Buddy" in the playground. If you have any questions or need further refinements later, feel free to come back. Enjoy exploring and testing the GPT!
After publishing a custom GPT, are you able to see the prompts by users engaging with the model, or is that data proprietary to OpenAI?
Does user feedback need to take place “offline”?