Forum in maintenance, we will back soon 🙂
This is the prompt I am using:
Act as a master computer science teacher.Generate five coding exercises using [{language}] and covering the provided [{concept}].The exercises should meet the chosen educational [{standard}].Include any additional [{conditions}] provided by the user.Provide a solution and test cases for each exercise.Provide the full name and full description of the standard.For AP or Advanced Placement standards include the Big Idea, Enduring Understanding, Learning Objective, and Essential Knowledge
It works great and gives me the output I want but the output never looks the same twice. What I am aiming for.... this is for (mostly high school) Computer Science teachers to generate coding exercises for their students. They provide the inputs shown. Every state has educational standards for classes. Advanced Placement (AP) classes have their own that are national standards. I want to provide the standard met for the exercises. If the standard is an AP class I want the specific parts of those standards listed as described in the last line of the prompt. Since each state is different I want to keep it generic but I want to provide the right information.
Â
Also I want consistency in the way the results output. Is there a way to format that prompt or the script to ensure the results look the same each time?
ChatGPT is never going to give you the same output twice unless you give it the output to give. Have you tried creating a Custom GPT to try to get more consistent results?
Regards,
Earnie Boyd, CEO
Seasoned Solutions Advisor LLC
Schedule 1-on-1 help
Join me on Slack
@ssadvisor I have not. How might one do that? (if that is in this course and I missed it I apologize for the redundant question)
@christopher-evans Hi, as @ssadvisor mentioned, LLms will never give you the exact same response.
To format your output, yes you can get a consistent structure like json or whatever by feeding the output examples to the model, or for json you can use Instructor with pydantic, check this post:Â learnwithhasan.com/create-ai-tools-fast-beginners/
going back to consistent results, if you can collect data of the expected output, then you can use RAG (retrieval, and then get what you want from documents); otherwise, GPT will generate different outputs every time, even with the same prompt.Â
What @ssadvisor mentioned about CustomGPTs, yes you can build a GPT that is tailored to do a specific task and power it with data and functions, check this video: https://youtu.be/RzIoeDKt538?si=mHanQn6zCrJi874e
Â
Â
