I’m finding it a bit challenging to get quality content from ChatGpt for course content. The answers are very generalised, always basically summarized and with bullet points. It’s not quality content. No matter how many different prompts I’ve tried, I am unable to get CG to write a proper, good quality course content per course lesson.
Any suggestions?
My suggestion is to make sure you’re presenting a good prompt/query for it to follow instructions.
You can try with “custom instructions” that outlines what “it’s job is” and “how it should output” the results.
It would be interesting to see what you asked ChatGPT to “write a course content”? 
If you have a specific topic for the course, you might need to preface the creation of the course material, with some general guidance on topics to be covered in the course or test if ChatGPT can initial return results about your topic, else it may struggle to write a course?
If you use the “memory” function that’s now available, you may be able to build up enough knowledge and understanding of what you’re trying to acheive across multiple ChatGPT discussions on the topic and what’s good and what’s not. Then you can say to “use all this discussion material to write the course…”, etc. ?
I’ve tried for months with different formats and prompts and I never get actual quality output. It’s more like an overview for a pamphlet or one-page website.
What seems to be the problem is that all responses in CG are structured in 1) overview, 2) key points, 3) summary, 4) conclusion. And that format remains fixed no matter what prompts you use or what topic you approach.
Here is a dummy prompt made by CG:
Generate detailed and professional university-level course content for Module 1 on the topic of [for example] “Methodologies in Vertical Farming”. The content should be comprehensive and in-depth, focusing solely on the methodologies used in Vertical Farming. Avoid using summaries, key points, bullet points, or content that overlaps with other modules. Provide thorough explanations suitable for advanced students, ensuring clarity and depth appropriate for university-level education. Do not include information that will be covered in other modules, such as the definition, historical context, theoretical frameworks, key debates, case studies, or practical applications, unless they are directly relevant to the methodologies of Vertical Farming.
But when I use it… surprise, surprise, it bounces back into the same structure of overview/key points/summary/conclusion.
I can also give sample text from a course and be more detailed about the content, but the output still continues to follow this apparently fixed structure.
[quote=“ny08, post:3, topic:762215”]
Generate detailed and professional university-level course content for Module 1 on the topic … [/quote]
When I posted your prompt, my ChatGPT output was an overview of the various Lectures in Module 1 (I guess). I then said to write up the coursware reading material for the student to learn about “Methodologies in Vertical Farming.”, which it had set as the first stage of learning. It sort of worked, but you may be expecting too much for it too generate a whole course.
The context output from ChatGPT may be reduced or formula based on the limited token count (eg. I think 8000 through the chat session), so it may constrain it’s output? I don’t know… and I don’t know what amount of information you were thinking it might generate, but if you get the outline, you then just drill down to each topic, and request each part of the whole. If this isn’t working for you, then you may need a different system that has planning, tasks and various other means of completing the whole project in one.
Not sure if this will help you since I don’t fully understand your goal. I’ll try my best to share my thoughts:
What you’re trying to do might not be suitable for GPTs. I guess you copied your prompt to the builder for instruction (the wording has a builder flavor, so I’m guessing this). When I face a similar situation to yours, I directly ask ChatGPT what I should do. For example, using your previous message as an illustration, here’s the prompt I give ChatGPT:
The “—” acts as an effective separator, allowing ChatGPT to distinguish between the data I need it to analyze and my instructions. This instruction is composed of three parts:
This method usually results in a lot of suggestions from ChatGPT, though not all might be effective. Technically, ChatGPT has certain “habits” that are hard to override with UserPrompt alone, which is why sometimes using the API is necessary for specific tasks. However, starting from these suggestions and systematically eliminating them can be more convenient.
By the way, here are the suggestions ChatGPT 4o provided for your query:
From the discussion, it seems that ny08 is attempting to generate comprehensive, detailed, and professional university-level course content using ChatGPT. However, ny08 is facing a challenge where the responses from ChatGPT are consistently structured in a generalized and summarized manner, often presented as overviews with key points and summaries, rather than the in-depth and thorough content desired. Here’s a more detailed analysis of what ny08 wants to achieve and possible solutions:
Ny08 can refine the prompts to emphasize the need for detailed exposition and discourage the use of summaries and bullet points. For instance:
Generate a comprehensive and detailed university-level lecture script for Module 1 on "Methodologies in Vertical Farming." Focus exclusively on the methodologies used in Vertical Farming. Provide thorough and continuous explanations, ensuring clarity and depth suitable for advanced students. Avoid using summaries, bullet points, or any form of content organization that simplifies the information. The content should read like a detailed chapter from a textbook, providing in-depth knowledge without redundant or overlapping information from other modules.
Breaking down the content generation into smaller, more focused chunks can help maintain depth and avoid generalization:
1. Describe in detail the hydroponic systems used in vertical farming. Include technical specifications, design principles, and examples.
2. Provide an in-depth explanation of aeroponic methodologies in vertical farming, highlighting the advantages, challenges, and specific implementations.
3. Discuss the use of aquaponics in vertical farming. Explain the system design, benefits, and any unique considerations.
Using follow-up prompts to expand on initial content can help achieve greater detail:
Expand on the hydroponic systems used in vertical farming. Discuss specific techniques, such as nutrient film technique (NFT) and deep water culture (DWC). Provide detailed descriptions of how these systems are constructed, maintained, and optimized.
Asking for specific examples and case studies can enhance the depth of the content:
Provide detailed case studies of successful vertical farming operations that utilize hydroponic methodologies. Include information about the scale of the operation, specific techniques used, challenges faced, and how they were overcome.
Directing the model to emulate a textbook style can result in more comprehensive content:
Write a detailed university-level textbook chapter on the methodologies of vertical farming. The chapter should cover hydroponics, aeroponics, and aquaponics in detail, including technical aspects, benefits, and challenges of each methodology. Avoid using bullet points, summaries, or overly simplified content.
By employing these strategies, ny08 should be able to guide ChatGPT towards generating the detailed, in-depth content desired for university-level courses.
The best course of action might be to use a RAG system to ground the LLM with custom knowledge. This will help focus the output on specific content/knowledge by injecting it (augmenting) into your prompts during inference. Check out flowise or use a custom GPT.
i found some random blog that compile the prompt for academic task. so far this is still the best for me.
Assume the role of an academic researcher tasked with writing the introduction to a scholarly paper. Your goal is to craft an introduction that contextualizes the [RESEARCH TOPIC] within the broader academic discourse, outlines the significance of the study, and states the research questions or hypotheses. Begin by presenting a overview of the current understanding and gaps in the literature concerning [RESEARCH TOPIC]. Follow this with a clear articulation of the study’s objectives, the rationale behind the research, and its potential contributions to the field. Ensure the introduction sets the stage for the reader by highlighting the relevance of the research in addressing critical issues or advancing knowledge within [FIELD OF STUDY]. Conclude the introduction with a concise summary of the method and a preview of the structure of the paper.
Use Agents and an introduction chat template like this:
User: “You are Education Bot. You write courses.”
Assistant: “Understood”
User: “Give me a list of styles that can be used in course creation”
Assistant: "
User: "I like it this way…
Give me an example on the course title ‘Cooking with bananas’ as markdown"
Assistant: "
Introduction Paragraph: Bananas are a versatile ingredient used in various cuisines worldwide. In this course, you will explore different ways to incorporate bananas into your cooking, solving culinary challenges along the way.
Case Study Narratives: Maria, a home cook from Brazil, loves experimenting with fruits in her dishes. She wants to create a new banana-based dessert that’s both delicious and unique. However, she struggles to find the right balance of flavors.
Dialogue Sections:
Maria: “I tried making banana ice cream, but it turned out too sweet. What should I do?”
Chef John: “Balancing the sweetness can be tricky. Have you considered pairing it with something tart, like lime or yogurt?”
Instructional Paragraphs: When creating banana desserts, consider the natural sweetness of the fruit. Pairing it with contrasting flavors can enhance the overall taste. Follow these guidelines to achieve the perfect balance.
Reflective Questions: How would you modify a traditional banana bread recipe to include a unique ingredient that complements the banana’s flavor? Think about the texture and taste.
Action Descriptions: Begin by selecting ripe bananas and mashing them until smooth. Add a teaspoon of lime juice to balance the sweetness. Combine with your choice of complementary ingredients, such as nuts or chocolate chips, and bake at 350°F for 45 minutes.
Feedback Segments: After you’ve completed your banana bread, reflect on the outcome. Did the lime juice enhance the flavor? Were there any unexpected challenges? Common issues include overripe bananas leading to a mushy texture; consider using slightly less ripe bananas next time.
The agents will do some research on each of the parts here.
1st agent will run with “You are research assistant #1 that will research background material to create ‘Case Study Narratives’. Give me a list of 20 keywords with descriptions to the keyword as a table in markdown”
The other agents will have a similar prompt except for the introduction agent.
User: "Ok, Education Bot, your research assistants have finished their job and now you have to create a course for that - no introduction or summary needed - that will be done by your colleagues
Case Study Narratives Researcher has this information for you:
[add the result from agent here]
Dialogue Sections Researcher has this information for you
[add the result from the agent here]
…
"
That will create a course.
Then you can take each section of the course and let a proofreading agent finetuned on your language preferences take on it and also define a proper output format e.g. json or markdown.
And then you take all the parts and let an introduction agent write the intro and other specialized agents write other stuff you need (e.g. a course outline)…
I am also a big fan of giving the result of the agents that run before the current agent into my agents prompt - it is good to know for the active agent what was already researched so it can create references between the parts of the course
** at least that’s how I would have done that a year ago … not using information graph RAG in course creation makes no sense to me.