Key takeaways
- Start with your own plan and ideas before involving an LLM.
- Give the model multiple reliable sources and precise instructions.
- Read every line and add your experience to keep the writing personal and accurate.
Starting a new creative project, such as writing a lesson for a course, can feel intimidating, especially when you do not know where to begin. Fortunately, AI tools like GPT-4o and Sonnet 3.5 can make this process much easier. But do not merely ask them to generate a lesson for you. There is a good way to use these tools so the result does not simply look like generated content, and that is exactly what I will show you today.
Let us review a few tips for using LLMs effectively when writing lessons.
1. Do not just use ChatGPT and consider the work finished.
Seriously, people can immediately tell when you rely only on AI-generated content.
Make sure you have a solid plan with your own ideas before involving AI in the process.
LLMs are known for using generic language, which can make your lesson less engaging.
You will also need to edit the generated content to make sure it is well written and keeps your audience interested. For some reason, ChatGPT tends to repeat itself again and again with slightly different phrasing. That is useful when learning a new concept, but it can quickly become too redundant in this context.
Trust me, your lesson will be much better if you take the time to add your own touch.
2. Always give the AI precise context.
This means gathering good data sources about your topic, ideally using multiple references, and including them in your prompt. It is a bit like creating your own retrieval system, or RAG.
Do not use AI to generate content only from its own memory.
Why?
Because LLMs can sometimes hallucinate or present incorrect information. Imagine writing a lesson about a complex topic you do not fully understand.
If you rely only on AI, you could include errors without even realizing it.
For example, I once asked an AI to explain quantum entanglement, and it gave me an explanation that sounded plausible but was wrong. If I had not checked it, I would have shared incorrect information with my audience.
Make sure you understand the topic and have reliable sources supporting your claims. You can easily find information with tools like Perplexity or by asking ChatGPT to search for sources online.
This not only helps you verify the accuracy of generated content, but also lets you add references or links to your final lesson.
The more reliable your content is, the more your students will trust you.
3. Always use multiple sources to provide context.
As with an essay, do not limit yourself to a single article or book.
By providing several sources, ideally three or more, you give AI a richer foundation for generating content.
Remember how we researched the topic earlier? You can use that material as context for the AI.
4. Choose higher-quality AI models.
Use strong options such as GPT-4o, Sonnet 3.5, Gemini Pro 1.5, or Llama 405b.
Avoid lower-quality models. They are more likely to produce errors or hallucinations even when you give them good sources.
For example, GPT-4o and Sonnet 3.5 are excellent choices for this kind of task. Although Claude Sonnet 3.5 offers a free version, ChatGPT’s free version only provides a mini GPT-4o, which is not really suited to this task.
Creating effective prompts
Now let us discuss how to write good prompts for creating a lesson from your reference material.
Here is an example:
“You are a highly knowledgeable LLM expert and technical writer specializing in AI. Use only the following information to create a high-quality lesson. The text may include sources to use when writing the lesson or an existing lesson draft. You may need to write a lesson from scratch or provide feedback on an existing draft.”
“Here are a few additional instructions:”
-
Write in detail and use complete paragraphs, not only bullet points or listicles.
-
The article should use a professional style and avoid exaggerated language such as “revolutionary” or “mind-blowing.”
-
Do not use overly strong terms such as “amazing,” “intriguing,” “fascinating,” “must-read,” or “exceptional.”
-
Strictly avoid the following words: “embark,” “explore,” “vibrant,” “realm,” “endeavor,” or “delve.”
“Here is the source material:”
[Your source material here]
Reading and reviewing the generated content
Always make sure to carefully read every line of the AI-generated content. Typos and errors will affect the final result.
Check the accuracy and watch for possible hallucinations.
This may mean AI does not save as much time as you expected, but it is essential. You do not want incorrect information in your lesson. AI is not magic and still requires work, but it makes the overall process much easier.
Add a personal touch
After that, edit and improve the text thoroughly.
Add your own ideas and anecdotes. For example, replace generated examples with examples from your experience. This will make the lesson more engaging and personal.
Structure and style
Also check the structure of the text.
Make sure the lesson follows a logical order and that points do not repeat. LLMs tend to be redundant, which can hurt the clarity and appeal of your content.
Finally, pay attention to the writing style.
Make it engaging by using short sentences and simple words. Explain why the content is useful and why readers should keep reading. Structure the text like a story in which each action or consequence follows from the one before it.
Conclusion
By following these steps, you can create high-quality, engaging lessons more efficiently. But remember, AI is only a tool. You are responsible for making the lesson truly exceptional.
Thanks for reading, and happy writing!
FAQ
Why shouldn't you ask ChatGPT to write the entire lesson?
A model tends to produce generic, repetitive, and sometimes inaccurate language. Your plan, ideas, and review are still necessary to create a useful and credible lesson.
What context should you provide to an LLM for writing?
Give it the topic, expected format, style instructions, and several reliable sources containing the facts and examples it should use.
Why use at least three sources?
Multiple sources give the model a richer foundation, reduce dependence on one interpretation, and help you verify important claims.
How can you limit hallucinations in generated writing?
Tell the model to use only the provided sources, then verify every fact, link, and explanation before adding the text to your lesson.
What does a good writing prompt look like?
It defines the model's role, content type, source material, structure, tone, and phrases to avoid instead of simply requesting text about a topic.
How do you keep a personal touch when using AI?
Replace generic examples with your own experience, reorganize the text according to your logic, and use sentences that sound like the way you actually explain things.