This reminds me a bit of this thread: Poor quality response on trained LLM with pdf files
GPT 4 has a max context length of 128 kilotokens https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo
And then there’s embeddings…
It’s certainly not a technology issue.
I think “learn” is the wrong word, unless you want the model to just emulate your style.
I think “work” is a better word. Use the AI to work through your stuff.
In my opinion* the best way is to sit down and think about your process of creating a lesson plan, and then using the models to handle the legwork. Finding the balance between automation and manual work will be crucial - I’d advise to not automate too much up front, and just iteratively use the models to solve more and more complex tasks as you get more familiar with how to use them.
Good luck!
*The field of prompt engineering is currently somewhere between science and alchemy, meaning different people can give you different answers that can also be equally irrefutably correct.