I have a task to take reference material (maybe 2,000 tokens long) and generate questions and other learning activities about the reference material. I don’t think RAG will work since I need to reference the whole document. Since they’re relatively short, I’m including the material in with the prompt.
Considering my use case, is there a better practice I should be following than using the prompt for everything? Can I use fine-tuning but still reference specific documents?
You don’t need RAG for this specific use case. RAG is generally not needed unless you need to reference a large number of documents or an extremely large document, and the outputs from a RAG system will never be as high quality as providing the FULL context(reference material) in the prompt.
Simply create a prompt template on a Google doc or similar - which describes what you want the AI to do (you can include examples).
In the prompt template have a placeholder for the reference material like “reference material”
Replace the placeholder with the actual reference material and paste whole prompt into the AI.
You can refine your prompt over time.
Another way to approach this is to create your own custom GPT to do exact same thing.