What is the best approach to fine tuning LLMs like BERT and ChatGPT for use as an expansive item bank for only middle and high school curriculum subjects? Not the specifics, just the approach. I am probably just going to use one shot and follow the directions. I am building a novel intelligent tutoring systems where this will be the expert domain for subject mastery for the bots to pull from.

I have just been Googling and reading primary literature papers. It Seems fairly straightforward.