Using openAI to create questions (not answer them)

[chinmay1]

Use case: I want to create a quiz (10 questions) using openAI on a specific topic. The topic details are contained in a 200 pages document. In other words, I want OpenAI to create a list of 10 questions based on 200 pages document that I feed. How do I do that?

1. If I try embedding, that would not work because embedding is very good at finding the answer from several embeddings. But it can’t combine several embeddings to ask insightful questions.

2. If I try fine-tuning the model, how do I feed 200 pages document to Open AI. I think the max is 10-12 pages. If I break the document to in several sections to fine tune, I am not sure it will be able to combine all the sections and ask “insightful questions”.

Thanks

Chinmay

[daveshapautomator]

I have two videos that might help:

This first one is about generating questions.

Fine-tuning GPT-3 to generate questions about anything

This second one is about extracting information from arbitrarily large documents.

Answer complex questions from an arbitrarily large set of documents with vector search and GPT-3

[chinmay1]

Contrary to what I posted, I was able to create questions (and very good one) from the embeddings.

[Wes]

This has opened my mind to how useful recursion can be in training the models, I can’t believe it didn’t occur to me before. Thank you.