Hi all,
As I am developing my own RAG solution, I was curious to see how chatgpt (web application) is able to process and analyze pdf documents which I plan to use for RAG. What I noticed when using gpt through web application is that the model is capable of giving a high level overview of the document and summarize the most important sections. At the same time, if asked, it can go into finer details of the document and provide detailed explanations of particular sections.
It seems like web application has the ability to dynamically decide whether to look at the document from high or low level perspective. How can this capability be replicated into RAG? For example, user of this RAG solution could be asking “What is document X all about?”. Only after getting an answer to this question user will be able to ask more specific questions about the document.
In general, are there techniques which can make the RAG solution capable of switching perspectives?
Thank you…