Right now I’m developing an API for reading and answering large financial PDFs, however I have a question, if GPT reads a PDF and answers a set of questions, if in 2 days I ask new question, does GPT have to “read again” the PDF and therefore I will be charge twice for the reading? or can I save the model somewhere?
Thanks in advance!
You cant save a model since you are using API. What you do, you save chats. Every new chat with their ids. To which when you send queries to openAI it looks from chat history in order to complete your prompt.
And every time you do send a query, you sent to int small chunk which it will generate answer from that. You can’t send a whole document from that. And that chunk are done by embeddings. What openAI do, is to complete that answer which u embed with vectordb or whatever, then it complete sentence from there. Imagine pdf has milion of tokens, you can’t senfmd them all. Thats why we use embeds