Who does this subscription make sense for? I’m not processing 25 PDFs per day, every day. I’m using it occasionally to look at a few long documents. Using my own API key with ChatPDF cost me ~$0.50 this month, and now I can’t use it anymore…
Hello, i’m impressed of your implementation
I’ve tested it with an Asimov novel, just about the AI Multivac
I did a question after reading a large paragraph of one of the protagonists,but although i repeated the question in serveral ways, it always answered “Sorry, the source file doesn’t provide information about this…”.
So i suspect that you used embeddings to find the most relevant paragraph of the document and ask chatGPT to process a good answer based on that paragraph. Or maybe you pass 2 paragraphs? How many paragraphs are you passing?
I know that this is a “delicate detail” because each word you pass to the endpoint is computed as billed tokens but maybe the trick to get it run better is to pass more information. Or maybe the previous and the next paragraph to each one selected as most similar by embeddings.
Can you talk about these details? Thanks anyway! it’s a good example to follow!
Find the most similar document embeddings to the question embedding, how do you implement this function in node, I only know that you can use the numpy.dot() function in python to return the similarity between two vectors.
Perhaps if we aspire to write useful chatgpt API integration we should tske care of obsoletism and share more.
Join more groups that are fun and not gulag inducing.
Personally one of the greatest achievements we can make as progenitors of this chrysalis that we’re in is to build the ultimate symbolic AI SDK environment.
Break out the hand gestures of Close Encounters of the 3rd Kind, kids.
If we as a people accomplish this can you imagine what a PDF tool would like then?
Does any of the both solutions (ChatPDF / pdfGPT) offer it’s own API? My use case is to implement something to “chat” with the PDF file(s) programatically. Do I need to tokenize the PDF by myself and use the ChatGPT API directly or can I take benefit of any existing solution?
I entered my openai.com API key I tested already twice today.
It’s stuck with a barebones copy of chrome
says “we’re wroking to get faster” but my PDF is only 8 pages long.
I wrote my own Python script to feed chunks of the book to ChatGPT API and ask questions about each chunk and collect the results. Costs a lot less and I can customize it as I want.