ChatPDF is the fast and easy way to chat with any PDF, free and without sign-in. Talk to books, research papers, manuals, essays, legal contracts, whatever you have!
Check it out at ChatPDF.com and let me know what you think!
Nice interface and crack at this problem, but the bot is deceiving as you can see here. Prompt engineering around omitting its internal database seems to help, but not always. May take some trial and error attempts to get it right.
That’s great, is there a maximum PDF file size? I’m assuming chunks of the PDF (based on search results) are provided for context (as per docs Answers Transition Guide | OpenAI Help Center)? Or is there some other approach used? Thanks for sharing.
When uploading there’s a maximum PDF size of 30 MB, but if that is actually 30 MB of pure text, probably the backend function will time out… so you have to try for large files.
Indeed, I’m using the embedding technique to augment the prompt with only the relevant paragraphs when answering a question.
Awesome app we are planning to implement same for a few of our clients and will be leveraging IBM Watson’s Discovery for the search result instead of OpenAI embeddings …wonder which is “better”.
Has anyone got a view of the Bot’s % accuracy (i.e. is it above 80%?) I believe this is difficult to gauge because there has only recently been released a Q&A knowledge base as a possible Q&A test baseline to measure against by IBM…and what are the possible current approaches there are to mitigating same i.e. disclaimers, citing the sources, fine-tuning, anything else?
It says my PDFs are too many pages, and “For more, use your own key.” But I’ve supplied my own key and it still says they’re too long.
Edit: Oh! I think it’s because it was using the URL input. When I upload the file instead, it works for a 280-page PDF. I don’t see why that would matter, but at least it’s working.
Really nice work. Could you talk a little bit about how you chunk up the doc? Was it a fixed number characters or some doc-depending strategy to semantically separate the file into pieces in a more meaningful way? Thanks!
Hi, it looks that doesn’t read all pages in the document. For instance, in some documents with 20 or 30 pages it keeps telling me that only have access to pages 1, 2 and 13, as an example. Whan I ask: what is the total amount of pages in this document, it tells me that has no access to this kind of information. Do you know how to resolve this? Many thanks
The App is amazing. Congratulations for the simple idea that changes the way we can read books. With this App, can we create a shareable link, which uses url to automatically instantiate the desired book?