2

I have a lot of PDF files collected using such tools as Mendeley or Zotero (thousands of files). I want to use ChatGPT-like to ask questions based on these PDF files (for example: "What is the biggest challenge in multimodal affective computing?"). I know I can use a tool like chatPDF, ChatGPT AI PDF extension, privateGPT, local GPT, or the similar. But none give local access to abundant local files (as long as I know). Some can only retrieve from one or several (not large numbers) PDFs; some need to upload PDFs (not offline).

I like elicit.com and used it, but I must upload my PDFs to get answers from my files. Their answer/reply is also almost as good as without uploading a PDF, but again, it is not local/offline.

Is there any ChatGPT-alike that we can use to ask from the given large PDF files? I believe this kind of approach will speed up academic writing from the current workflow (I used control+F in Mendeley instead of asking to chatGPT, or maybe Mendeley/Zotero could build their AI assistant inside the app in the future).

6
  • Are you willing to pay money for such a product? Commented Nov 15, 2023 at 3:44
  • If it is a one-time payment, yes; if it is subscription, no. But prefer open-source if it is available.
    – sugab
    Commented Nov 15, 2023 at 4:11
  • 1
    Why do you rule out solutions like privateGPT? As far as I can tell, it can ingest large numbers of local PDFs. Is the issue with restricting the scope of the search to only those documents, or just with the quality of the embedding?
    – Anyon
    Commented Nov 16, 2023 at 6:53
  • Good dreams, I hope they become true. right now, any tool that does that is unreliably if you are looking for factual information. Chagpt is a stochastic parrot that spits things that sound correct and are most of the time not. Commented Nov 16, 2023 at 17:57
  • 1
    @sugab The related h2ogpt is probably more performant, but I doubt it would be fast on reasonably priced hardware.
    – Anyon
    Commented Dec 1, 2023 at 3:21

0

You must log in to answer this question.