Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

ChatPDF – Chat with Any PDF

chatpdf.com

Post reply on HN
Print view View original

91–100 of 141 posts

Re: ChatPDF – Chat with Any PDF

#91

Post by qxxx »

pedro_hab wrote: post #72

This was pretty bad for me, I tried asking the name of a person references in the PDF and it couldn't find it. I asked who is the claimant in this PDF and it said the claimant was empty. But if I asked if the claimant name was in the PDF it answered yes. I am assuming the PDF to Text is not working great here, which I supposed is the whole point.

yea, same here. I upload some test text. I asked how many children does my coworker have. It said "Your coworker has no children". I said but in the text it says that my coworker has 2 children. The answer was, "You are right, your coworker has 2 children as mentioned on page 2"

Re: ChatPDF – Chat with Any PDF

#92

Post by sergiotapia »

cs702 wrote: post #55
angadsg wrote: post #43

IMO folks are better off deploying their own version where they can adjust a few knobs (e.g. split chunk size) to get better results, given that PDF Q&A is such a commodity application. Wrote a https://github.com/angad/dharamshala/blob/main/docs.py return_source_documents is particularly helpful to get a sense of what is being sent in the prompt.

Consider adding a bit of overlap to the text chunks. Say, 300 tokens: text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=300) Otherwise, you'll likely end up with too many edge cases in which only part of a relevant context is retrieved :-)

We did chunks with a sliding window of previous page + current page + next page, with overlaps. That produced the best results.

Re: ChatPDF – Chat with Any PDF

#93

Post by croes »

pbhjpbhj wrote: post #39
taneq wrote: post #16

Earlier quoted context omitted.

You might not dump your internal documentation or confidential files to it, but I can see something like this being very useful if you can chuck a user manual for a product into it and ask common-sense questions about the product. So many parts these days come with a multi-hundred-page, questionably-written manual that technically does contain all the required information but buries it in waffle.

Or for legal contracts ... though no-one is going to go there with a commercial product unless they can indemnify themselves somehow against erroneous answers.

But can you trust ChatGPT's explanations of a legal text?

Re: ChatPDF – Chat with Any PDF

#96

Post by matchagaucho »

simonw wrote: post #20

How are you solving for PDFs that are too large to fit in the token context? I know of a few approaches for that: - Ignore the problem and let it hallucinate answers to anything that's not in the first 5-10 pages - Attempt to recursively summarize the PDF at the start - so summarize e.g. pages 1-3, then 4-6 etc, then if the resulting summaries are still too long for the context window run a summary of those summaries…

Chunk the PDF text and create embeddings. Get cosine similarity between user query and each chunk, and send the top N chunks to OpenAI that fit within token memory.

Re: ChatPDF – Chat with Any PDF

#98

Post by anymoonus »

Docalysis wrote: post #89
simonw wrote: post #20

How are you solving for PDFs that are too large to fit in the token context? I know of a few approaches for that: - Ignore the problem and let it hallucinate answers to anything that's not in the first 5-10 pages - Attempt to recursively summarize the PDF at the start - so summarize e.g. pages 1-3, then 4-6 etc, then if the resulting summaries are still too long for the context window run a summary of those summaries…

I can answer for my site ( https://docalysis.com/ ) which does a semantic search to figure out which parts of the document are most relevant. Then you just use those parts. Docalysis also shows you the PDF side-by-side, has page numbers, and overall responses are of better quality according to users that have emailed comparisons to ChatPDF.

Consider adding the ability to try your service before signup.

Re: ChatPDF – Chat with Any PDF

#99

Post by glonq »

dfaslkjvalkj wrote: post #88
glonq wrote: post #60

Other possibilities to fuel the ChatGPT hype train... ChatPNG - apply OCR to an image, extract text, feed it to GPT. ChatMP3 - apply speech-to-text to a recording, feed it to GPT. ChatGPS - hmm. not sure yet. something location-based obviously... If any VC's are interested, I'm selling 10% stake in these projects for only $20k right now. /s

Forget that I'm currently selling NFTs for these projects-- ChatTXT - extract text from plain text files and feed it to GPT for analysis. ChatPDF - extract text from a PDF document and feed it to GPT for analysis. ChatDOC - extract text from a Microsoft Word document and feed it to GPT for analysis. ChatDOCX - extract text from a Microsoft Word document and feed it to GPT for analysis. ChatPPT - extract text from a M…

I look forward to feeding your NFT's into my ChatNFT project :P

Re: ChatPDF – Chat with Any PDF

#100

Post by teacpde »

jeroenhd wrote: post #87
glonq wrote: post #60

Other possibilities to fuel the ChatGPT hype train... ChatPNG - apply OCR to an image, extract text, feed it to GPT. ChatMP3 - apply speech-to-text to a recording, feed it to GPT. ChatGPS - hmm. not sure yet. something location-based obviously... If any VC's are interested, I'm selling 10% stake in these projects for only $20k right now. /s

After the success of ColorGPT ( https://twitter.com/TheRundownAI/status/1640054184635449344 ) I don't think you need to bother with actually getting any AI into your app, just make sure the name ends with GPT.

From the project's readme:

> It uses ChatGPT API to generate color name from color hex.

It does use ChatGPT, but yeah, I get the sentiment.

Post reply on HN