Does anyone know a way to get an AI to read a PDF file accurately - short of copying and pasting the text from the PDF?

[msc545]

[Effably]

AFAIK both Claude and Gemini allow you to upload pdfs and will read and process them just fine. I assume most AI do that now.

It can get a tad messier if the information is stored in pictures and not text. Did you try at all? Claude 4 can do this with ease, often limited uses if you use a free version but probably the best free AI available to the public, especially deepsearching.

Gemini can also do this for free with no real limitations.

Did you try it or did you just ask before trying?

[msc545]

I've tried to get Deepseek and Chatgtp to do it and both made errors in parsing numbers

[Effably]

So my advice is still to use a google account and use Claude 4's deepresearch mode.

Failing that even Gemini's is fine.

The key really is to use deep-research modes. An AI normally uses a bunch of symbols, tokens, and calculates and infers the next one. Deep research modes also do that but basically re-run and do it parallel multiple times. The end result is the difference of a paper written by a grammatically apt high school student based on whatever wikipedia and the first google result said about it or a Bachelors, masters level university student who spent a few dozen hours researching.

Normally doesn't matter if you just want to know basic facts. But if you are doing any level of work requiring processing, absolutely always use it.

Google's notebooklm can also directly accept your pdf. Summarize, explain, turn it into podcasts, studyguides etc.

[msc545]

Thank you very much. This data consists mainly of numerical tables.

[Effably]

Absolutely. Good luck mate.

[msc545]

It turns out that Claude has a length limit, and my datasets are very large. I think Gemini is next.

[jshm2]

Depends on the text.

AI isn't a "proof reader" so it will not know the images, and what they mean to the text, nor what it all actually is.

Hence why it's recommended to split the PDF/DOC/Epub into chunks and chapters to make it easier for the AI to figure out what you want it to work on.

It's why LLM devs get paid more than AI devs. You can have a top tier AI - but it can't do anything without properly formatted data.

[Amy10223]

I am not sure what you mean, but I got AI, I don't know which one it was because I didn't expect it, to come up with the Epstein file list. I saved this. Yes, Trump's name is on it.

[Anonymous]

Watch Stephen Robles on YouTube, he talks about how to do stuff like that.

[ProbablyClueless]

Don't Gemini and CoPilot do a pretty good job with that? You just upload the PDF and let it do it's thing.