Hacker News Reader

Comment by dgan | Hacker News Reader

Tried Claude yesterday to help me extract rows from a financial statement PDf. Let's automate boring stuff !! After multiple failures , I did it myself

View on HN

I sat down at a Piano yesterday and tried to play a beautiful song even though I never really used a piano before. Sounded horrible, must be the pianos fault.

One reason to use LLMs is to understand when and how to use them properly. They're like an instrument you have to learn. Once you do that, they're like a bicycle for your mind.

except the piano's seller advertised that i don't need to know how to play "just ask it, it will play!"

Here ya go, fixed your analogy

LLMs can't disappoint. They can only be disappointed.

Which model have you picked to learn and keep using? Because each new model means learning new instrument.

ah yes, all those LLM models have such unique and different system prompts, impossible to re-use your knowledge from model A for model B...

or wait... https://github.com/elder-plinius/CL4R1T4S nah, they are very much the same when it comes to prompting

system prompts aren't actually part of the model at all; obvious you don't know what you're talking about.

wauw. your point was that you needed to relearn "all the prompting" when a model changes. which is not the case. but hey. at least you know what you're talking about.

It was actually mine point, not theirs. But yes, you do actually need to lelearn "all the prompting" when switching models. Would gladly see any tests you have made or any published studies that show it is not the case anymore with newer models!

If you buy a machine to play the piano for you, you won't learn how to play the piano. You'll just become really good at using that machine.

LLMs are like a child with undiagnosed ADD, it is a constant roller coaster with peaks of amazement and and equally dips of disappointment.

Currently my biggest gripe is asking AI to make one modification to file, it then decides to do 4 other "extra" changes I didn't ask for. AI please stay focused!

As with any other tool, half the art of using LLMs productively is the ability to recognize when they're a good fit and when they're not. So long as AGI doesn't come around, there will always be cases where they fall on their face and as far as I'm concerned, we're pretty far from AGI.

Think how much energy would be saved if you could embed a sqlite database in a pdf with the data in it.

There are formats that are structured - XBRL.

If you want it to be human readable you can use iXBRL which is embedded in HTML.

Try Mistral OCR. They trained their model to extract data from PDF. Used it to convert pdf forms to json schema. Worked excellently.

i will give it a try tonight & report back!

Okey so Mistral is definitely better at reading PDFs and extracting usefull information

Most people don't realize that LLMs by design were not made for document processing, data extraction etc. For that, you would have to use a dedicated tool like Klippa DocHorizon, which built its own AI OCR from scratch. It also provides an API that you can use to send your documents and receive formatted data. It's less popular than, say, Textract or Tesseract, but it's far more accurate, especially if you're dealing with sensitive data that you wouldn't want an LLM to hallucinate.

Claude is not good for thus, try Mistral OCR. I'm using it on a daily basis on English and non-English texts and it works very well.

Did you try Claude to just extract it or to write a python script to do so?

Reply on news.ycombinator.com