Hacker News new | ask | show | jobs
by Maxion 760 days ago
And, today, with LLMs it'll take you a few seconds to digitize the document, too. For this reason I've also been considering a typewriter...
3 comments

OCR has been a solved problem for years. Long before LLMs started being hyped.

At least from typewritten documents that you did not torch or shred etc.

No it hasn't. Just 1.5 years ago I tried all the latest OCR tools, including AWS, GCP and Azure services, and none of them could consistently and reliably read a receipt printed at a store.
Receipts are hard.

- cheap paper

- cheap ink

- misprints

- abbreviations

- every store does it differently

Yes. Which makes OCR not a solved problem.
OCR is merely step one.

Interpreting recognized characters is another matter.

I was OCRing documents with ABBYY or Tesseract in 2000s if not a little bit earlier. I have been OCRing text documents with my phone for the last 6 years or so, with Prizmo.

It was taking seconds back then, too.

Reminds me of those tablets (and pens) that you can write on / with and they automatically digitize and OCR whatever was written, as if by magic.
Does one exist that actually works?
The iPad, with the Apple Pencil is pretty much there. It’s actually amazingly good. I have terrible handwriting, and it doesn’t seem to have a problem with it.

If anyone ever tried using a Newton, there was a series of Doonesbury comics[0] about its awful handwriting recognition.

[0] https://www.computerhistory.org/revolution/mobile-computing/...

I got pretty good at writing with the Newton, but it was me adapting rather than the Newton understanding my natural handwriting (which is fairly neat given my parents are both teachers).
Yup, iPad and Apple Pencil do an amazing job, either with the built-in Notes app or several third party apps. Even better with a screen protector like the Paperlike that gives a little tooth to the screen to make it a bit more like writing on paper.