Hacker News new | ask | show | jobs
by abadger9 1650 days ago
I've done some OCR side projects during hackathon weeks over the years (with google tesseract). This is a neat idea, I can only imagine the difficulty with which transcribing the variety of terrible handwriting will cause frustration and an eventual flood of refunds.
3 comments

Yeaaaah, his handwriting is actual certified art compared to mine . I can't get anything to recognize my handwriting reliably, including the text recognition built into iOS with Pencil - it's just useless.
To second what the other commenter said to you: I have seen someone with terrible handwriting fix it. He decided he cared, found pens and notebooks he really liked, and started using them attentively. People treat pens like they're all the same, but the drying time, line thickness, weight, drag across the paper, etc. all vary enormously, and if you care, you can probably find the right tool and get your handwriting to a point where you're happy with it.
I decided I hated having chicken-scratch handwriting around the end of high school / beginning of college. I literally did writing worksheets (tracing over letters) like I was in kindergarten. And if I wrote a messy word / letter on my homework I'd cross the word out and do it again. It made a huge difference, and I started getting compliments all the time on my writing. It's slipped back into average territory since, but it definitely works!
I disagree on the suggested method, but I agree that handwriting can be improved. Someone who writes in prescription cursive isn't going to get better with a new pen and different paper. It takes deliberate practice and attention to detail.

This is why "a poor craftsman blames his tools" is an adage.

I have terrible handwriting and have had the idea in the past to try and relearn with a new style. Like maybe the way they teach French children in school! But when I looked online for a book on how to learn handwriting as an adult they were all for veterans who had lost their dominant hand and that was incredibly depressing. Anyone have a good resource?
Just copy writing you think is good, and keep a daily journal where you write exclusively in that style. It'll be slow going at first, but it works. Takes about a year before it's totally natural.
My handwriting is much cleaner with a fountain pen than a ballpoint pen. Tools matter. It doesn't have to be an expensive fountain pen or paper, a cheap Faber Castell on printer paper works fine.
Of course your voice is going to sound much better on a condenser mic than on a $1 mic from RadioShack, but that doesn’t mean your voice changed. We’re talking about improving handwriting, if you write someone a note illegibly your excuse should not be that you forgot your good pen at home.
I can't even recognize my handwriting reliably.
Same - I find I typically use writing as a tool for thought moreso than a record that I will come back to, though I occasionally take a pass through old notes to see if I had any forgotten gems.
You can change that.

After I started using a Palm Pilot, my handwriting improved significantly and the changes seem to have been permanent. I get basically 100% accuracy with the Apple Pencil in iOS.

There is a delicious irony in the fact that you are training yourself using a reinforcement learning approach to meet the needs of an imperfect machine learning application.
An inexpensive bit of consumer tech. was able to accomplish something that years of human teachers could not.
The insta-feedback. I learned to spell because a little red squiggly line told me to try again. I avoided the right-click to autocorrect. Now everything just autocorrects and I assume people are more and more reliant on it.
I don’t see an issue in that, we can meet machines part way to make it easier for both of us
He really glosses over how he uses GPT-3 to correct the text...
It looks like he’s first using tesseract to recognize his handwriting and convert it into text. Tesseract doesn’t do a perfect job so the recognized text is full of mistakes. He treats the mistakes as spelling mistakes and “asks” GPT-3 to correct them. This is a very clever idea and will greatly improve current OCR efforts.
That's a really interesting part, and probably why the OCR works good enough for such case.

Not sure about legal implications of using it though:

https://www.technologyreview.com/2020/09/23/1008729/openai-i...

EDIT: it seems I have misunderstood the article - OP probably uses MS API to access GPT-3 anyway, so the point is moot.

They might consider it a trade secret of sorts. If I were them, I wouldn't want someone to just take the idea and undercut me.
I think from his explanation in the article it's quite straightforward to implement it yourself: get a GPT-3 subscription at OpenAI or MS Azure, use the API as described in the article, voilà.

But the idea is genius indeed.

From the screenshot I suspect the real secret is that it gives the user a chance to correct errors after scanning.
I have the same issue with my remarkable 2 : My handwriting is not OCR-able
Practice! I found a guide for writing letter like architect, helped me a bunch, after like 30 days of learning new letter shapes
Could you post a link to this guide? My handwriting while not horrible could use some improvement, and is not helped by the fact that I am left handed.
I'm uncertain if this is the one parent mentioned, but I found this guide with a quick search: https://artdepartmental.com/blog/architectural-lettering/
This is good too! I got books from the library.