I wish Remarkable took this idea -- they really oversold their OCR capabilities[1]. It works great in their support and promo videos, but I found the actual performance to be absolutely terrible.
[1] https://support.remarkable.com/hc/en-us/articles/36000266143...
But if you know you're going to use a language model after the OCR, then you don't OCR to a single character, but rather to a distribution of character similarity (e.g. the 90% least similar or clipping at a certain similarity threshold). Then the language model should have more to work with (although TBH its work becomes more complicated).
[1] https://tesseract-ocr.github.io/docs/Limits_on_the_Applicati...
[2] https://tesseract-ocr.github.io/docs/Improving_Book_OCR_by_A...
“ The error does occur because image segments, that are considered as identical by the pattern matching engine of the Xerox scan copiers, are only saved once and getting reused across the page. If the pattern matching engine works not accurately, image segments get replaced by other segments that are not identical at all, e.g. a 6 gets replaced by an 8.”
Then I only now realised I don’t actually want my notes public, I hope there is some form of access control built into this! :D
It would also be great if you could delete a page by taking a picture of the notebook entry completely scribbled out, or a video of tossing it into the fire. We need a product roadmap for this!
It also amazes me how some countries make it so easy to open a business. Doing so in Brazil would be a legal nightmare. Your projects are super inspiring and I always have a mind to leave my job and start doing the same. I should probably move to the US or Canada first, otherwise it might not be possible for me.
A company in Brazil needs to fit into categories. So if you have an e-commerce company that sells food, you can't use the same entity to provide a service, for example, the email provider with emojis the original poster did previously. You need a new legal entity for that.
If you have one of the companies not pay by itself, closing it is a nightmare. I have a company that is closed, not debts, not a single problem, for 15 years already inactive. That company is considered a liability for my current company and me as a person. Once you have 3 companies in your name, you start having trouble in Brazil, as you fall into "risk" territory for taking credit, opening accounts, renting offices or apartments, etc. And if one company wants to receive payment in foreign currency, you also have to generate quarterly financial reports, and all sorts of bureaucracy. Each step of the way you find new problems.
The rules are so extensive and so hard to navigate that you can't be the only person working on that. If you want to be like the original poster, you need an accountant that will charge you per company a fair amount of money, and it will still give you a lot of work to communicate with your accountant about each of the issues.
Yes, it is OK for someone that was always an employee to open a restaurant and have a living from that afterwards. Specially since it is a well stablished business category. But serially opening companies in Brazil is not a good time at all.
Disclaimer: I am not a professional in this area, I am a software engineer and I do have 2 companies in Brazil, one operating and one closed. Most of my knowledge is either self-taught or learning through my business accountant.
Closing a business is a bit of a bother, but it is mostly a question of waiting.
First you need to file all the paperwork. Then you wait. Once you get approvals, you can't do anything without a bank account, and opening the bank account is a lot of trouble too, because they want to triple check everything.
From zero to operational is a long way, and with lots of legal liability along the way. I have a company closed for 15 years that still counts as a liability to me when I try to do anything, like renting an apartment.
You might be right in "opening" paperwork only being kinda OK. But you certainly can't do it like the original poster, that codes for a week or two, opens a new business and move on to new projects in series. (Yeah, not every project of his is a new company, but IIRC the larger ones become new companies).
That brio pen; the light bulb illuminating your workspace; the words you write; they’re all extensions of your body — from transcribing your internal thoughts, through your extremities, through transmission and distribution.
If the author is reading this, I hope you’ll consider connecting with a few reputable online stationary retailers to offer a curated collection of reasonably-priced premium notebooks and pens. Additionally, consider expanding your original scope to include tags, which would open up your platform to paper-based automation opportunities.
If you’ve made it this far and are just looking for the best cross-platform handwriting recognition, check out the Nebo app. It requires a stylus, but it’s the best I’ve come across and it reads PDFs, too.
Check out TarokoShop's notebooks: https://www.etsy.com/shop/TarokoShop?ref=simple-shop-header-....
So I had it for years and it went unused. And then I used a pencil! (Mitsubishi 9850 HB, thanks for asking.) [1] Oh and it is the most fun in the world! It’s like writing on a pat of butter with the tail of a fox. It’s the smoothest thing you’ve ever felt.
Now it lives in the kitchen and I record meat temperatures and whisky cocktail ratios and it is my favourite book.
That is all.
Too bad that since I moved to America, I haven't been able to find paper that good for this affordable a price
But for the pen and paper I was manually transcribing it to digital (and still only transcribed about half of it). I didn't know OCR had gotten that good (and still suspect my writing isn't clean enough for great OCR).
But maybe I should give this a try, might be enough to get me back in the habit. Also trying to avoid doing as much typing lately (because of some arthritic-like pain in the fingers on one hand, although it's my writing hand :/)
I tried using Windows Speech Recognition, and it's unfortunately seems to be pretty garbage. Tons of mistakes I had to manually correct, couldn't say too much at all without it being so garbled I didn't remember what I really said to correct it manually.
But then I found out about built-in Apple Dictation on Macbook, which sends it to Siri, and I tried reading some old journal entries, and it's actually pretty darn good! I might be able to get through transcribing my other notes using it with minimal corrections. Just need to make sure you state punctuation, or else it doesn't really put any into it.
Still didn't seem that great for programming code though. Would be cool if I could find something decent for that.
Dragon Naturally Speaking (a paid product) is pretty good at normal dictation. My workplace bought me a copy when I was recovering from wrist surgery and it wasn't bad, especially since I could still use one hand. I've seen people mention using it for programming, with a bit of difficulty and a learning curve. But you can create your own custom commands, which is pretty much required if with keywords in a language that don't have a dictionary entry.
Oh yeah, camera quality makes a big difference.
On the downside, the Moleskin IDE the custom one the author had made in China are extremely biased against people with bad handwriting. More attention to accessibility may be required.
Seriously though this seems great if you have a bunch of notes and journals that you'd like to digitize. I have a small journal with my own recipes from over the years, and digitizing it has been in the back of my mind for a while-- if Paper Website can defeat my astoundingly awful handwriting.
"Paper Website: Start a tiny website from your notebook" https://news.ycombinator.com/item?id=29174478 (32 days ago, 271 points, 70 comments)
Makes me wonder how interesting the web might be if I just started doing that again, and how boring it might have been if I'd just had a working search engine back then.
Then again I also seem to remember getting bored enough with the web to only spend an hour or two on it at a time. Also I was in middle / elementary school so that might have played a role too.
It was nice back in the early days when you wanted U.S. state government information, you could almost always enter something like http://state.xx.us and get the state's home page, then explore from there. (Where xx was the state abbreviation.)
Cities were very often http://city.state.xx.us.
Now many (most?) states have vanity URLs, and the cities are worse. I think Chicago's changed its URL at least three times.
I guess it's what the people running it want but I find myself going there less and less every day and only look at a few curated subs
I don't really think it's that the niche stuff has moved away from the web - it's that nearly every functional discovery mechanism (that my now 30ish year old self knows about) has been captured by advertising or killed.
When all you ever get served up is links to the same drivel promoted by folks who have no honest interest or curiosity, but are essentially mercenary marketing/sales (sorry - influencers blegh...), then the web starts to feel like a bland wasteland.
Some of this is entirely related to being older - but I do genuinely think the current tech powerhouses on the web are trying their damn hardest to kill off any & all organic discovery mechanisms they can. Often through completely disingenuous means. If that fails, they buy them and shutter them, or roll them into the brand where it becomes the same drivel again.
However true or untrue all of the political intrigue, journalistic threats, etc., it's just crazy to me that such an innocent corner of the web that I loved so much in the mid-late 2000s was sending death threats to journos in London not 8 years later.
I also have something that I'm working on for this -- I've used my tool (which isn't productized yet, it's still ~4 templated repos) to launch 2 projects (in bio) so far and am working on a third product, but BulletTrain/lemonsqueezy are two much more polished options I know about -- maybe you should give them a try?
Is the thing you're trying to sell an API? Or is it a resource-provisioning sort of SaaS (they sign up, you spin off a persistent worker or something). I model them as the same in my thing (the provisioned "resource" is the API key), but I think that's one of the big differences in which tool will work easily for you.
Curious - I'm looking at your other projects as well and the design is quite good. Are you using a firm for design, or do you have any front end frameworks to recommend? For some reason design consistency the way you have it is extremely hard for me.
[1] https://docs.aws.amazon.com/Route53/latest/APIReference/API_...
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.
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.
I registered a few and messaged him suggesting it would be cool if I could transfer the Mailoji email addresses with a code so I could hide them in NFTs only the owner could see…I think by the very next day he added the transfer code feature (and didn’t fail to give it the attention of his own style complete with an emoji gift box).
You and the author might also check out the new ENS (Ethereum Name Service), it support emojis for use as crypto identity/wallets. "Triple pures" (three base-level emoji) are popular as a wallet address.
And I’m a giant fan of ENS as a protocol. FYI even though there is a 3 character minimum because they use Unicode there are a few hundred emojis that are technically 3 characters allowing for registration of single character emoji ENS names. Beware though Unicode also allows Zero Width Joinder characters so there are people who add them in bad faith in attempt to sell desirable names to unsuspecting buyers that don’t get what they think they are paying for.
Edit: the link was bad, but example of legit single character emoji ENS is [pirate flag].eth you can search it directly in the ENS App
But I think I can't use this service:
(1) It requires "editing" for web publication, and I know I'm too lazy to keep up with that.
(2) I find myself mostly scribble with pencil and paper, and write on computer. This is partially because my handwriting is in another dimension in terms of recognizability.
(3) I sweat a lot, and that ruins paper notebooks pretty quickly, normally within 2~3 months of daily use. (So I use legal pads.)
So, personally, I've always thought about the reverse: write pages with computer, and make a book out of it for archival, like, yearly.
Still, I'm yet to carry out this idea, because:
(1) I'm unsatisfied with currently available text-based document formats - either too limited (markdown), too biased (ReST), or too verbose(HTML/XML). I'm hoping to build something like Notion(block-driven) out of plain-text document format.
(2) I fiercely hate proprietary note-taking services and apps. I've already had enough headaches: pages lost, broken import/export features, backup restoration failures, etc. Never gonna spend a single penny on them.
Paper website will likely be cloned if it works.
And if I could plug my own project, you could also do it by just sending emails: https://publicemails.com.
do you make a separate LLC for each project?
how much did it cost to launch the business and what does is cost to keep running (lawyers/paperwork/admin)?
i'm scared of starting my own tiny projects, because of all the bureaucracy involved to even get started
I live in the US. To incorporate here, you file paperwork with your state. You don't need a lawyer, just send in the filing fee(s) with the completed paperwork. If I remember correctly, fees were somewhere around $100-$200. I have an accountant do my corporate taxes. He charges me $400. I file the sales tax paperwork myself on my state's web site. It's basically: How much do you owe us? And then you pay it. Most eCommerce storefronts keep track of the sales-tax stuff for you so it's easy.
You probably don’t even need the LLC, but I like having a bit of a legal umbrella (though chances are, no one is going to sue you unless your project gets big)
I don’t see anything in the text that maps to either in the output?
Something like this is almost a sweet spot of keeping the paper version as a "draft copy" while being able to enrich a digital version of your journal. As someone in the thread mentioned, being able to have private pages would likely encourage people to try it out for their journalling purposes. Otherwise, the project looks amazing!
This year I switched from fountain pens and clairfontaine notbook to E-ink tablet, Supernote A5x.
the main purpose is to keep the record in digital formats, And so far I am satisfied. The OCR sucks, though. I hope Supernote may adopt the GPT-3.
Computers are useful tools but they can be quite attention-destroying.
Sparing my hands from using a keyboard!
Brilliant idea!
Now _there_s a startup idea for you! How did you manage to do that? Much more interesting story :-P
im just wondering if GPT-3 can be used as a spell check for speech to text use cases
nice project!
Also this seems like a lot of effort for 4k a year. Of course, hopefully you get some more subscribers to make it worth while.
Well, the simplest and most greyface application is forms; you can define particular areas of each page as being particular form fields. If you're blogging, you might have a field for a "slug" that appears in the URL, for example, or a field for tags, or checkboxes for some tags (plus a special page to declare the meanings of the checkboxes). Or, if you're tracking expenses, you could have a checkbox for each expense category and columns for the date and the amount. For recipes, you might have a section for listing ingredients, with a column for unit of measure, a column for quantity, and a column for the ingredient name. Etc.
For me, the special feature of paper notebooks that cellphones and other computers suck at is drawing. If I want to draw a diagram or illustration, it just works much better on paper: my pencil point occludes much less drawing area than my finger does, there's no tracking error where the ink appears 2 mm to the side of the point, it has much lower latency, and I can draw finer lines. But scanning those drawings into a computer is a pain, because I have to illuminate them evenly and hold them flat while I photograph them, which still probably involves some perspective distortion. Barcodes on the paper, together with reference lines and reference color swatches, could solve that problem, as well as providing information about which parts of the paper are occluded, if any.
For a few special applications like numismatics and entomology, the paper could provide a precise physical measurement reference for specimens.
Combining drawing with filling out forms, you can make a font from your handwriting; this is enormously easier if you can correct the various distortions. In http://canonical.org/~kragen/oilpencil/ I spent about 24 hours fiddling with various graphics programs, but there was a website I found somewhere where you can print out a form, draw the font on it, upload the scan, and download your TrueType font. This kind of thing might help with training OCR, too, especially if you don't have access to GPT-3. (Or if OpenAI decides to peremptorily destroy everything you've built because one of your users uses your service to write about their dead fiancee: https://towardsdatascience.com/openai-opens-gpt-3-for-everyo...)
Other ways to combine drawing and filling out forms include sketching orthographic projections to build 3-D models; coloring a coloring book; drawing maps for Minetest and similar grid-cell games (especially 2-D ones); drawing heightfields; and sketching different keyframes of an animation to automatically morph between. You could even draw a 2-D continuum of keyframes, thus providing an animation character that's continuously variable along two different axes; you might put time on the theta axis and some sort of emotion along the radius axis.
(You can also apply these ideas with drawings that are input via other media, such as touchscreens, Wacom tablets, and mice, not only scanning paper. When you're scanning paper it's hard to get feedback as you're drawing, although you could maybe glance at your cellphone screen periodically, or use a projector like DynamicLand, or have a continuously updated monitor using a webcam feed. It could even use the occlusion information from the barcode to patch in remembered images wherever your hands were occluding the paper.)
What should the barcodes look like?
In 02001 Anoto announced their "Digital Paper" approach: https://www.wired.com/2001/04/anoto/. As explained in https://en.wikipedia.org/wiki/Digital_paper this uses an unobtrusive 2-D barcode scanned by a camera in a "digital pen" (later called the "Fly Pen", 02005) to locate the pen in an enormous global "virtual desktop"; I think the NeoLAB "Neo smartpen" works the same way. This was all before cameraphones went mainstream and high resolution. They got 300 patents but fortunately everything they filed in 02001 expires this year. Anoto's barcodes use a grid of slightly displaced grid dots.
The Fly Pen provided a sort of graphical user interface on the paper, using audio for output. It was sort of aimed at kids doing schoolwork and playing games. It failed in 02009. The founder started a new company called Livescribe focusing on notetaking; the Livescribe smartpen allows you to spatially organize and annotate a continuous audio recording. It has been more commercially successful: https://en.wikipedia.org/wiki/Livescribe
Tiny unobtrusive dots might not reproduce reliably on a cellphone camera, though having been published in WIRED in 02001 means the technique is in the public domain (or will be next year). A better idea might be to use thin horizontal and vertical grid lines whose thickness varies slightly, perhaps in a pastel subtractive primary like cyan, magenta, or yellow; then you can optionally remove them in software after scanning. Scanning a whole page at a time, instead of a tiny area around a pen point like the "digital pens" described above, gives you a great deal more space for redundant page ID data in the barcode; probably 48 bits or so is sufficient.
Like, you've still got to edit the page to add links, images, colours, etc. In fact, that seems like the most complicated part.
It's a fun gimmick and a nice selling point. But someone will be able to use this idea for Wordpress/Wix/Squarespace plugin. Would be surprised if they didn't produce their own feature to do the same thing eventually.
It'd be way cooler if you could draw links on the page and it would figure that out. Or draw a box with some links in it and generate a header. Draw a box for a place holder image and generate that. The next iteration would obviously be using paper to design the actual site and then using CV to generate the markup/styling.
* I primarily see this as, not a traditional website builder, but as a way to take paper journals and transform them into web journals/blogs.
* As someone who has tried all manner of tools and software for journaling (literally like tens of software apps), I've found that my brain just works best with a nice paper journal and a great pen (I highly recommend the prismacolor premier fine marker). It just works far better for me than anything else.
I'm so glad this project is at the top of HN. It is the ultimate "hacker project" of someone that scratched a personal itch, and found out other folks would be willing to pay for it.
You could name pages like "Home", "About", "My trip to Paris". And then write on the piece of paper "[link to My trip to Paris]"
Or you could write "[link from Posts]" to add it to a common listing.
You'd basically just create a written form of Markdown.
When I'm reading a post, I rarely click any links at all.
I was a little surprised that the paper part is used only for plain text. It would sort of make sense to have at least a few formatting and layout features.
Here is dang's comment about that topic/meme [1]
It's cool, I've never shipped a project like this, and probably never will. But I've worked on my fair share of software so I know what it takes behind the scenes.
Oh well.