Notices by Quin (thequinbox@dragonscave.space)

My personal stance on AI over the past 6 months has slowly shifted from being very strongly opposed to much more of an optimist. I still think that so much of the hype around it is overblown, it's being shoved in all sorts of places it doesn't need to be, etc., but since I started using AI coding agents, my productivity has gone through the roof. Notice too that I didn't say speed, because honestly I think I take longer to ship code now, but productivity. I actually get more done. I spend a while writing out a good prompt, let Claude run for 20-30 minutes while I go get some food and stimulants, then spend a couple/few hours at least tweaking the code, reviewing it, testing it, etc. But I can now just throw out ideas! If I want to quickly try adding oCR to Paperback, I can tell Claude to try it, go eat, then come back, read the code, learn not what to do when I try this later, and git reset. I forgot where I heard this, and I'm paraphrasing, but the quote goes something like: "AI does not create fast experts; it makes experts faster". Can definitely say that's been my experience. A lot of these dumb tiny AI startups will probably die, I hope AI browsers go to the software graveyard, etc., but LLMs are here to stay and honestly I've come around to it. I still get queezy when thinking about what it's doing to the environment, but also, humans are doing plenty of horrible things to the earth right now that I'm not even aware of that are taking out endangered species and eliminating lifeforms that make the chemical we need to survive, so with or without LLMs we're fucked and headed for global warming. That doesn't mean I don't think we should solve the problem, but it's not the only problem either.

@mrkrabs Have you tried RIM? https://getrim.app. It can share your entire system audio, work with any screen reader, and even let a blind user control a sighted person without them knowing!

@jonathan859 @mrkrabs Way easier said than done. Without going into detail, RIM has to do some pretty insane shit to allow for passing full system audio over the wire, and it also requires a much more stable network than just sending JSON packets with speech fragments does.

Blind Office 365 users: please, please tell me I'm missing something and there's some sort of setting I can tweak to make NVDA Not lag by half a second when trying to read any line in the text field in word? This is utterly absurd. I hit down arrow, hear blank. 500 MS later, I hear List Level 1 Solid Bullet. And it doesn't even read the list item I focused half the time! How and why is this so bad? I guess Copilot doesn't make lag-free edit controls...

Paperback 0.5 is out, sporting the largest changelog in Paperback's history thus far! We even had a new contributor majorly assist in adding a new document format! Full changelog:
* Added Microsoft Word document support!
* Added support for PowerPoint presentations!
* Fixed certain menu items not being disabled with no documents open.
* Fixed the orientation of the go to percent slider.
* Fixed the table of contents in Epub books with URL-encoded file paths and/or fragment IDs.
* Fixed whitespace being stripped from XHTML headings in weird ways.
* Fixed whitespace handling inside of nested pre tags in HTML documents.
* HTML and Markdown documents now support the table of contents feature! When you load an HTML/Markdown document, Paperback will build its own table of contents out of the structure of the headings in your document, and it will show that to you in the control+t dialog.
* HTML documents will now have the title as set in the title tag, if it exists. Otherwise, they'll continue to use the filename without the extension.
* Switched from UniversalSpeech to using a live region to report speech. This means no screen reader DLLs are shipped alongside the program anymore, and more screen readers will now be supported, such as Microsoft Narrator.
* Switched zip libraries to allow for opening a wider array of epub books.
* The dialog asking you if you want to open your document as plain text has been completely redone, and it now allows you to open your document as plain text, HTML, or Markdown.
* The go to percent dialog now includes a text field allowing you to manually input a percentage to jump to.
* The HTML parser will now recognize dd, dt, and dl as list elements.
* The table of contents in Epub books will once again be preserved exactly.
* The Unicode non-breaking space is now considered when stripping blank lines.
* You will no longer be asked how you want to open an unrecognized file every single time you load it, only the first time.
Zip: https://github.com/trypsynth/paperback/releases/download/0.5/paperback.zip
Setup: https://github.com/trypsynth/paperback/releases/download/0.5/paperback_setup.exe
PayPal donation link: https://paypal.me/tygillespie05
GitHub sponsors link: https://github.com/sponsors/trypsynth
Enjoy!

Version 0.3 of Paperback, my incredibly fast and light-weight ebook and document reader for Windows, is out! What's new:
• Fixed the table of contents in epub books with URL-encoded manifests.
• Fixed heading navigation in HTML documents containing multi-byte Unicode characters.
• Fixed high CPU usage in documents with long titles due to a regression in wxWidgets.
• Fixed loading UTF-8 text files.
• Fixed nested TOC items in EPub books putting your cursor at the wrong position.
• Fixed a crash on application exit in certain cases.
• Added a checkbox in the options dialog to enable or disable word wrap!
• It is now possible to donate to Paperback’s development, either through the new donate item in the help menu or through the sponsor this project link at the bottom of the GitHub repository’s main page.
• Markdown documents will now always have a title, and Paperback should now be able to load virtually any Markdown file.
• PDF documents will now always have a title, even if the metadata is missing.
• Switched PDF libraries to the one used in Chromium, leading to far more reliable PDF parsing across the board.
• You can now only have one instance of Paperback running at a time. Running paperback.exe with a filename while it’s already running will open that document in the already running instance.
• You can now press delete on a document in the tab control to close it.
Download: https://github.com/trypsynth/paperback/releases/download/0.3/paperback.zip
Enjoy, star, open issues, open pull requests, do what you do.

Let's be JAWS! Instead of writing the semi-complex logic that makes making word wrap in Paperback apply right upon closing the dialog, I'll pop up a dialog in your face that tells you you have to restart paperback, and when you hit OK, I'll just exit paperback and never restart it! Great idea and amazing UX, right?

This week, I've gone back to college for my first semester, and continued work as a software developer. And what do I choose to do with my weekend? Benchmark the difference in speed between std::endl and \n in C++ streams. Just FYI, in case anyone cares, endl is orders of magnitude slower than raw \n. Endl always flushes right away, which is fine if you're just writing a few lines, but every single flush makes a system call. Meanwhile if you just do it with \n, the OS can flush the buffer whenever it wants. On my 16-core Ryzen 9, writing to stdout directed to a file took 40 ms with \n, and 1624 with endl. Writing to an ofstream directly, e.g. a text file without redirection, \n took 27 MS and endl took 1585. And now you know that.

As against vibe coding and the like as I generally am, I'm not really worried about an LLM taking my job, at least anytime soon. AI's can write react components and Flask apps, sure. But I'm a systems level developer. I'm the kind of insane person who enjoys writing Rust and is looking at Zig during the hours I'm not working just because I enjoy the intellectual challenge of learning another hard language and the different ways of managing memory and low-level resources are fascinating to me. Five years ago I couldn't have predicted ChatGPT, so I'm not even going to try and predict what's going to come next. But an LLM replacing me seems pretty unlikely. Systems level work requires much more than just spitting out code. It requires hardware intuition and debugging instincts, something you can't get through training data. It makes you think about computers in a different way, and I think that if we get to a point where an AI can do what I spend 8 hours a day doing, the way we think about computers will probably have shifted in general, and it won't just be an LLM doing it.

I've just released the first version of my ebook and document reader application, Paperback. It's still in a fairly early stage, but I think it's good enough now to be called 0.1. Please do open issues if you find any bugs! https://github.com/trypsynth/paperback/releases/download/0.1/paperback.zip

Hello from Linux. More specifically, Fedora. Fedora, running on an M1 Mac with Asahi. This is cursed. Everything about this is cursed. But it works! Sort of. Sometimes. If I don't breathe on it wrong.

My older sibling: *opens their car window to give a homeless guy a dollar*.
My mom: "He's probably just going to use it to buy drugs."
My sibling: "I mean, I was just going to use it to buy drugs too. I get cash out for the bar, the dispensary, and drag shows."
That response made my week.

Imagine not having to worry about what OS you use because of accessibility so much that you can literally refuse to run stock Android for privacy reasons. Day in the life of a sighted nerd, a pipedream for blind ones.

@Fragglemuppet Did you recently start narrator? If so, restart NVDA. All will fix then.

You know what's a pain in the ass? Being an adult.
You know what kicks ass? Being an adult.

@nixCraft GNU-make and Go both say use tabs. So, use tabs, I say.

I wrote a CLI tool wrapping the NVDA.zip API, so you can just run nvdl to download the latest NVDA version, nvdl alpha to download the latest alpha, etc. It works on all platforms, and on Windows you will be asked if you want to run the installer after downloading. You can also use -u or --url to get the download URL only, not actually download the installer. https://github.com/trypsynth/nvdl

Recently got my internet upgraded to Ting Fiber, and it's been amazing so far. Super fast and super reliable. I have one super strange problem though. My main desktop computer will periodically drop the connection. It only lasts for about 10 seconds, but that's long enough to make TTCom timeout, to nuke my Putty session, and more. I've tried a USB wi-fi card as well as my on board wi-fi, they both do the same thing. The eero is sitting less than 6 inches from the antenna of my wi-fi card, and NVDA says I have 100% signal strength. I've tried upgrading my wi-fi driver too, and no luck. Anyone got tips here? I doubt it matters, but the network security type is WPA3-SAE, and this is the only device with such an issue.

My recent adventures with Rust inspired me to rewrite Easymark, my simple Rust utility to render a markdown file in your default browser, with many more features and better code. Here's the result, on GitHub and completely open source. If you run it with no arguments you're prompted with an open file dialog. If you want to skip this step, just pass the path to a file on the command line. In either case, the selected filed should be rendered in your default browser.
Source code: https://github.com/trypsynth/easymark
Binary release: https://github.com/trypsynth/easymark/releases/download/1.0/easymark.exe
Enjoy!

JAWS users, especially programmers: is there a way for me to make it report indentation much more reliably? I'd prefer a sound changing pitch like NVDA's indent beeps, but I'd honestly take it even reporting it via speech if I could make it actually see tab characters as indentation and if it worked in all applications.