Lanie

@RareBird15@allovertheplace.ca · Joined ⁨Feb⁩ ⁨2025⁩

#Christian | Totally #blind & #autistic | Living with multiple #chronicIllnesses in long-term care. #Writer (30+ articles on faith, tech, & disability) & #UsabilityTester for Fable. Developer of WordPredictor & WindowState #NVDA add-ons. Creator of Everrealm, a screen-reader-first strategy game. Advocate for #DisabilityRights. Human to Squeaker (Miniature Pinscher). Runs support groups for multiply disabled people.

Website
lanie.work verified
Work
Accessibility Tester at Fable | Writer at Lanie: Faith, Tech, and Advocacy
Projects
WordPredictor & WindowState (NVDA add-ons) | Everrealm (accessible game)
Profile and Header Pics
Profile: Me with my Miniature Pinscher, Squeaker. Header: Squeaker wrapped in a blanket with only his face showing.

Replying to @⁨lnkr⁩

@lnkr @menelion I actually used to try to clear out the trigger words and some of the other patterns, but I found them hard to recognize at times and often wound up exhausting myself just trying to do that.
What you said about having to hide the tool's fingerprints is exactly it. I use the thing to get my own thoughts out, and then I have to scrub it so people don't see the tool before they see me. That's backwards. The tool is supposed to be the accommodation, not the thing I have to apologize for.
I don't have an answer to the tension you're describing. I think it's real. The same technology is my screen reader for writing, and it's also the thing flooding the internet with slop and hurting creators. Both are true, and I don't know how to separate them from the outside any better than you do.
What I do know is that "just use your brain" isn't the answer. It assumes a brain and a body that work the way the speaker's do. Mine don't, not every day. And the people who say it aren't trying to tell the difference between thoughtful use and slop. They've already decided there isn't one.
I appreciate you actually reading it and thinking about it. That's more than most people do.

WordPredictor v1.7.0 is out! 🎉

Big accuracy improvements this time:

  • Retrained the LSTM on 20 million tokens of modern movie and TV dialogue instead of 19th-century literature, so predictions sound like how people actually talk.
  • Extended the context window from 4 to 6 words.
  • The LSTM now blends with the n-gram model instead of replacing it, combining context-aware suggestions with words learned from your own writing.
  • Fixed a long-standing bug where accepting a prediction corrupted the context, making the next prediction ignore everything you'd typed before.

github.com/RareBird15/wordpred

#NVDA #Accessibility #BlindTech #AI

github.com/RareBird15/wordpredictor/releases/tag/v1.7.0

WordPredictor v1.6.0 is out! 🎉

This update adds an optional LSTM neural network predictor trained on 116 Project Gutenberg books. It provides context-aware predictions that complement the n-gram model. Toggle it on in Settings > Word Predictor.

The n-gram model remains the default. The LSTM is opt-in.

Bundled onnxruntime DLLs are included, so no extra install is needed.

github.com/RareBird15/wordpred

#NVDA #Accessibility #BlindTech #AI

github.com/RareBird15/wordpredictor/releases/tag/v1.6.0