completely blind computer geek, lover of science fiction and fantasy (especially LitRPG). I work in accessibility, but my opinions are my own, not that of my employer. Fandoms: Harry Potter, Discworld, My Little Pony: Friendship is Magic, Buffy, Dead Like Me, Glee, and I'll read fanfic of pretty much anything that crosses over with one of those. keyoxide: aspe:keyoxide.org:PFAQDLXSBNO7MZRNPUMWWKQ7TQ
@dpk None of the alternatives care about accessibility. Codeberg continues to have accessibility issues. Not only do they go unfixed, codeberg keeps putting up inaccessible captchas, fixing them for a while, then adding yet another inaccessible "prove you're human" test a month later. Gitlab is just as bad as github, so no point moving there. Hosting my own repo means finding an accessible solution for issues, downloading releases, etc. All of the existing solutions have shown no commitment to accessibility, so I'm not willing to spend months putting in the work to fix them.
@cachondo I wanted to get the pile of gold coins and the pile of silver coins. In theory, get piles should have recognized that there were two different "pile" kinds of object in the room, and performed get on both of them. But instead it decided that "piles" is not the plural of pile, and invented a rule about a painful medical condition and gave it to me.
I hate English so, so much. Piles is not the plural of pile. It's swelling external or internal to the anal sphincter. Well, if you pick the wrong sense. And of course my rules do. Anyway typing get piles did not do what I expected or wanted.
TIL: apparently the reason AI suggests naming so many things Ember when you ask for ideas is because Ember means Human in Hungarian. So it showed up a lot in training data as a possible human name, and a lot in translation as a meaningful word. And also from the training data, humans want to give things meaningful names that sound human. So Ember it is! Is this true, I wonder? @Tamasg
@KaraLG84 I wish this was an option in settings. Sometimes text in other alphabets is just ascii spam for art or formatting or whatever. And I don't want my phone to load a new voice and switch to it just to read gibberish. So it's an improvement in that way. But I'd like to be able to set the behavior per activity, so I can have switching when I am, in fact, expecting to encounter text in other alphabets.
Finally getting to a place where the rules my mud dynamically invents mostly make sense!
rules for mine: takes: direct (touchable), instrument (carried, optional) means: get from the earth by excavation check: any tunnel it must not already be mined out everywhere it must not already be depleted everywhere you must be able to reach what you act on everywhere you must be able to act carry out: any tunnel extract ore from the tunnel walls any crawlspace mine the rubble for valuable materials everywhere the source becomes depleted of its materials
This may be the worst image description I've ever gotten from AI. I don't think this model is going to be a part of my workflow, guys. The tip everyone always gives to write good alt text is: imagine you're describing the image to someone over the phone. And now all I can think of is some deranged person shouting this over the phone at me. At least I don't have to upload the actual image; I'm sure you can all picture it in your heads based on this alt text.
Right! With the template updates, so I no longer have to read anyone's icky display names anywhere, #fedra is now the perfect fediverse client for windows. Uses standard notifications. Doesn't make nine thousand noises constantly. Easy to show and hide, with no hotkey steeling globals or invisible timeline. Just, you know, a minimal app for adults that behaves like an adult itself.
And I don't care how much you plan. You will always find problems, and need to re-plan, and then the plan will take you twice as long as you thought. AI fixes none of these things. For example, if you introduce pronouns, you're going to end up with things like "You give Jessica her the sword". And you need to account for "They give you the sowrd" and "she gives you the sword". And if you're keeping an NPC memory, as I am, you should really be storing events as "Jessica gave Britney the sword", not "Jessica gived", or worse not using the correct tense at all. None of this was in the beautiful plan I spent five hours on. Thank goodness I never need to account for "will give" and "is giving". Also Evennia had most of this stuff for free, so I just had to learn to use it, not build it from scratch. But it still derailed the plan and added a bunch of unexpected work.
Planning is always, always the hard part. If it isn't, even with AI, you're doing it wrong. I started planning at 8 AM this morning. By the time I did my own research and settled the entire plan, it was 1:30 PM. But now I have a clear enough plan that I can do it myself, or let the AI do it, or some of column A, some of column B. The advantage of AI is that it can hold more in its head at one time than I can, and can write out documents with more correctness than I would. When a plan exists, any bot or most humans can do the actual work. github.com/fastfinge/aimud/blob/master/docs/pronouns-and-ownership.md
Well, looks like I don't get to create anything today. Instead I get to spend all day figuring out if I agree with the AI or not. Again, you can learn from working with AI, but only if you want to. I understand a total of zero things in this statement, and I won't continue until I've done my own education and understand all of them: "Centering theory (Grosz, Joshi & Weinstein) asks exactly your question, and its Rule 1 is: if anything in an utterance is pronominalised, the backward-looking centre must be. The Cb is the highest-ranked entity from the previous utterance that also appears in this one, ranked by grammatical role — which maps straight onto roles you already have: actor > direct > target > container > source > instrument."
