• 5 Posts
  • 26 Comments
Joined 17 days ago
cake
Cake day: February 6th, 2025

help-circle

  • I also think the point here is considerably more nuanced than non-passing=bad.

    I read it as a critique of how trans people are stereotyped as non-passing. The fact of the matter is, a lot of us do pass and go about our daily lives, and the average cis person doesn’t even realize. We assimilate more often than most would admit.

    So while I don’t think there should be moral value to passing per se, I also think it’s pretty fucked up the way we’re all sort of stereotyped as being the most visible until proven otherwise…


  • To be honest I’ve never tried this on a PC.

    Also as a heads up sorry if you know already what I’m about to say, I don’t know what your background is so I’m kinda assuming it’s “tech literate but hasn’t done a project like this before”. If it is too complex or too patronizing please correct me by asking questions or telling me what to skip next time, respectively.

    I think most companies handle training by renting an instance on a cloud service provider and then running the training there. That’s the expensive part of development but can be mitigated by being smart with how often & what sort of training you choose. At minimum, a lot of debugging can be done on a PC.

    Inference tends to be cheaper, so perhaps that could be done on a PC. I’ve never tried personally but could help look into it.

    An application like this is usually referred to in industry as a Retrieval-Augmented Generation (RAG) LLM. The premise is you have an external database—such as a wiki, training materials, or set of news articles—and you teach the LLM how to reference that database when generating it’s replies. The database does not need to be large and you can also put in place measures such that it says “I don’t know” if unable to come up with satisfactory grounded answers.

    Something to consider would be what sort of database you’d like to reference. I know someone who has a complete backup of the Transgender_Surgeries wiki including the posts it links to for example. I think that would help that community as there is an outstanding issue of most users finding it hard to read the wiki or simply not doing it. If you wanted to create an MVP I would be down to help advise with that and provide guidance/support if stuck or possibly if in need of funds.

    I’d do it myself but my free time is mostly taken up by doing some online community organizing & having a boyfriend who exists in meat space. Would rather support/advise instead. I might also be able to pull some IRL contacts from my workplace (with connections to the trans community) as needed, can’t speak for them though.

    It’s a really great idea tbh. It’s well defined & frankly I wish more clients came to us with projects this well defined with this good of a dataset. Usually it’s not so straightforward. If you decide its something you wanna keep working on please keep me in the loop. Otherwise I might circle back to it myself & will definitely ping you.

    EDIT: some more notes re: the dataset. Once a working MVP is up using something like the wiki, we could figure out how to increase the scope of the database by e.g. ingesting subreddit data (look up Arctic Shift project on GitHub for ways around the API issue) or, better yet, using Lemmy’s APIs since those are all still open (as far as I understand it—still new to Lemmy). Compiling such a dataset and self-hosting a chat bot like that could be a way to ensure community knowledge lives on even if, say, Lemmy instances go dark as the platform evolves.




  • Honestly, it could be centralized for sure! Though I was also thinking more of just handing them a list of links to distribute on their own pages.

    I think for most communities the first step is probably to have any sort of website (even a super basic HTML one) self-hosted outside the US for their members to bookmark. Even if they don’t migrate until a ban, their members can at least visit the site and see where the mods decided everyone should internet-meet up.

    If they then wanted to link a list of additional backup resources, that could help spread by word of mouth where to go if other resources are down. I’m basically thinking some members will get lost and isolated if one day their main safe spaces vanish & we should try to limit that problem so long we have time on our side.

    re:LLMs it’s an interesting idea IF you use an open weight model and take steps you’re not sharing the data with tech companies. I don’t for example think we want to help ChatGPT find out where all the trans.

    But it is certainly possible to do a RAG LLM system IF you use an open weight model AND you have a dataset of trans resources! I’ve done these things before and don’t consider it too hard. You do need to have the dataset already compiled which can be a fair amount of work to begin with—after that it’s straightforward.

    Right now there are efforts to back up a lot of resources in trans spaces that might go dark. I could see it being nice to for example convert the wikis and subreddit dumps into chatbots you can ask questions and get back answers with citations to where the info comes from.

    It could help solve some of the issues faced by people e.g. not checking the wiki and then being confused…

    EDIT: the more I think about it that’s a great idea and I’m adding it to my backlog. I could do it myself, don’t have much time, but could definitely advise on it and maaaybe pull in some real life contacts who do it as their career. Let me know if the LLM thing is a project you want to work on & I’ll try to facilitate





  • So this is a kinda interesting thing. I’ve had two people in my networks who expressed hesitation because of UX. One was excited to find out that her fav 3rd party apps moved here. The other was excited to find old reddit.

    I genuinely think a lot of people feel new Reddit is just enshittification they don’t need or want.

    I had a few questions though, if you don’t mind me picking your brain!

    • How does this old Reddit cloning work? Is it something each instance can decide to host, or is it native to Lemmy?
    • If it’s tied to a specific instance, how do you know that this option exists? Is there a general way to find out specifics like this that I can point people to? It looks like it’s listed in this instances sidebar. Is that how it usually works on most instances?

    I’ll pass the info along to at least one person!



  • I don’t know about the average person but certainly enough people yes! But most don’t try unless they have a bit of encouragement and community

    I’m actually surprised at how many people will say they think Lemmy is unuseable. I have found that instead of explaining the fediverse they respond better to “it’s Reddit but better because it’s not beholden to any company or country” or something similar