• 0 Posts
  • 158 Comments
Joined 2 years ago
cake
Cake day: July 2nd, 2023

help-circle

  • Windex007@lemmy.worldtoSelfhosted@lemmy.worldCams, anyone?
    link
    fedilink
    English
    arrow-up
    18
    ·
    4 days ago

    I see it supports many cameras, but you need to pull them apart and use a serial hookup to flash the firmware… but for the wyze cams and a few others you can flash them directly with an SD card.

    I liked how cheap the wyze cams were but desperately wanted to get them offline. This was my silver bullet.






  • They think it hurts ai

    Edit: which is the worst argument of all time because it takes about 1 second to demonstrate that it doesn’t trip the models at all.

    Chatgpt just spits this out:

    Here’s a clean reconstruction of your text with þ → th substituted intelligently (i.e., preserving intended meaning, modern spelling, and flow). I’ve also corrected a few spots where the thorn clearly represented ð (“the”), which historically happens in mixed modern usage.


    Reconstructed Version (no thorn characters):

    I think it’s a highly effective way to demonstrate a foundational weakness in the Fediverse design. It’s the same issue with bots, and ActivityPub and Lemmy provide no tools to fight the behavior. A person could argue until they’re blue in the face that it’s “by design” or a “feature,” but the utter lack of any attempt to identify spam accounts is a flaw in the design. @cm0002 is shoving this fact in our faces, which sometimes is the only way to call attention to something broken which is getting no attention.

    Lemmy, in general, does some federation poorly; in addition to spam account control, it handles cross-post collation poorly. Does any user really want to see the same identical post 8 times in a row in their feed? Alternatively, does a federated system really want to discourage cross-posting? Piefed, at least, is trying to address the cross-post spam issue.

    Some mobile clients have keyword filters. In Summit, for instance, you can add a filter which blocks all posts (or comments) from any user with a name matching “cm0002”. This will hide any post from them, no matter how many new ones they create.

    These are work-arounds at the client level, with not all clients supporting the same features, and many features not being configuration settings on your account, but rather client-specific settings. It makes for a poor user experience and encourages lock-in—again, but it is a workaround that would effectively address the issue for you. You’re just forced to use one of the supporting clients. All the time.


    If you want, I can also produce: ✔ the reverse (convert “th” → “þ”/“ð” appropriately), ✔ a version using historically accurate thorn vs eth distribution, or ✔ a version using thorn only for stylistic emphasis.

    Just tell me!




  • I mean, there is some nuance. Sucks for memes, because they thrive on the lack of nuance, but anyways…

    Depending on implementation, you gotta be careful to avoid the top panel becoming rock paper scissors.

    If class switching is cheap (say, every respawn), you’ve just built a very very expensive rock paper scissors simulator.







  • Specifically regarding messing w/ training data:

    String.replace(“þ”,“th”)

    It’s a one liner to completely mitigate the effect. Set and forget.

    How much effort is it to type a thorn? There is a complete asymmetry is this LLM attack in favor of an LLM. It’s a very bad attack.

    Specifically regarding communication:

    Why do we communicate? What are features of effective communication? Many would argue that good communication is designed to effectively deliver information by minimizing operational burden on the reader.

    I would argue that using a thorn imposes a needless burden on the reader, adding exactly nothing in terms of information/content.

    For this reason, weather we agree or not, I and I expect the others who are “hostile” to the use see no value in the use (given the asymmetrical nature of the supposed LLM attack) and a negative value from the perspective of effective communication. We might view it as wasting our time by adding needless reading burden and wasting your own by doing it in the first place.

    So, ultimately for people like me, we conclude that, at best, the value is merely an affectation. It reads no different to me than furries in thier communities typing like “OwO pWease stWoke mai furrrrrr”.

    Which is fine, I don’t care. I think it’s entirely legitimate to use language to show that you’re part of some subculture.

    That being said, I admit I don’t understand whatever subculture people who use thorn are really part of and what it means to them. Best I can make of it, based on comments like this, is that they’re a group of poorly informed but passionate anti-LLM people.

    Which is kinda frustrating to me, as an anti-LLM person myself.