Skip Navigation
InitialsDiceBearhttps://github.com/dicebear/dicebearhttps://creativecommons.org/publicdomain/zero/1.0/„Initials” (https://github.com/dicebear/dicebear) by „DiceBear”, licensed under „CC0 1.0” (https://creativecommons.org/publicdomain/zero/1.0/)DA
Posts
2
Comments
1,853
Joined
2 yr. ago
  • If they're local they'd be basically useless due to a lack of computing power and potential lack of indexing for a search engine chatbot, so I doubt it. It would also have to be so polished that it wouldn't require further user knowledge / input, and that's just not a thing with any local LLM I've come across. Mozilla can gladly prove me wrong though. I certainly wouldn't mind if they generally can make the whole process of local LLMs easier and more viable.

  • Yes, but what would a local model do for you in this case? Chatbots in browsers are typically used as an alternative / more contextualized search engine. For that you need proper access to an index of search results. Most people will also not have enough computing power to make use of any complex chatbot / larger context sizes.

  • All of those could be terrible to be honest, because AI is a data tracking vacuum. An AI adblocker or content filter sounds cool at first, but it would mean it reads and analyzes your data, just like the shit you do with chatbots too. Reading your mails? That's basically what Google does for years with gmail, that's why they have such a good spam filter. I agree that a chatbot would be kinda useless though, even if privacy friendly, which in of itself would be great but I just don't see the use. This could simply be outsourced to a website.

  • The answers to both of those things depends very heavily on the details. I think focusing on their main products is a good thing, but adding AI sounds like one of those likely terrible decisions. We definitely need privacy friendly & open source based AI though, in all areas, so I hope this is Mozilla pushing for something sensible here.

  • The bots (what the actual girlfriends or whatever other characters are) aren't the problem. You can find them on chub.ai for example or write them yourself fairly easily. The issue the software, and even more so the hardware. You need something like the mentioned Kobold.ccp or oobabooga, and then you'd also need a trained LLM model that you can get on huggingface.co, which is already where it gets complicated (they'll be loaded within kobold or oobabooga). You also need to understand how they work in regards to context sizes & bytes, because they need a lot, and I mean A LOT of vram to work properly. Basically, the more vram you have, the better the contextual understanding, their memory is. Otherwise you'd have a bot that maybe knows to only contextualize the last couple messages. For paid services like novelai.net you basically have your bots run through big ass server farms with lots of GPUs that bundle their vram and processing power, giving you "decent" context sizes (imo the greatest weak point of LLMs and it is deeply rooted in how they work) and decent speed. NovelAI also supports front-ends like SillyTavern which is great for local bot management and settings, regardless if you self host or use a paid service (NOT EVERY PAID SERVICE HAS AN API FOR THIS! OpenAI's ChatGPT technically does too but they do not allow NSFW content and can ban you for that if caught).
    There's a bunch of "free" online services too, like janitorai.com but most of them have slow speeds and the chat degrades significantly after just a few messages, because they have low context sizes. The better / paid models suffer from this degradation too but it is slower and less noticeable, at least at first. You can use that to get an idea of how LLMs work though.

    Edit: Should technically self explanatory / common sense, but I would advise not to share ANY personal information through online service chats that could identify you as a person!

  • I tried oobabooga and it basically always crashes when I try to generate anything, no matter what model I try. But honestly, as far as I can tell all the good models require absurd amounts of vram, much more than consumer cards have, so you'd need at least like a small gpu server farm to local host them reliably yourself. Unless of course you want like practically nonexistent context sizes.

  • Starfield @lemmy.zip
    DarkThoughts @kbin.social

    NG+ Temples limited?

    I'm in my first NG+ run. I've skipped the story, thinking it makes sense to grind out the temples first until you've maxed your powers and equipment but after finding 14 temples Vlad does not give me any new locations anymore and there's still 10 of the powers at level 1. I found a planet that has anomalies, but no temple quest triggered there either. So, can you only upgrade a limited amount of powers per run? If there's only about 14 temples per run, then I would need almost twice as many runs than the ones needed to get the last armor, no?

    Or is this just bugged and I'm basically stuck on that save?

    Edit: Bigot mods love to censor me I guess.

    Artemis App @kbin.social
    DarkThoughts @kbin.social

    Public Beta initial feedback

    Okay, disclaimer this is just a "not even 5 minutes in" type of thing at this point, but I wanted to start the complaining about a few things already. ;)
    Also, I use it without an account at this point. I don't know if that affects any of the app features right now, besides the obvious ones to communicate & vote of course.

    • Compact mode thumbnails are microscopic in size. Thumbnail size is already a sore spot for me because most Reddit & Lemmy apps are also pretty bad in that regard (likely focusing more on the big cards, dunno), but this is by far the worst example I've seen. It's especially sad considering that the desktop kbin page is the polar opposite to this. In compact mode it actually fills the entire space out that is available to it.
    • Thumbnails also cannot be swapped to the left side.
    • I love that there's an option to have all comments collapsed. However, I'd like it if I manually collapse a comment again (if we talk about the top level one), that it does not collaps