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/)BO
Posts
2
Comments
129
Joined
2 yr. ago

  • how they fool the AI while keeping it invisible to the human eye

    My guess is that AI companies will try to scrape as much as possible without a human ever looking at the data.

    When poisoned data start to become enough of a problem, that humans have to look over very sample, then this would increase training cost to to a point where it's no longer worth to bother with it in the first place.

  • Workflow/Prompt

    I asked chat gpt + dalle

    the topic is “Sustainable Ecumenopolis”. Come up with a novel concept for this.

    The make some award winning high-quality images about it

    then I rejected the first batch of result and said

    vertical farming and solar panels are stale and boring.

    the images should be award winning high quality and atmospheric.

    then it generated some images. The one I selected for the competition was this

    Render: Atmospheric night plaza in Biosphere Nexus. Architectural structures inspired by biomimicry surround the area. Trees with bioluminescent leaves light up the ambiance, with diverse citizens enjoying the scene.

    If I win, the next topic will be "Torment Nexus"

  • what is the bare minimum of security measures you can do?

    I guess just the normal things with p2p stuff: make sure no ports are exposed except for the essentials, update software, use SSL wherever possible.

    When you don't use VPN, people will see your actual IP adress and will launch the same kind of attacks, they also launch on servers [1] to try to hijack your system and add them to their bot net.

    [1] port scans, login-attemps, applying known exploits. If this doesn't sound scary, you should try operating a server that is exposed on the internet and then look at the number of login attemps.

  • Thanks, I was not aware of this. This changes a lot.

    Could you point me in the right direction? (i.e. are there good/neutral articles elaborating on this specific case?)

  • Great adoption of the manga. I am hyped for the next episode.


    The way demons are depicted reminds me a lot of LLMs (and AI in general).

    • Less powerful demons have no understanding of human thought and are really bad at fooling people (see the girl-demon in the flashback).
    • Medium powerful demons still have no understanding of human thought, but are somewhat good at imitating people (see the present-day demons from todays EP).
    • More powerwull demons... ::: spoiler Spoiler start to develop more complex strategies for fooling humans. :::
    • While the most powerfull demons ::: spoiler Spoiler are implementing strategies that trancent both human strategies and the original demon strategies. :::

    this article by scott alexander does a good job of explaining how AI's are inhuman intelligences, that are trying to fool humans into thinking they have morals. https://www.astralcodexten.com/p/perhaps-it-is-a-bad-thing-that-the

  • I like it. you can visualize sizes with 3 orders of magnitude between them without one being microscopic.

    What makes this graph shitty, is that the spheres don't look very 3D.

  • is religious fanatics fighting religious fanatics. Fuck all of 'em

    I am with you.

    Though it would be nice if refugees from palestia would get asylum in Europe (which they don't), then the non-fanatics could get out of there, while the crazies do their thing.

  • I generally agree with subscriptions. "if you're not paying for the product, you are the product".

    But there are better platforms to support with your money. Youtubes main selling point was that it was free. Now that it is not, it's a good time to look into paid alternatives. For music I am already using bandcamp and spotify. For general documentaries I plan to switch to Nebula/CuriosityStream (I planned this for some time now, but I this may be the nail in the coffin)

  • I recommend to use relevativ paths in the compose files. e.g.

     
              - '/home/${USER}/server/configs/heimdall:/config'
    
    
      

    becomes

     
              - './configs/heimdall:/config'
    
    
      

    you may want to add ":ro" to configs while you are at it.


    also I like to put my service in /srv/ instead of home.


    also I don't see anything about https/ssl. I recommend adding a section for letsencrypt.


    when services rely on each other, it's a good idea to put them into the same compose file. (on 2nd thought: I am not sure if you already do that? To me it is not clear, if you use 1 big compose file for everything or many small ones. I would prefer to have 1 big one)

    you can use "depends_on" to link services together.


    you should be consistent with conventions between configurations. And you should remove config-properties that serve no purpose.:

    • you don't need to specifiy "container_name", when it would be same name as the service
    • PUID=1000 and PGID=1000 shouldn't be needed, I think.
    • sometimes you add explicit ":latest" to the version, and sometimes you don't

    while you are at it, you may want to consider using an .env file where you could move everything that would differ between different deployment. e.g.

    • PUID
    • TZ
    • exposed ports, maybe

    consider using podman instead of docker. The configuration is pretty much identical to docker-syntax. The main difference is, that it doesn't require a deamon with root privileges.


    you may want to consider to pin version for the containers.

    pro version pinning:

    • no unexpected changes, when you restart the container (e.g. because you accidentally pulled)

    con version pinning:

    • when you DO want to make an update, you have to spent 2 minutes to go to docker hub to find out which version you want.