• KevonLooney@lemm.ee
    111·
    6 months ago

    They said AI would take you places. They never said they were places you wanted to go.

  • xantoxis@lemmy.world
    925·
    6 months ago

    When we say LLMs don’t know or understand anything, this is what we mean. This is a perfect example of an “AI” just not having any idea what it’s doing.

    • I’ll start with a bit of praise: It does do a fairly good job of decomposing the elements of Python and the actuary profession into bits that would be representative of those realms.

    But:

    • In the text version of the response, there are already far too many elements for a good tattoo, demonstrating it doesn’t understand tattoo design or even just design
    • In the drawn version, the design uses big blocks of color with no detail, which (even if they looked good on a white background; and they don’t;) would look like shit inked on someone’s skin. So again, no understand of tattoo art.
    • It produces a “simplified version” of the python logo. I assume those elements are the blue and yellow hexagons, which are at least the correct colors. But it doesn’t understand that, for this to be PART OF THE SAME DESIGN, they must be visually connected, not just near each other. It also doesn’t understand that the design is more like a plus; nor that the design is composed of two snakes; nor that the Python logo is ALREADY VERY SIMPLE, nor that the logo, lacking snakes, loses any meaning in its role of representing Python.
    • It says there’s a briefcase and glasses in there. Maybe the brown rectangle? Or is the gray rectangle meant to be a briefcase lying on its side so the handle is visible? No understanding here of how humans process visual information, or what makes a visual representation recognizable to a human brain.
    • Math stuff can be very visually interesting. Lots of mathematical constructs have compelling visuals that go with them. A competent designer could even tie them into the Python stuff in a unified way; like, imagine a bar graph where the bars were snakes, twining around each other in a double helix. You got math, you got Python, you got data analysis. None of this ties together, or is even made to look good on its own. No understanding of what makes something interesting.
    • Everything is just randomly scattered. Once again, no understanding of what design is.

    AIs do not understand anything. They just regurgitate in ways that the algorithm chooses. There’s no attempt to make the algorithm right, or smart, or relevant, or anything except an algorithm that’s just mashing up strings and vectors.

    • driving_crooner@lemmy.eco.brOP
      291·
      6 months ago

      I was hoping for a sand clock and the python snake, but now I’m not sure if the sand clock is an international actuarial thing, or if is just a brazillian one. But for mathematical notation related to actuarial sciences the annuanity [1] is the main one, so 2/10.

      • xantoxis@lemmy.world
        29·
        6 months ago

        See, that’s a cool symbol. Make the right angle part of that symbol into a snake, you’re done. 1000% better than the AI’s mess.

      • Jaq@lemmy.blahaj.zone
        2·
        6 months ago

        Might be a Brazilian thing with the sand timer, but the annuity ä is imprinted on my brain, and it’s been years since those exams. The tattoo needs some element of “what was this value last year?”

    • tjsauce@lemmy.world
      101·
      6 months ago

      It’s kind of adorable, like a child designing an album cover using concepts they recognize but don’t understand

  • set_secret@lemmy.world
    611·
    6 months ago

    I put in the same prompt and got this. Gpt never breaks down a request with here are some elements I can include etc; do you think this is in the custom instructions or something?

  • rbn@sopuli.xyz
    121·
    6 months ago

    In case you get it tattoed, also put the entire conversation next to it. Would be funny at least for a few years. Then, probably no one will remember what a ChatGPT is.

    • ___@l.djw.liEnglish
      21·
      6 months ago

      Lord, we can only hope.

      Getting both generic and specific models shoved down my throat by $multiNationalCorp on a daily basis is exhausting.

      If I write an email that annoys someone and costs the company money, I’m thrilled to bits to own it and admit I fucked it up. But I dont have the time or energy to make sure that AI isn’t turning a reasonable email into something rage-inducing just by missing an obvious nuance. I’m certainly not hanging the quality of my code on the strength of an AI “helper.”

  • pfm@scribe.disroot.org
    102·
    6 months ago

    My main concern with people making fun of such cases is about deficiencies of “AI” being harder to find/detect but obviously present.

    Whenever someone publishes a proof of a system’s limitations, the company behind it gets a test case to use to improve it. The next time we - the reasonable people arguing that cybernetic hallucinations aren’t AI yet and are dangerous - try using such point, we would only get a reply of “oh yeah, but they’ve fixed it”. Even people in IT often don’t understand what they’re dealing with, so the non-IT people may have even more difficulties…

    Myself - I just boycott this rubbish. I’ve never tried any LLM and don’t plan to, unless it’s used to work with language, not knowledge.