• Deebster@infosec.pub
        link
        fedilink
        English
        arrow-up
        3
        ·
        4 days ago

        Really? I would be shocked if an LLM’s been anywhere near this, it’s not in AI style at all.

        • thingsiplay@beehaw.org
          link
          fedilink
          arrow-up
          2
          ·
          4 days ago

          @syklemil@discuss.tchncs.de Just for curiosity I checked the text with an Ai detector. Mind you, this is not proof for or against, as I don’t take these tools too serious too. They are Ai tools themselves.

          https://www.scribbr.com/ai-detector/ says 0% likely is written by Ai.

          https://stealthwriter.ai/ says 93% is written by Ai. This tool also shows the passages it thinks it is.

          Remember, this is not a proof. But it looks suspicious to me. Test the services with text you know for a fact it is written by human (like yourself). And test is with some text you know for a fact it is output by an Ai (maybe directly ask an Ai for some text). In example the second tool gives me 48% written by Ai, for some text from my own blog post.

          • eah@programming.dev
            link
            fedilink
            arrow-up
            4
            ·
            3 days ago

            At some point, we’re going to have to have verified real human identity crap because the present situation of having to question everything I come across on the internet and essentially CAPTCHA myself to everyone every time I post is giving me a level of stress that makes me want to log off forever and I can’t be the only one.

          • Deebster@infosec.pub
            link
            fedilink
            English
            arrow-up
            3
            ·
            edit-2
            3 days ago

            That StealthWriter thing doesn’t feel very reliable to me, e.g. it pick up the first paragraph:

            After 4 years with Rust, I love the language – but I’m starting to think the ecosystem has an abstraction addiction. Or: why every Rust crate feels like a research paper on abstraction.

            I think AI would write “four” instead of 4, would use an em dash, and the construction of the second sentence (a Doctor Strangelove reference?) doesn’t feel like most LLM’s style.


            Anyway, there is this massive red flag at the end which suggests I was completely wrong:

            P.S grammarly forked me over vro 🥀

            I’m not 100% sure what that means, but Grammarly is an AI writer thingy (or maybe it’s only their premium level?). I know that AI tools do pick up on Grammarly-filtered stuff as these somewhat false positives are a problem in educational circles.

            • thingsiplay@beehaw.org
              link
              fedilink
              arrow-up
              2
              ·
              3 days ago

              Yeah, I noticed too it does not work well. After some other comments, I tested it on more stuff where I know for a fact it was written by human (myself). And it even gave me over 40% ai probability. It’s entirely possible it does that by design, because they have a functionality to obfuscate text for other Ai to detect.

              I need to stop talking and posting about these tools, they are as bad as Ai can get.

      • Flipper@feddit.org
        link
        fedilink
        arrow-up
        2
        ·
        3 days ago

        Somewhere else this was posted the author stated they wrote it themselves, but took grammar corrections from grammarly. That probably created the Ai vibes.

        They already got a lot of flak for it there.