Want to find out if the text you're reading online was written by an real human or spat out by a large language model (LLM) trying to sound like one?

Want to find out if the text you’re reading online was written by an real human or spat out by a large language model (LLM) trying to sound like one? Mozilla’s Fakespot Deepfake Detector Firefox add-on may can help give you an indication. Similar to online AI detector tools, the add-on can analyse text (of 32 words or more) to identify patterns, traits, and tells common in AI generated or manipulated text. It uses Mozilla’s proprietary ApolloDFT engine and a set of open-source detection models.

  • ArkyonVeil@lemmy.dbzer0.com
    12·
    8 hours ago

    I like the idea of a general AI detection approach. Problem is that it’s very easy to get false positives depending on the writer’s personal style. Accusing garbage of being garbage does nothing, but falsely accusing an individual of using AI when none is used will just lead to harmful witch hunt behavior.

    Also, putting your trust in a flawed tool like this, which might miss actual AI written speech (or human bullshit speech) will just give a false sense of security.

    Be careful out there fine folks. The unscrupulous used to lie using just humans, now they also use robots to do it.

  • kbal@fedia.io
    1·
    5 hours ago

    I’d rather have a tool to detect intelligence in general, weeding out the stupidity regardless of who it comes from.

  • GenderNeutralBro@lemmy.sdf.orgEnglish
    2·
    6 hours ago

    I don’t see any mention of whether this uses local models or cloud models. I’m not interested in sending anything I care about it into the cloud.

    • thingsiplay@beehaw.org
      2·
      5 hours ago

      I think that I can decide better than an Ai tool if a text is written by an Ai. Not a fan of this. I find it weird to use an Ai tool, to detect Ai text. Also I wonder if those Ai fake tools can utilize Fakespot detection to improve the actual fakeness. They can train their Ai until Fakespot does not detect anymore.

      They talk about their proprietary model, plus several other tools. Highlight any text online and request an analysis. Nowhere in the article, the download page and their Fakespot website is once mentioned if this is local and offline. So, I’m 99% sure the data is sent to their server for analyzing the highlighted text.

  • Ebby@lemmy.ssba.com
    33·
    16 hours ago

    Sounds like a great idea, but a significant portion of 1-star reviews point out massive data collection issues.

    More information necessary before I install this on my system.

  • perishthethought@lemm.eeOPEnglish
    18·
    17 hours ago

    Oh man - right after I posted this, I got the DM others have reported from the Polish tech chick from Toronto, with links to her cool videos. I passed her message to this extension but it says it looks like Human text. Oh well.

      • TehPers@beehaw.orgEnglish
        3·
        9 hours ago

        There’s an entire category of machine learning dedicated to having two AIs “fight” against each other. One generates something while the other classifies it as either AI generated or genuine.

        Anyway, this is a complete tangent. Just thought it was interesting. AI detector tools for LLMs aren’t usually very accurate, unfortunately.

      • NuXCOM_90Percent@lemmy.zip
        14·
        16 hours ago

        Like others said: That is probably just “normal” spam

        But also? This is the same logic as “AI can’t draw hands” and all the other gotchas. We are in an arms race and have been for decades. Captchas have always been about “bots” and have increasingly been defeated by various forms of computer vision which are the basis for how a lot of “AI” works. And we are in the same place now.

        • burgersc12@mander.xyzEnglish
          7·
          16 hours ago

          I just think its almost impossible to detect AI, same way you are talking about how it is almost impossible to build a captcha that is both useful and bot proof.