• ViatorOmnium@piefed.social
    link
    fedilink
    English
    arrow-up
    46
    arrow-down
    2
    ·
    6 days ago

    The answer is AI slop. Kagi translate uses a slopinator in the background, you can pass anything in the to query parameter.

    • thedeadwalking4242@lemmy.world
      link
      fedilink
      English
      arrow-up
      77
      arrow-down
      1
      ·
      6 days ago

      Translation is actually one of the real valid uses for LLMs. It’s actually what they where originally designed for. They are DAMN good at it to.

      It picks up on local dialects and cultural euphemism very well and give you very natural output.

      • Digitalprimate@lemmy.world
        link
        fedilink
        arrow-up
        8
        arrow-down
        1
        ·
        6 days ago

        I had to very, very grudgingly but rightfully apologize to one of my traders for using an LLM to translate some Vietnamese financial accounts to English. I read to fast and missed the part where he said he used it for translation. I still worry the actual financial ratios are off, so I’ll have my other guy do that, but yeah, ok I hate these things but they are good at translation.

      • kspatlas@piefed.blahaj.zone
        link
        fedilink
        English
        arrow-up
        1
        ·
        5 days ago

        I find they’re also decent at helping you find a starting point for translating unfamiliar sentences in languages you know the basics of

    • AItoothbrush@lemmy.zip
      link
      fedilink
      English
      arrow-up
      48
      ·
      6 days ago

      I mean at least its a valid use of a LARGE LANGUAGE MODEL, focus on it being a language model. If they actually optimize it for translating it could be a good thing.

      • idiomaddict@lemmy.world
        link
        fedilink
        English
        arrow-up
        19
        ·
        6 days ago

        I’m a language teacher and tbh, deepl is difficult to detect at a high level. It’s easier if you know their style and if you can recognize phrasing from their native language, but if you’re giving a native Dutch speaker you don’t know an essay to write at C1 academic German, good fucking luck telling if it’s an ai translation or not.

        • Digitalprimate@lemmy.world
          link
          fedilink
          arrow-up
          6
          ·
          6 days ago

          I’m a C1 Dutch speaker (barely, technically) and before I understood how evil LLMs are, I played around with some of them to have them tweak things I wrote in Dutch into different styles. It was pretty eye opening.

        • arrow74@lemmy.zip
          link
          fedilink
          English
          arrow-up
          2
          ·
          6 days ago

          DeepL does a fairly good job translating academic articles. It has been extremely helpful in that regard. Of course I’ve seen some issues, but they’ve been minor and if you know the subject easy to figure out

    • chuso@fedia.io
      link
      fedilink
      arrow-up
      30
      ·
      6 days ago

      Indeed, you can choose anything you want as the destination language.

      A screenshot that shows that the same "I got a new job" that was used in Kagi's example of translation to LinkedIn speak is now used to translate it to Slavoj Žižek speak, offering a long and rambling speech in a Žižek way and offering options to adjust the theoretical focus or the level of Žižekian intensity.