Comment on Please allow ads! JK, you need to subscribe, too! LOL

• Vespair@lemm.ee ⁨5⁩ ⁨weeks⁩ ago

  It doesn’t though. Because AI was trained on human data, it contains and can replicate human errors. Its extremely rare yes, especially compared to real human output, but I have personally seen ai make misspellings and other human-like errors in its output.

  source

  • Lumidaub@feddit.org ⁨5⁩ ⁨weeks⁩ ago

    But actual typos? I might see it using the wrong word because people keep using it (can’t think of a good example right now but like their/there) but “Minneosta” isn’t a word at all. A simple spell check would catch that.

    source

    • Vespair@lemm.ee ⁨5⁩ ⁨weeks⁩ ago

      Yes, actual typos. LLMs don’t perform spell checks. Yes, they can “spell check” you input, but that isn’t what is actually happening. It’s all predictive text, and if they’ve learned to predict that the appropriate word in this context in 98% of cases is Minnesota, but since their dataset includes real human errors, it’s not unrealistic that they the LLM could also conclude that the appropriate word in 2% of cases is “Minneosta” instead, which means given enough output variables the misspelling will appear.

      Again, I have personally had LLMs generate output with typos, not just factual errors.

      source

      • Lumidaub@feddit.org ⁨5⁩ ⁨weeks⁩ ago

        Goddammit. Well thanks a lot for ruining this silver lining in all this. 😠

        source

        • -> View More Comments
• Lumidaub@feddit.org ⁨5⁩ ⁨weeks⁩ ago

  Excellent point. Huh.

  source