• Stefan_S_from_H@discuss.tchncs.deOP
    link
    fedilink
    arrow-up
    36
    arrow-down
    1
    ·
    5 days ago

    A few months ago, someone asked Google for countries that start with an H. In German.

    It listed Hungary, which is called “Ungarn” in German. OK, an understandable mistake. But then it gave the additional information that Hungary sometimes gets called Holland.

    Two things upset me:

    1. People believe the answers.
    2. Nobody is talking about all the stupid mistakes AI is making. It should have all stopped after LLM AIs thought blueberry has 3 Bs in it.
    • FishFace@piefed.social
      link
      fedilink
      English
      arrow-up
      13
      ·
      5 days ago

      Nobody is talking about all the stupid mistakes AI is making. It should have all stopped after LLM AIs thought blueberry has 3 Bs in it.

      I never hear about anything else

    • TheRealKuni@piefed.social
      link
      fedilink
      English
      arrow-up
      10
      ·
      5 days ago

      LLMs are really bad with letters, and in my limited understanding that’s because they don’t see words as strings of letters, they see them as tokens. It’s all numbers by the time the LLM is processing it.

      • exasperation@lemmy.dbzer0.com
        link
        fedilink
        arrow-up
        2
        ·
        4 days ago

        We think in terms of tokens, too, but we have the ability to look under the hood at some of how our knowledge is constructed.

        For the typical literate English speaker, we seamlessly pronounce certain letter combinations as different from the component parts (like ch, sh, ph, or looking ahead to see if the syllable ends in an E to decide how to pronounce the vowel in the middle). Then, entire words or phrases have a single meaning that doesn’t get broken apart. Similarly, people who are fluent in multiple languages, including languages that use the same script (e.g., latin letters), can look at the whole string of text to quickly figure out which language they’re reading, and consult that part of their knowledge base.

        And usually our brains process things completely separately from how we read or write text. Even the question of asking how many r’s are in “raspberry” requires us to go and count, because it isn’t inherent in the knowledge we have at the tip of tongue. Someone can memorize a speech but not know how many times the word “the” appears in it, even if their knowledge contains all the information necessary to answer the question.

        Even if we are actively thinking in the context of how words are constructed, like doing crosswords, these things tend to be more fun when mixed with other modes of thinking: Wordle’s mix of both logic and spelling, a classic crossword’s clever style of hints, etc.

        Manipulation of letters is simply one mode of thinking. We’re really good at seamlessly switching between modes.

  • criss_cross@lemmy.world
    link
    fedilink
    arrow-up
    13
    ·
    edit-2
    5 days ago

    I can’t replicate it but at one point I asked why there were only 5 playable characters in a specific video game. Which the AI snippet tried to tell me that there were actually 120 playable characters.

    Yeah.

    EDIT. I’ll put this in a spoiler as it sort of is but the game in question was

    Tap for spoiler

    Clair Obscur: Expedition 33

    You could argue there’s 6 total. But there’s only 5 at a given time , which given that you have a party of 3 and can have the other 2 come in as support I thought it was odd you didn’t have 6 at a given time. So I went to DuckDuckGo for speculation and it proudly said there were 120.

  • 👍Maximum Derek👍@discuss.tchncs.de
    link
    fedilink
    English
    arrow-up
    11
    ·
    5 days ago

    I was asked to evaluate the last version of Gemini for work. I set up the agent, gave it a robust gemini.md file, and asked it about a bug I was seeing. It told me the bug was because I had spelled “Arrange” with 3 Rs. Except:

    1. I hadn’t
    2. Doing so would not have caused the bug I was seeing
    3. 3 consecutive letter R’s didn’t exist anywhere in the code base

    Gemini 3 really had no where to go but up.

    • criss_cross@lemmy.world
      link
      fedilink
      arrow-up
      11
      ·
      5 days ago

      I once got an Agentic AI stuck in a loop where it kept proposing unrelated code to fix an issue. It would add it, break, then detect it was unrelated, delete it, retry and see the issue was still there. So it would add the code back and redo the loop.

      Clearly we are on the path to replacing humans.

  • frida@lemmy.world
    link
    fedilink
    arrow-up
    5
    ·
    5 days ago

    this shit is so annoying when i click on all but accidentally click ai mode… fuck off