• RandomVideos
    link
    fedilink
    arrow-up
    26
    ·
    13 hours ago

    Why hope they sanitize their inputs?

    Why are they trusting an AI that cant even do math to give notes to tests?

    • CandleTiger
      link
      fedilink
      arrow-up
      4
      ·
      3 hours ago

      The problem with LLM AIs Ous that you can’t sanitize the inputs safely. There is no difference between the program (initial prompt from the developer) and the data (your form input)

      • RandomVideos
        link
        fedilink
        arrow-up
        2
        ·
        3 hours ago

        You can make it more resistant to overwriting instructions at least

        • CandleTiger
          link
          fedilink
          arrow-up
          2
          ·
          2 hours ago

          You can try, but you can’t make it correct. My ideal is to write code once that is bug-free. That’s very difficult, but not fundamentally impossible. Especially in small well-scrutinized areas that are critical for security it is possible with enough care and effort to write code with no security bugs. With LLM AI tools that’s not even theoretically possible, let alone practical. You will just need to be forever updating your prompt to mitigate the free latest most fashionable prompt injections.

    • bleistift2@sopuli.xyz
      link
      fedilink
      English
      arrow-up
      10
      ·
      11 hours ago

      Because it’s the most efficient. With students handing in AI theses, it’s only sensible to have teachers use AI to grade them. No we only need to have teachers use AI to create exam questions and education becomes a fully automated process. Then everyone can go home early.