Comment on AI agents wrong ~70% of time: Carnegie Mellon study

• Knock_Knock_Lemmy_In@lemmy.world ⁨1⁩ ⁨week⁩ ago

  Run something with a 70% failure rate 10x and you get to a cumulative 98% pass rate. LLMs don’t get tired and they can be run in parallel.

  source

  • MangoCats@feddit.it ⁨1⁩ ⁨week⁩ ago

    I have actually been doing this lately: iteratively prompting AI to write software and fix its errors until something useful comes out. It’s a lot like machine translation. I speak fluent C++, but I don’t speak Rust, but I can hammer away on the AI (with English language prompts) until it produces passable Rust for something I could write for myself in C++ in half the time and effort.

    I also don’t speak Finnish, but Google Translate can take what I say in English and put it into at least somewhat comprehensible Finnish without egregious translation errors most of the time.

    Is this useful? When C++ is getting banned for “security concerns” and Rust is the required language, it’s at least a little helpful.

    source

    • jsomae@lemmy.ml ⁨1⁩ ⁨week⁩ ago

      I’m impressed you can make strides with Rust with AI. I am in a similar boat, except I’ve found LLMs are terrible with Rust.

      source

      • MangoCats@feddit.it ⁨1⁩ ⁨week⁩ ago

        I was 0/6 on various trials of AI for Rust over the past 6 months, then I caught a success. Turns out, I was asking it to use a difficult library - I can’t make the thing I want work in that library either (library docs say it’s possible, but…) when I posed a more open ended request without specifying the library to use, it succeeded - after a fashion. It will give you code with cargo build errors, I copy-paste the error back to it like “address: <pasted error message>” and a bit more than half of the time it is able to respond with a working fix.

        source

        • -> View More Comments
  • jsomae@lemmy.ml ⁨1⁩ ⁨week⁩ ago

    The problem is they are not i.i.d., so this doesn’t really work. It works a bit, which is in my opinion why chain-of-thought is effective (it gives the LLM a chance to posit a couple answers first). However, we’re already looking at “agents,” so they’re probably already doing chain-of-thought.

    source

    • Knock_Knock_Lemmy_In@lemmy.world ⁨1⁩ ⁨week⁩ ago

      Very fair comment. In my experience even increasing the temperature you get stuck in local minimums

      I was just trying to illustrate how 70% failure rates can still be useful.

      source
  • davidagain@lemmy.world ⁨1⁩ ⁨week⁩ ago

    What’s 0.7^10?

    source

    • Knock_Knock_Lemmy_In@lemmy.world ⁨1⁩ ⁨week⁩ ago

      About 0.02

      source

      • davidagain@lemmy.world ⁨1⁩ ⁨week⁩ ago

        So the chances of it being right ten times in a row are 2%.

        source

        • -> View More Comments
• outhouseperilous@lemmy.dbzer0.com ⁨1⁩ ⁨week⁩ ago

  Less broadly useful than 20 tons of mixed texture human shit.

  source

  • jsomae@lemmy.ml ⁨1⁩ ⁨week⁩ ago

    Are you just trolling or do you seriously not understand how something which can do a task correctly with 30% reliability can be made useful if the result can be automatically verified.

    source

    • outhouseperilous@lemmy.dbzer0.com ⁨1⁩ ⁨week⁩ ago

      Its not a magical 30%, factors apply. It’s not even a mind that thinks and just isnt very good.

      This isnt like a magical dice that gives you truth on a 5 or a 6, and lies on 1,2,3,7, and for.

      This is a (very complicated very large) language or other data graph that programmatically identifies an average. 30% of the time. Which means the more possible that is, the easier it is to either use a simpler cheaper tool that will give you a better more reliable answer much faster.

      And 20 tons of human shit has uses! If you know its providence, there’s all sorts of population level public health surveillance you can do to get ahead of disease trends! Its also got some good agricultural stuff in it-phosphorous and stuff, if you can extract it.

      source

      • jsomae@lemmy.ml ⁨1⁩ ⁨week⁩ ago

        I think everyone in the universe is aware of how LLMs work by now, you don’t need to explain it to someone just because they think LLMs are more useful than you do.

        IDK what you mean by glazing but if by “glaze” you mean “understanding the potential threat of AI to society instead of hiding under a rock and pretending it’s as useless as a plastic radio,” then no, I won’t stop.

        source

        • -> View More Comments