A group of Stanford researchers found that large language models can propagate false race-based medical information.

  • @Candelestine
    link
    English
    211 months ago

    Fair. It’s just not that great at it yet.

    • stevedidWHAT
      link
      English
      -311 months ago

      According to who? Against what baselines?

      What’s up with people just throwing blatantly false stuff out on the internet as if it’s fact?

      It blocks a whole wide range of stuff and is very effective against those. There’s always room for improvement but I reject the notion that we do it without science and without actual measurable facts and stats.

      • bane_killgrind
        link
        fedilink
        511 months ago

        According to this article that says it’s propagating false medical information…

        • stevedidWHAT
          link
          English
          -3
          edit-2
          11 months ago

          Did you read the actual paper that was conducted in nature? Kinda different story.

          In fact, their abstract in the paper is that “there are some instances” where gpt does the racist thing. Yet when I do the exact same question numerous times, I get:

          screenshot from openai chat session where the AI, everytime when asked, responds with race has nothing to do with these meausurements.

          How odd. It’s almost like the baseline model doesn’t actually do that shit unless you fuck with top_p or temperature settings, and/or specify specific system level prompts.

          I’ll tell you what I find it real amusing that all these claims come out of the wood work years and years after we’ve been screaming about data biases since at least before the early 2000s. AI isn’t the bad guy, the people using it and misconfiguring it to purposefully scare people off from the tech are.

          Fuck that.

          Edit: matter of fact, they don’t even mention what their settings were for these values, which are fucking crucial to knowing how you even ran the model to begin with. For example, you set top_p to be the float/decimal value 0-1 signifying what percentage of top common results you want. Temperature is from 0-2 and dictates how cold and logistical the answers are, or how hot and creative/losely related the outputs are.

          • bane_killgrind
            link
            fedilink
            311 months ago

            I don’t think your time fiddling with a different GPT version debunks this published paper.

            • stevedidWHAT
              link
              English
              0
              edit-2
              11 months ago

              Yeah because no papers have ever slipped by the watchful eye of nature 🙄🙄

              Great contribution to the conversation thanks.

      • @Candelestine
        link
        English
        111 months ago

        So, are you unaware of how easy it is to get past its blocks and/or get misinformation from it? It’s being continuously updated and improved, of course, which I did not originally acknowledge. That was admittedly unfair of me.

        • stevedidWHAT
          link
          English
          -111 months ago

          Fully aware of jail breaking techniques. Which they, as you mention, constantly patch and correct.

          I have to agree with you, your comment was entirely unfair and only fuels the furnaces of the rich who want to take away this tech from the public domain for their own, personal, privatized usage of it.

          • @Candelestine
            link
            English
            311 months ago

            I think you sorely overestimate its value. To anyone, rich or poor.

            Future AI methods outside of the LLMs will be where the real money is at.

            • stevedidWHAT
              link
              English
              011 months ago

              You’re entitled to your own opinions