Language models generate text based on statistical probabilities. This led to serious false accusations against a veteran court reporter by Microsoft's Copilot.

German journalist Martin Bernklau typed his name and location into Microsoft’s Copilot to see how his culture blog articles would be picked up by the chatbot, according to German public broadcaster SWR.

The answers shocked Bernklau. Copilot falsely claimed Bernklau had been charged with and convicted of child abuse and exploiting dependents. It also claimed that he had been involved in a dramatic escape from a psychiatric hospital and had exploited grieving women as an unethical mortician.

…

Bernklau believes the false claims may stem from his decades of court reporting in Tübingen on abuse, violence, and fraud cases. The AI seems to have combined this online information and mistakenly cast the journalist as a perpetrator.

Microsoft attempted to remove the false entries but only succeeded temporarily. They reappeared after a few days, SWR reports. The company’s terms of service disclaim liability for generated responses.

…

  • @Grimy
    -21•20 days ago

    So just to be clear, if you can sue companies for this, there is no open source scene and we end up with only Microsoft and Google in the game since they will be the only one able to eat the fines.

    There’s no easy way to solve this problem, especially with the tech being so recent and the scope so big. In any case, it’s user error. Llms aren’t expected to be right at all times, especially when it’s a coding model about obscure journalists. They are tools to help the user, and every step requires verification from the user.

    They aren’t a replacement for truth, they can’t stand in for wikipedia and news articles, they aren’t meant to be cited in papers, etc.

    • robsutoEnglish
      24•20 days ago

      What do you mean by ‘there’s no open source scene’?

      I don’t understand what open source has to do with this.

      • Vaquedoso
        8•20 days ago

        He’s saying that the only corporations with the fighting power to take on legal battles will end up being the big ones. So we may end up in a situation where AI will only be in the hands of the mega wealthy, instead of in the hands of regular people.

        • 2xsaiko
          12•19 days ago

          “Open source” models usually run on your local hardware instead of accessing it through some corporation’s website. Who are you gonna sue when your own computer spits out garbage about you, yourself?

          • @Grimy
            1•19 days ago

            I imagine the ones creating and distributing the model. Even if you only got sued when you hosted a model and not when you shared it, it still doesn’t make for a good ecosystem. Regular people should have the choice to use models even if it spits out garbage for certain tasks, it might suit their needs for their own task perfectly.

            There’s no reason to gatekeep llms and lock them behind hardware requirements, it’s up to people to understand their limitations and what they are for.

            • 2xsaiko
              2•19 days ago

              I mean I’m not a lawyer but this is what I think is relevant here:

              1. This is a public service provided by Microsoft (or whoever really)
              2. It prints libel
              3. They’re responsible for the libel it prints as it’s not user generated content (I think there’s a law about that that excludes specifically this so running social media sites is viable)

              I really don’t think it matters whether what’s behind it is an LLM or an underpaid Indian writing the text in real time or if it’s just static pages the site owner wrote. They’re still responsible for it.

              If you run it locally, none of it is public (until you publish what it generated, in which case you’re responsible for the content).

              • @Grimy
                0•19 days ago

                It would be relevant if Microsoft or any of the LLM companies presented their models outputs as truths. It’s been repeated multiple times that the outputs should be reviewed and verified. This is some serious “Reddit lied to me” vibes. Copilot literally says it uses AI and to check for mistake on the chat page.

                On top of that, these could be viewed as bugs. Can you actually imagine suing over bugs about a novel type of software that is realistically two years old? Though tbh it will be a long time before we reach tech that cannot make a mistake. The general public expectations are a bit ridiculous imo.

    • @[email protected]
      6•19 days ago

      There’s no easy way to solve this problem

      How about not replacing search engines with this evidently non-functional scam, for instance…?

      It’s user error

      No. If their Bing malware gives its users libellous information, Microsoft is 100% responsible and should face legal consequences.

      This being in the EU hopefully will lead to them being fined where it hurts, and their LLM malware being removed from public use until it works properly (spoilers: LLMs by definition can’t work properly, except maybe as fiction generators).

      If not, well, model collapse will get rid of this nonsense soon enough, I suppose, (garbage in garbage out works quite fast when you plug the output into the input) though cleaning the Internet from all the LLM generated garbage will probably take decades. Hopefully the idiots responsible will be fined to pay for the costs.

      • @Eranziel
        3•19 days ago

        Agreed. The solution to this is to stop using LLMs to present info authoritatively, especially when facing directly at the general public. The average person has no idea how an LLM works, and therefore no idea why they shouldn’t trust it.