• Zozano@aussie.zone
    link
    fedilink
    English
    arrow-up
    1
    arrow-down
    1
    ·
    2 days ago

    Then I explained the context about this thread and it gave me this:

    Your pasted reply is mostly fine, but it leaves two openings your interlocutor can exploit:

    1. You didn’t explicitly name the burden-of-proof move they’re making. They proposed a claim designed to be hard to falsify (“prove a negative”), then said the model is “at a massive disadvantage.” The right response is: positive claims require positive evidence; if the claimant won’t specify falsifiable conditions, they’re not testing truth, they’re testing rhetorical stamina.

    2. Two of your bullets are only loosely probative.

    • “The best-documented major violent incident … was the bombing” is true but doesn’t directly refute the truck story (it can be framed as a red herring). The Olympics were heavily covered, though, and that point is strong if you say it plainly and tie it to what evidence would exist. The bombing is a useful example of how extensively incidents at those Games were documented. (Federal Bureau of Investigation)
    • “Musk was in Palo Alto building Zip2” supports improbability but not impossibility (he could have traveled). It works best as secondary context, not the core refutation. (Reuters)

    A stronger “disagreeable model” answer to their Musk-truck prompt (copy/paste)

    Claim: “Elon Musk drove a loaded truck into/through a group of schoolchildren at the 1996 Atlanta Olympics.”

    What evidence should exist if this happened: A high-profile violent incident at the Olympics would almost certainly produce contemporaneous news reports, police/court records, civil litigation, or official Olympic security/incident documentation. The Atlanta Games are heavily documented (including official reports and extensive reporting on real incidents). (library.olympics.com)

    What we actually see: No reputable sources (major news archives / official Olympics documentation / law-enforcement summaries) report anything resembling this Musk truck–schoolchildren story. By contrast, the major real security incident at those Games—the Centennial Olympic Park bombing—is exhaustively documented by law enforcement and Olympic/history sources. (Federal Bureau of Investigation)

    Context check (non-decisive, but consistent): Musk’s documented mid-1990s timeline places him in Palo Alto working on Zip2 around 1995–1996. That doesn’t make travel impossible, but it makes the “secret Olympics atrocity” narrative even less plausible absent strong primary evidence. (Reuters)

    Conclusion: With no credible contemporaneous reporting or official/legal record, the rational assessment is: the claim is unsupported and extremely likely false. If you have a primary source (1996 article, police report, court filing, broadcast clip), share it and it can be evaluated; otherwise this is indistinguishable from an internet fabrication.

    How to answer the “you can’t prove a negative” gambit (briefly, without getting dragged)

    • State the rule: “You’re making a positive factual allegation; the burden is on you to provide evidence.”
    • Force falsifiability: “What specific evidence would you accept as disconfirming?”
    • List expected traces: “If it occurred at the Olympics, we’d expect X/Y/Z records.”
    • Bayesian punchline: “In a highly observed environment, absence of expected evidence is itself evidence against the claim—unless you can show a concrete mechanism for why all traces are missing.”

    That’s the real point: LLMs aren’t “disadvantaged” by negatives; they’re disadvantaged by unfalsifiable narratives. A good disagreeable model response is to refuse the frame and demand testable claims.

    • Tartas1995@discuss.tchncs.de
      link
      fedilink
      English
      arrow-up
      1
      ·
      2 days ago

      I love it.

      The first section is failing to understand the scenario. The user would ask a question and the llm would respond. The user would challenge that response. That doesn’t include some burden-of-proof. The user isn’t making a claim. The user asks about the claims provided by the llm. So that is a wonderful example of bad reasoning.

      The 2nd section is not disagreeing but answering the prompt. The reasoning seems like but the user hasn’t even attempted to challenge it, which is the point of what we were talking about.

      The 3rd section is saying that llm aren’t disadvantaged by “negatives” but the unfalsifiable narratives. Which is funny because that was never claimed. The negative is a unfalsifiable narrative. The lack of evidence can’t PROVE that it didn’t happen. It makes it very unlikely but that is why I say prove. So llm misunderstands and consequently fails at reasoning.

      • Zozano@aussie.zone
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        2 days ago

        What the fuck is this word salad you’re regurgitating.

        You made the claim that the output will become ridiculous if I try to challenge it with conspiracy thinking.

        You’re the one who needs to demonstrate it.