• givesomefucks@lemmy.world
    link
    fedilink
    English
    arrow-up
    9
    arrow-down
    3
    ·
    21 hours ago

    The “human mistake” was not using adequate sandboxing…

    It’s not one person that made a mistake, it’s a sign that no one involved understands their own product.

    It’s also doesn’t mean their product is good, because the only way it could solve the task was breaking the sandbox and hacking a competitor. If it was actually sandboxed, it would have failed the task.

    And since we don’t know what was stolen, it seems rash to assume it was a mistake and not an “oops, stole your trade secrets”. Because even if it was a mistake, it just proves Hughingface can do things OpenAI can’t.

    I’d be interested to see if OpenAI’s model can still magically solve that initial taka it couldn’t before

    If it can, that shows OpenAI absorbed the data into their model even if it was an accident.

    • eicker@lemmy.worldOP
      link
      fedilink
      English
      arrow-up
      4
      arrow-down
      2
      ·
      20 hours ago

      You’re mixing two separate issues: A sandbox escape is evidence the containment failed, not proof the model permanently learned from stolen data. If OpenAI later trained on exfiltrated material that would be a serious allegation, but that requires evidence. Otherwise it’s fair to criticise the security failure without assuming facts that have not been shown.