Close Menu
    Technik

    Der CEO von Anthropic sagt, dass obligatorische Sicherheitstests für KI-Modelle erforderlich sind


    Der CEO von Anthropic sagt, dass obligatorische Sicherheitstests für KI-Modelle erforderlich sind

    https://www.bloomberg.com/news/articles/2024-11-20/anthropic-ceo-says-mandatory-safety-tests-needed-for-ai-models?embedded-checkout=true

    View 9 Comments

    9 Comments

    1. Jeb19780101 on

      i think it is the CEOs that need the safety tests more than the AIs.

    3. Bob_Spud on

      A time waster, AI could have appropriate replies built into it. They have to keep constantly changes the tests to fool AI tech.

      AIssS for criminal and other nefarious activity is more of a worry.

    4. mulligan on

      Ask him if they are needed when used by the military 

      The number one way these things will kill people is exactly the place most of these guys seek an exemption from safety 

    6. EmbarrassedHelp on

      Depending on the intended use case of a model, safety testing should probably be mandatory. You don’t want a mental health assistant telling users to do something stupid.

      However, for creative tools they need to drop the bullshit puritan ideas that sex related stuff and certain publicly available knowledge is “unsafe”.

    7. 1king-of-diamonds1 on

      Just in case anyone is unaware, while these models definitely need guardrails AI techbros aren’t pushing this out of genuine concern.

      The purpose of these sorts of headlines is just to make the AI space impossible to enter, cementing their lead and making disruption very unlikely.

      Techniques such as retroactively paying for IP or rigorous testing are being used to raise the barrier to entry in the space. While these things aren’t in themselves problematic (ideally they should have been in place all along) it’s crucial that these sorts of restrictions are also bundled with antitrust legislation to try and prevent this space from being perpetually locked in biased monopolies.

      As others have said – it’s the companies and CEOs that require scrutiny as much or more then the models themselves

    8. PMzyox on

      Guys, I have a solution to this mess:

      Mutually Assured Destruction + AI

    Leave A Reply