• @[email protected]
    link
    fedilink
    English
    24 months ago

    If anyone has a good document that fucks up AI models… Now would be the time to share it.

      • @[email protected]
        link
        fedilink
        English
        24 months ago

        Not that I’m keen on AI, but doesn’t happen anymore with o3-mini.

        Good thing is that it still lacks on complex tasks.

      • Teknikal
        link
        fedilink
        English
        94 months ago

        I tried that recently on the Gemini 2.0 flash and it got it wildly wrong as well. Seems strange AI seems to struggle with it.

        • @[email protected]
          link
          fedilink
          English
          12
          edit-2
          4 months ago

          It’s something to do with the word being 2 tokens and it not knowing the tokens before or after the current one.

          It’s a simple example of It’s inability to actually think and reason.

    • @[email protected]
      link
      fedilink
      English
      84
      edit-2
      4 months ago

      Man, I want to be at that shareholder meeting; “how about we just don’t have a CEO and pocket the savings?”

          • @[email protected]
            link
            fedilink
            English
            54 months ago

            Ironically, a lot of stock trading already is done by bot, it’s a problem from a while ago. Not LLM, of course, much simpler but it’s why it sometimes goes hive mind.

            • @[email protected]
              link
              fedilink
              English
              54 months ago

              Sure, but a bot initiating a trade is a whole lot different than a bot owning the stock.

  • @[email protected]
    link
    fedilink
    English
    145
    edit-2
    4 months ago

    McDonald’s couldn’t even get AI to take drive-through orders properly. Musk wants it to run the government without even doing a test run first.

    • @[email protected]
      link
      fedilink
      English
      364 months ago

      He’s either

      • an incompetent moron who thinks AI eventually will be like in the movies, you just need to push throught its teething pains,
      • someone that tries to destroy the government and its services through weaponized incompetence, so they can be privatized,
      • both, because I saw way too many things in real life.
    • @[email protected]
      link
      fedilink
      English
      92
      edit-2
      4 months ago

      That’s because he and his kind believe government is useless and can just be broken without losing anything important. From their point of view, government is just a thing that takes money from them and spends it on people who don’t deserve to live because they’re not asshole billionaire techbros. And it makes poor people’s lives slightly less unpleasant by giving them money and services, which billionaires don’t like because it makes the poor less desperate and exploitable.

      • @[email protected]
        link
        fedilink
        English
        334 months ago

        That’s because he and his kind believe government is useless and can just be broken without losing anything important.

        kinda like racks of servers that keep a social media site used by 100s of millions of users? the ones muskrat just yanked willy-nilly out of a data center and loaded onto uhauls? it took weeks for twitter to repair that damage.

  • Kairos
    link
    fedilink
    English
    174 months ago

    They want this so they can blame the computer when it makes a bad decision which is actually just parroting what they want

  • @[email protected]
    link
    fedilink
    English
    7
    edit-2
    4 months ago

    The overall goal is to cut the agency’s budget by fifty percent. Shedd suggested using AI to analyze contracts for redundancies, root out fraud, and facilitate a reduction in the federal workforce by automating much of their work.

    I am bullish on AI in the long run.

    I am skeptical that given the state of affairs in 2025, you can reasonably automate half of the federal government, via AI or any other means.

    I also don’t think that the way to do this is to lay off half of the federal workforce and then, after the fact, see what can be automated. If you look at the private sector automating things, it tends to hedge its bets. Take self-service point-of-sale kiosks. We didn’t just see companies simply lay off all cashiers. Instead, we saw them brought in as an option, then had the company look at what worked and what didn’t work – and some of those were really bad at first – and then increase the rate of deployment once it had confidence in the solution and a handle on the issues that came with them.

    • @[email protected]
      link
      fedilink
      English
      44 months ago

      Yeah, but you and much of the business world have intelligence and strategy. Elon is the guy who thinks he can just pay some Chinese gamer to play a game for him, then pretend he did it himself; that’s his version of “brilliant strategist.”

      It’s no wonder he can’t figure out how to automate anything safely or correctly, because he doesn’t actually understand how to do anything himself, and he can’t just pay some Chinese rando to do it for him.

      • @[email protected]
        link
        fedilink
        English
        44 months ago

        I worked as a consultant for a long time. I learned that anyone who starts a question with “Why don’t we just…” generally doesn’t understand the problem.

  • @[email protected]
    link
    fedilink
    English
    314 months ago

    Assuming we correct course and stop our spiral into authoritarianism in 4 years, we’re going to have to completely rebuild our entire federal tech infrastructure. It simply can’t be trusted after these traitors. The possibility of backdoors is simply too high. It’s going to cost billions and billions of dollars.

    We’re all going to need new Social Security numbers, because this dumbfuck now has them all on his private servers.

  • @[email protected]
    link
    fedilink
    English
    18
    edit-2
    4 months ago

    And if when the AI fails, buddy billionaires will be there to offer privatized alternative, for a fee of course.

  • @[email protected]
    link
    fedilink
    English
    344 months ago

    DOGE Plan to Push AI Across the US Federal Government is Wildly Dangerous.

    Fixed that for you.

  • @[email protected]
    link
    fedilink
    English
    844 months ago

    All of DOGE “plans” seem like shit that was cooked up by immature teenagers and drug addled brains, because they are.

  • @[email protected]
    link
    fedilink
    English
    14 months ago

    You generally won’t understand another person (and adversary especially) if you don’t see how their actions perfectly make sense for them, and without conspiracies.

    So - there is one matching variant, that Musk sincerely hates bureaucratic kinds of power, but not proprietary kinds of power. Replacing a bureaucrat with (some imagined good) AI in another assumption would be replacing a mediocre human with inherent lust for power with an unreliable automaton, but without lust for power. The good part here is that humans are unreliable too and working bureaucracies compensate for that.

    The bad part is that for every failure a person should be responsible proportionally to their input. I’m not sure they’ll do that, or I’m sure they won’t.

      • @[email protected]
        link
        fedilink
        English
        14 months ago

        Yes, but corporate bureaucracy is someone’s property, so ultimately there is a responsible person, always.