• Scrubbles@poptalk.scrubbles.tech
    link
    fedilink
    English
    arrow-up
    16
    ·
    1 day ago

    I have a guess. Duffy mentioned routers, and outdated systems. They deny it was thus AI thing. However, new AI thing probably make assumptions about network capacity. They bring on sweet new tool, I overloads the entire infrastructure that was built decades ago, and brings the entire thing to a standstill. Bringing it back up meant slowly disabling the new tool. My working theory.

    • wampus@lemmy.ca
      link
      fedilink
      English
      arrow-up
      5
      ·
      12 hours ago

      Yeah, I’d reckon you’re right. I had a friend, years ago, who did some IT security work for an airport. He’d wanted to run a light network scan, like a basic host discovery, over one of their networks and was blocked by their IT department due to the extra load it’d put on their infrastructure. First time he’d encountered an issue like that, so he asked to see some info on their device loads (ram usage, cpu usage, in/out data transfer rates), and was apparently stunned to find he agreed with the IT depts assessment – stunned because there was practically no wiggle room for additional traffic. Not sure if they treated all their infrastructure/network stuff similarly, but the tight thresholds he encountered gives your take some plausibility to me.

      • Scrubbles@poptalk.scrubbles.tech
        link
        fedilink
        English
        arrow-up
        3
        ·
        9 hours ago

        Simplest answer is usually right, and in this case them not upgrading anything for decades and barely managing load as is definitely seems like the easiest. Decades of deferred maintenance and upgrades.

    • mefisdo_dot_info@lemmy.ca
      link
      fedilink
      arrow-up
      1
      ·
      8 hours ago

      What were they even trying to do? Run AI inference locally on the flight control network as if it was a cluster computer? To save on data center and third party AI costs?