• 0 Posts
  • 15 Comments
Joined 9 days ago
cake
Cake day: January 22nd, 2025

help-circle

  • A WELL TRAINED AI can be a very useful tool. However the AI models that corporations want to use aren’t exactly what I’d call “well trained” because that costs money. So they figure “we’ll just let it learn by doing. Who cares if people get hurt in the meantime. We’ll just blame the devs for it being bad.”

    Edit: to add this is partly why AI gets a bad rap from folks on the outside looking it. Corporations institute barebones, born yesterday AI models that don’t know their ass from their elbow because they can’t be bothered to pay the devs to actually train them but when shit goes south they turn around and blame the devs for a bad product instead of admitting they cut corners. It’s China Syndrome but instead of nuclear reactors it’s AI.












  • I can agree with both points but I understand Viking’s points too.

    I work with high pressure hydraulic systems and our gauges that we use for testing and operational checks have to be calibrated to within a very small margin per the manufacturer, if it’s unable to be calibrated to within that margin or if it’s been recalibrated X amount of times we have to toss it and replace it (they’re expensive btw…). To the lay man, a difference of 50psi might seem inconsequential and a stupid reason to throw it out but for someone like me who is standing next to something that’s already pressurized to its “proof pressure” which is usually multiple times higher than it’s operating pressure, that 50psi is the difference between me going home that evening or me getting impaled by an accumulator that blows apart and pins me to the wall.

    For an example more akin to what Viking is talking about, see the Therac-25 radiotherapy machine, where improper use, training and maintenance led to several otherwise preventable deaths.