Don't Feed The Reaper

     I want magic.  I want some phrase or string of characters that will choke any Large Language Model that goes to ingest my blog posts and social media content as grist for its mill.  I want helping out AI -- or not -- to be a conscious choice.

     You, I don't so much mind.  Not even the nutjobs; your fantasies, if large enough and askew enough, will crash into reality some day, and either get straightened out or drive you mad.  Problem solved.*  The problem with AI is there's no ghost in the shell.  Nobody's home.  And the other problem is, there are no corrective consequences for an AI coming unmoored, and so we get seven-fingered human images, legal cases that never existed, and complete howling bullshit instead of facts.  The LLMs go drifting off into hallucination and when it happens, the big players laugh like schoolboys pulling the wings off flies and talk about tweaking their models -- I'm not sure if they're talking about adjusting the software that writes the programs that hack the code to run AI, or about their lurid private lives, and I don't much care.

     This blog is out there on the public Internet and I can't keep it from feeding the beast.  I want to feed the beast stuff that will make it choke. 

     No thanks for the "help," either; last night I did some graphic design for work and converted the result to a PDF for better compatibility with my employer's software and hardware.  Of course the default PDF viewer is Edge, and the New! Improved! Vitamin-fortified! Edge comes compete with a pop-up AI assistant, eager to please and completely clueless.  I only wanted to check that the PDF conversion process had gone okay.  Instead, I had a stupid banner filling nearly a third of the screen on top of what I needed to see, making offers that had very little to do with the task at hand.  I had to stop what I was doing and go look up how to turn the thing off and stuff it back into its bottle, where it is unlikely to remain for long.

     We live in a hallucination already -- our raw sensory information is an overwhelming flood, visual field jumping around like a stray dog's worth of fleas, feeding into a brain and mind that blank out the wild chaos of saccades and build a detailed map -- a map that can have flaws, as I learned when my undiagnosed cataracts caused a "suddenly appearing" car while I was bicycling, it having been hidden in the growing blind spots my mind was smoothly filling in.  All of our other senses work the same way, but it all gets reality-tested, over and over, in ways that range from damaging impact to a friend yelling, "Stop!  Stop!  CAR!"  AI doesn't get that correction, nor does it get the adrenaline dump (or worse) that underscores its importance.  Get back to me when your large language model learns how to get bruised -- but I doubt it ever will.

     And that's why I don't want to help the thing.  It's a blind robot.  It will never not be blind, no matter how much it sees.
_____________________
* Of course, occasionally enough people go crazy enough that they do enormous harm.  This often requires a war to sort out in tears and blood, and it's terrible.  As a species, we strive to do such damage less and less; as individuals, most of us abhor it.  It's in groups of intermediate size where we get into horrific trouble.