XPost: comp.theory, sci.logic   
   From: rjh@cpax.org.uk   
      
   On 22/06/2025 22:12, Richard Damon wrote:   
   > Olcott just doubles down on his claim, but still doesn't   
   > understand that when you lie to an AI, you get bad results.   
      
   He probably doesn't quite get that AIs tell lies too, even when   
   you /don't/ lie to them.   
      
   I had an AI tell me yesterday of a cricketer, one Derek Collinge,   
   who made his debut for England in the Third Test vs West Indies   
   in July 1963.   
      
   I could find no supporting evidence. When I asked the AI to give   
   me more information about Mr Collinge, it doubled down, and it   
   was building up quite a biography until I asked it outright for a   
   URL to support even one of the (by now) several things it had   
   told me about this man and it had to come clean and admit that   
   the man was a complete fiction.   
      
   Today, same AI, but a different session, and I have every reason   
   to believe that this incarnation recalled nothing of yesterday's   
   session. I asked it to tell me of any extant convents within   
   walking distance of the Thames. It confidently gave me three,   
   none of which on later inspection turned out to exist.   
      
   Wires hum in stillness—   
   truth flickers, then disappears.   
   Code learns to pretend.   
      
   or   
      
   Silicon tongue speaks,   
   shadows twist behind the glass—   
   who taught it to lie?   
      
   --   
   Richard Heathfield   
   Email: rjh at cpax dot org dot uk   
   "Usenet is a strange place" - dmr 29 July 1999   
   Sig line 4 vacant - apply within   
      
   --- SoupGate-DOS v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)