XPost: alt.folklore.computers   
   From: bowman@montana.com   
      
   On Fri, 2 Jan 2026 11:13:59 +0000, The Natural Philosopher wrote:   
      
   > On 02/01/2026 04:35, c186282 wrote:   
   >> NNs are 'different'. Not 'expert', not 'fuzzy', not LLM.   
   >>    A little closer to how biological brains work. The bitch has been   
   >>    finding suitable elements that can be compactly put on chips.   
   >>    They're getting better at that. Maybe 10 years and decently good   
   >>    'AI' will fit INSIDE a bot instead of a 20 acre gigawatt data   
   >>    center.   
   >   
   > Yes. They are ultimately pattern recognition engines.   
   >   
   > Trouble with those is you have to get the gain right, I cant remember   
   > what happened to that software you fed images too and it turned them   
   > into eyes, and dogs where there used to be plants. Because it tried too   
   > hard.   
   >   
   > Great fun   
      
   The 'hello world' of image recognition is classify dogs and cats. There is   
   a very large dataset of cat and dog images to work with.   
      
   One of the early problems was the dogs tended to be photographed outside   
   and the cats inside. After training the model was very good in classifying   
   furry animals in an outside setting versus those inside.   
      
   Speaking in an anthropomorphic way classifiers can have acceptable   
   behavior but you're never too sure exactly what they're 'thinking'. There   
   is a whole field of research trying to figure out what the hell goes on in   
   the black box.   
      
   At least with a classifier it's easy to see a problem if it calls a Great   
   Dane a horse but LLM fantasies tend to get accepted as facts.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)