Post B8ffm0LIeKgI8lrgy8 by icedquinn@blob.cat
 (DIR) More posts by icedquinn@blob.cat
 (DIR) Post #B8fdayqLgR1OOg0H2W by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @atax1a ah, i think that is the point of confusion: when those people say it is "getting better", what they mean is that it can slop in more cases
       
 (DIR) Post #B8fdg9GeM0CMAulnSi by atax1a@infosec.exchange
       0 likes, 0 repeats
       
       @ariadne i, too, can improve my ability to take a big nasty shit in more cases
       
 (DIR) Post #B8feqjPQw12t7HULuS by icedquinn@blob.cat
       0 likes, 0 repeats
       
       @atax1a actually ended up taking out all of the code from GLM and went back to generating it with macros from xml in the one project i ever used a bot on.its a real crapshoot when the things properly infer "unroll this stream in to an outline based on this stack machine principle" and when they absolutely fuck up "your job is to just check this in a single order."
       
 (DIR) Post #B8ffm0LIeKgI8lrgy8 by icedquinn@blob.cat
       0 likes, 0 repeats
       
       @atax1a i think there's a fundamental issue the LLM is basically a cheap intern but its still sidestepping that you shouldn't need interns.people will deadass use a hyperscaler datacenter and two million tokens instead of just write a little xml file and template expand it to code :blobcatgoogly: