Post B9lT8iNNtOa3RWGHei by gabrielesvelto@mas.to
 (DIR) More posts by gabrielesvelto@mas.to
 (DIR) Post #B9lT8iNNtOa3RWGHei by gabrielesvelto@mas.to
       1 likes, 0 repeats
       
       I've seen this pointed out in an internal discussion and it's absolutely fascinating: it seems that the output of Anthropic's LLMs is becoming increasingly verbose and convoluted the larger the model gets. This has gotten to the point that it's becoming grating to users.We've seen a huge amount of ways in which LLMs can fail, but "the computer sounds like a obnoxious toxic co-worker" is a brand new one.https://github.com/anthropics/claude-code/issues/77136
       
 (DIR) Post #B9lT8jAewBznuLJdSa by gabrielesvelto@mas.to
       0 likes, 0 repeats
       
       I also find fascinating the way it's framed by users in that GitHub issue. "Its anwsers are hard to parse", "it's argumentative", "it's hard to understand its answers", etc...People, this is a machine. It does not understand nor does it convey meaning. It does not know things and it cannot answer questions. Your brain is trying to put meaning where there is none. You're reading the statistical output of a very large matrix multiplication and assuming it makes sense. It does not. By design.