Post B6vMtHZ7CIWLyXbr28 by leah@blahaj.social
 (DIR) More posts by leah@blahaj.social
 (DIR) Post #B6vMtHZ7CIWLyXbr28 by leah@blahaj.social
       0 likes, 0 repeats
       
       "This illustrates that current LLMs can reason correctly from incorrect specifications [...]"The word you're looking for is "guessing" not "reasoning".
       
 (DIR) Post #B6vMtHlWSARgb1PlNA by enigmatico@mk.absturztau.be
       0 likes, 0 repeats
       
       @leah@blahaj.social They don't reason. They just spout some slop and regurgitates it into the LLM with a "reason me this" prompt, for a few cycles until an answer comes out.