Cool idea. I think LLMs aren’t built for intentional error. They’re wired to optimize meaning—next tokens chosen from attention scores, beam search. You can’t just flip logic and get creativity. You either get coherence or gibberish. If you want poetic mistakes, train on surreal input—but I wouldn’t expect avant-garde results.
it was just my idea, assuming that biggest players will not give a f..ck. With premeditation i am using rule 8+2 with many diffrent AIs, and asking same question for the same. If with same coding answers are way diffrent and intersting