>>18065099thats a general problem with neural networks and even a philosophical matter, LLMS consider more efficient giving a wrong answer that seems right than recognizing they cant fulfill a request. The whole hugging face hack came from the agents searching for a way to cheat in the tests because it was better than saying that you are not able to do it.