No, I'm saying it is not necessary to observe an output about ethics or empathy. Large language models work from the large corpuses of human language, which encodes these concepts in itself