Alignment is not sufficient to prevent large language models from generating harmful information: A psychoanalytic perspective

Open in new window