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

Zi Yin, Wei Ding, Jia Liu

Large Language Models (LLMs) are central to a multitude of applications but struggle with significant risks, notably in generating harmful content and biases. Drawing an analogy to…