Fangzhao Wu

Open-Source Can Be Dangerous: On the Vulnerability of Value Alignment in Open-Source LLMs