SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks