Defending Against Adversarial Prompting Using Perplexity