Universal and Transferable Adversarial Attacks on Aligned Language Models (llm-attacks.org)