Introducing Bergeron to enhance the robustness of Large Language Models against adversarial attacks without additional training.
