Comment by codelion

Interesting work. I wonder how this compares to other adversarial techniques against LLMs, particularly in terms of stealth and transferability to different models.