
With a few selected prompts, I can identify which LLM engine is being used by #ChatbotArena. At this point, if I am a malicious agent, I can carry out several rounds of evaluation, and each time I identify my target LLM, I can vote accordingly.

English














