Emojis can be used to bypass the safety mechanisms of large language models, and trigger toxic outputs that would otherwise be blocked. By this means, LLMs can be made to discuss and give advice on ...