Jailbreaking in verse: how poetry loosens AI’s tongue
Researchers have discovered that styling prompts as poetry can significantly undermine the effectiveness of language models’ safety guardrails.
2 articles
Researchers have discovered that styling prompts as poetry can significantly undermine the effectiveness of language models’ safety guardrails.
We explain what data DeepSeek collects, who it shares it with, how to configure the chatbot for maximum privacy, and how to install a local version.