About this document
Graph of Attacks With Pruning: Optimizing Stealthy Jailbreak Prompt Generation For Enhanced LLM Content Moderation by WalterHu is a document available to read on EtoBox.
The document presents the GAP (Graph of Attacks with Pruning) framework, which enhances the generation of stealthy jailbreak prompts for large language models (LLMs) to improve content moderation. GAP outperforms existing methods by achieving a 20.8% increase in attack success rates and a 62.7% reduction in query costs, while also providing specialized variants for automated seed generation and multimodal attacks. The framework
- Author
- WalterHu
- Language
- EN