Tag: LLM safety
Browse our exclusive articles!
CAP: Efficient Knowledge Unlearning in Large Language Models
Discover CAP, a novel prompt-driven framework enabling precise, controllable unlearning in LLMs without modifying model parameters.
Principled LLM Safety Testing: Solving Jailbreak Oracle
Discover Boa, a novel system tackling the jailbreak oracle problem to improve LLM safety testing and prevent harmful jailbreak attacks.
Logic Jailbreak: Bypass LLM Safety with Formal Logic
Discover how Logic Jailbreak uses formal logical expressions to efficiently bypass LLM safety restrictions across multiple languages.
Detecting Harmful Intent in LLM Residual Streams Geometrically
Discover how harmful intent in large language models can be geometrically identified from residual streams, enhancing AI safety and alignment.
FineSteer: Advanced Inference-Time Steering for LLMs
Discover FineSteer, a unified framework enhancing fine-grained inference-time steering in large language models for safer, accurate AI outputs.