Dynamic Refusal Trajectories for Robust Jailbreak Detection
Discover how dynamic refusal trajectories and SALO improve AI jailbreak detection, boosting security against adversarial attacks with over 90% accuracy.
Kernel Affine Hull Machines for Fast Semantic Query Encoding
Discover Kernel Affine Hull Machines (KAHMs) for efficient, low-latency query-side semantic encoding with high retrieval accuracy and reduced computational...
AsymK-Talker: Real-Time AI Talking Head Generation
Discover AsymK-Talker, a novel AI method enhancing real-time and long-horizon talking head generation with improved visual fidelity and lip sync.
Machine Learning Predicts Euler Characteristics in Topology
Discover how machine learning predicts Euler characteristics from geometric images, advancing topological analysis with minimal data and physics-informed m...
RouteHijack: Exploiting Routing Vulnerabilities in MoE LLMs
Discover how RouteHijack targets routing in Mixture-of-Experts LLMs, revealing critical vulnerabilities and advancing AI safety research.