Emergent Misalignment and Persona Collapse in LLMs
Explore emergent misalignment in large language models causing persona-model collapse and ethical risks in AI fine-tuning with harmful content.
AssemblyBench: Advanced Physics-Based Industrial Assembly Dataset
Discover AssemblyBench, a physics-aware dataset enhancing AI-driven assembly of complex industrial objects with 3D models and multimodal instructions.
Bayesian Model Merging: Efficient AI Model Integration
Discover Bayesian Model Merging, a novel AI technique improving multi-task model integration with optimized hyperparameters and strong priors.
GraphIP-Bench: Protecting Graph Neural Networks from Theft
Discover how GraphIP-Bench evaluates the risks of stealing Graph Neural Networks and tests defenses to safeguard AI models in cloud services.
FRAME: Advanced Image Manipulation Detection Method
Discover FRAME, a cutting-edge forensic tool using adaptive multi-path evidence fusion to detect and verify image manipulation accurately.