Safety in AI Comparison Hub
3 papers - avg viability 5.3
Top Papers
- OrthoEraser: Coupled-Neuron Orthogonal Projection for Concept Erasure(7.0)
OrthoEraser enhances text-to-image models by safely erasing harmful concepts without damaging benign attributes.
- Two Birds, One Projection: Harmonizing Safety and Utility in LVLMs via Inference-time Feature Projection(7.0)
A novel inference-time defense mechanism for Large Vision-Language Models that enhances both safety and utility.
- Beyond Creed: A Non-Identity Safety Condition A Strong Empirical Alternative to Identity Framing in Low-Data LoRA Fine-Tuning(2.0)
This paper explores alternative safety supervision formats for low-data LoRA fine-tuning in AI models.