The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety
arxiv
On the Identifiability of Steering Vectors in Large Language Models
arxiv
UniSAFE: A Comprehensive Benchmark for Safety Evaluation of Unified Multimodal Models
arxiv
ImplicitBBQ: Benchmarking Implicit Bias in Large Language Models through Characteristic Based Cues
arxiv
Overcoming Joint Intractability with Lossless Hierarchical Speculative Decoding
arxiv