1 paper
Zhengyi Jin, Ru Zhang, Xiao Chen +5
Fragmented safety evaluation undermines the governance of dangerous AI capabilities. We present a modular framework that evaluates each model through three orthogonal pipelines---K…