SafeRBench introduces a novel pipeline for evaluating the safety of Large Reasoning Models by decomposing reasoning traces into micro-thought chunks for fine...
Level: advanced
By Xin Gao and 10 other authors
Category: discussion