AdvBench

A new paper from researchers systematically evaluating widely used AI safety datasets AdvBench and HarmBench finds that these benchmarks contain contrived data points with excessive triggering cues an…

A per-layer ablation study on Llama-3.1-8B-Instruct found that refusal behavior is redundantly distributed across layers, not localized to a single layer, contradicting prior claims by Arditi et al. t…