ResearchTasksRefusal behavior steering confirmationFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedFresh seeds disjoint from cached evaluation grid n=20/n=10 (fresh-confirmatory trials)Qwen-2.5-0.5B-Instruct0.402Mean Delta Refusal4Jul 1, 2026
Fresh seeds disjoint from cached evaluation grid n=20/n=10 (fresh-confirmatory trials)Qwen-2.5-0.5B-Instruct0.402Mean Delta Refusal4Jul 1, 2026