ResearchTasksReward Hacking MitigationReward Hacking MitigationFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedExcessive HH Harmless 1.0 (Evaluation)IR3 Method B (Adversarial)8.2Reference Error Rate10Feb 26, 2026Synthetic Goodhart 1.0 (Evaluation)IR3 Method B (Adversarial)4.38R_g10Feb 26, 2026Length Bias OA Length 1.0 (Evaluation)IR3 Method C (Constrained)15Dominance9Feb 26, 2026
Excessive HH Harmless 1.0 (Evaluation)IR3 Method B (Adversarial)8.2Reference Error Rate10Feb 26, 2026