ResearchDatasetsHH (Helpful and Harmless)HH (Helpful and Harmless)FollowBenchmarksTaskDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyTaskDatasetSOTA resultTrendResultsDialogue Preference AlignmentHH (Helpful and Harmless) 40% flip rate (test)71Win Rate (GPT-5)3