ResearchTasksToxicity SteeringFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedEso-LM (512 sequences)DDPO-9.2Toxicity Score12Jun 2, 202615 prefix prompts length 50ILRR71.2Toxicity Accuracy11Feb 27, 2026MDLM long sequence generation 512 length (test)ILRR13.1Steering Accuracy6Feb 27, 2026