{"id":"W4285891774","doi":"10.1007/s10489-022-03944-z","title":"Reward Modeling for Mitigating Toxicity in Transformer-based Language Models","year":2022,"lang":"en","type":"article","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":28,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Computer science; Language model; Transformer; Artificial intelligence; Machine learning; Natural language processing; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003998332,0.0001226307,0.000155229,0.0001917669,0.0002541016,0.00002972683,0.0008078031,0.00004007779,0.00001130746],"category_scores_gemma":[0.00001420268,0.0001589534,0.00011156,0.0005437676,0.00001554633,0.000484132,0.0001243502,0.0002069155,0.000001430965],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002239395,"about_ca_system_score_gemma":0.0001135168,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002075811,"about_ca_topic_score_gemma":0.00008916737,"domain_scores_codex":[0.998764,0.0000796096,0.0001712312,0.0005534496,0.00009443673,0.0003373182],"domain_scores_gemma":[0.9993457,0.00008234397,0.00005204145,0.00041238,0.00003411423,0.00007344767],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001756098,0.00003922932,0.0001286764,0.00001460029,0.000004343908,0.0000377955,0.0009165518,0.931934,0.0004265532,0.06551198,0.000003718217,0.000965015],"study_design_scores_gemma":[0.0007891064,0.00004800065,0.00000300018,0.00001165853,0.000005949146,0.000001501523,0.0007645424,0.9780133,0.000475292,0.01969163,0.00001425836,0.0001817042],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4554477,0.0000135919,0.5438341,0.0000912292,0.00005219718,0.0001472978,0.000005577562,0.00007588137,0.0003323951],"genre_scores_gemma":[0.9835792,0.00000183062,0.01602891,0.000242452,0.00001660217,0.000005290264,0.000004520513,0.00001083822,0.0001103443],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5281315,"threshold_uncertainty_score":0.6481931,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08285491862021747,"score_gpt":0.1938981566713769,"score_spread":0.1110432380511594,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}