{"id":"W4389523803","doi":"10.18653/v1/2023.findings-emnlp.983","title":"GTA: Gated Toxicity Avoidance for LM Performance Preservation","year":2023,"lang":"en","type":"article","venue":"","topic":"Hate Speech and Cyberbullying Detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Institute for Information and Communications Technology Promotion; Ministry of Science and ICT, South Korea","keywords":"Offensive; Perplexity; Computer science; Consistency (knowledge bases); Artificial intelligence; Grammar; Generative grammar; Natural language processing; Machine learning; Language model; Operations research; Engineering; Linguistics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001514917,0.002171777,0.0009988782,0.001278974,0.0008663701,0.001511549,0.002518502,0.001877315,0.01112438],"category_scores_gemma":[0.006957088,0.000507078,0.001039812,0.0009927214,0.0008212399,0.001730979,0.002141338,0.002006841,0.007118806],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005754212,"about_ca_system_score_gemma":0.001672906,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003068563,"about_ca_topic_score_gemma":0.005256106,"domain_scores_codex":[0.9987621,0.0004008902,0.0001053452,0.000318889,0.0002985625,0.0001141004],"domain_scores_gemma":[0.9974648,0.001046107,0.0001528735,0.000840971,0.0003745579,0.0001206886],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0006578675,0.0001892235,0.001971397,0.0006030217,0.0001958567,0.0007439964,0.0002907043,0.0522173,0.1155389,0.004144333,0.04170085,0.7817466],"study_design_scores_gemma":[0.0002833123,0.0005484291,0.002508499,0.00007099917,0.0001744551,0.0009534032,0.0002046578,0.815977,0.135752,0.01151488,0.03190439,0.0001079612],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07083689,0.00318011,0.809918,0.001279743,0.001189294,0.0005182513,0.002115721,0.1025669,0.008395255],"genre_scores_gemma":[0.5266415,0.000697048,0.4428727,0.001635449,0.0004434347,0.0005691639,0.006942438,0.006527246,0.0136711],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01112438,"threshold_uncertainty_score":0.03721476,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02641855170120785,"score_gpt":0.2470416496224299,"score_spread":0.2206230979212221,"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."}}