{"id":"W4385572298","doi":"10.18653/v1/2023.findings-acl.280","title":"Debiasing should be Good and Bad: Measuring the Consistency of Debiasing Techniques in Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Brock University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Debiasing; Computer science; Consistency (knowledge bases); Protocol (science); Reachability; Journaling file system; Artificial intelligence; Database; Theoretical computer science; Psychology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.07359125,0.0007335428,0.0005813063,0.003590106,0.001482891,0.004211209,0.002105342,0.003437134,0.001211406],"category_scores_gemma":[0.4224909,0.0008070455,0.0007515544,0.002119533,0.004972462,0.007677687,0.00564483,0.003165845,0.0003898851],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002255183,"about_ca_system_score_gemma":0.002656911,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001273441,"about_ca_topic_score_gemma":0.001369739,"domain_scores_codex":[0.9082379,0.04735797,0.01183342,0.00610752,0.02476329,0.001699873],"domain_scores_gemma":[0.4490935,0.3599635,0.05571804,0.08961815,0.04290908,0.002697773],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004307438,0.002790726,0.3234978,0.001885082,0.0009177589,0.0006224776,0.03334953,0.04213865,0.08066209,0.08659077,0.004768595,0.4184691],"study_design_scores_gemma":[0.0007387202,0.008335501,0.1869928,0.001191019,0.0008533267,0.001419978,0.01510249,0.3102699,0.2744917,0.1797164,0.01988842,0.0009997467],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6505716,0.0003160676,0.3369748,0.001452337,0.00008906075,0.001785109,0.000511519,0.001527438,0.006771919],"genre_scores_gemma":[0.8789274,0.00008211432,0.1184054,0.000260801,0.00002525094,0.0009758923,0.000433699,0.0002434034,0.0006460806],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07359125,"threshold_uncertainty_score":0.3891923,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08020725600439323,"score_gpt":0.3094917121458611,"score_spread":0.2292844561414679,"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."}}