{"id":"W4402671660","doi":"10.18653/v1/2024.acl-long.600","title":"Measuring Political Bias in Large Language Models: What Is Said and How It Is Said","year":2024,"lang":"en","type":"article","venue":"","topic":"Hate Speech and Cyberbullying Detection","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Politics; Computer science; Natural language processing; Political science; Law","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":[0.0110216,0.001046633,0.0008535319,0.003192497,0.0008141097,0.003853251,0.0007657533,0.001247813,0.002428508],"category_scores_gemma":[0.05670015,0.0004871775,0.0009275966,0.002317437,0.001218073,0.005487596,0.002288592,0.002216849,0.001242648],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001145064,"about_ca_system_score_gemma":0.00130039,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00414972,"about_ca_topic_score_gemma":0.005693628,"domain_scores_codex":[0.993291,0.004025837,0.0004274652,0.0009301833,0.001043278,0.0002821506],"domain_scores_gemma":[0.9418938,0.04661741,0.003980986,0.003799727,0.003083139,0.0006248553],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001940952,0.0007797292,0.3947831,0.001588943,0.001445904,0.0004979383,0.01001834,0.08012399,0.02712578,0.05292872,0.01388292,0.4148837],"study_design_scores_gemma":[0.0001633285,0.000382008,0.08159788,0.0004134031,0.000566185,0.0003548249,0.002123673,0.7006428,0.01386154,0.179619,0.02002865,0.0002466394],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.542671,0.002429552,0.4267532,0.00497771,0.0002678049,0.0003473307,0.006646812,0.003596889,0.01230977],"genre_scores_gemma":[0.9470702,0.0003083803,0.04770927,0.0006852928,0.0002070586,0.0002328016,0.002655972,0.0002547854,0.0008762337],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0110216,"threshold_uncertainty_score":0.05828846,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05210208125202521,"score_gpt":0.263815868988379,"score_spread":0.2117137877363538,"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."}}