{"id":"W4412373306","doi":"10.56553/popets-2025-0122","title":"An Analysis of Chinese Censorship Bias in LLMs","year":2025,"lang":"en","type":"article","venue":"Proceedings on Privacy Enhancing Technologies","topic":"Computational and Text Analysis Methods","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Censorship; Psychology; 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":[0.01818955,0.0008290499,0.0006844205,0.001970052,0.001047153,0.001769808,0.001223587,0.0009708332,0.002154488],"category_scores_gemma":[0.07300234,0.0002986905,0.000849986,0.001709368,0.001682742,0.002984586,0.001807586,0.00186702,0.0006443021],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002898565,"about_ca_system_score_gemma":0.001549791,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01641624,"about_ca_topic_score_gemma":0.01311202,"domain_scores_codex":[0.9926409,0.00353001,0.0006076111,0.001284288,0.00155675,0.0003803724],"domain_scores_gemma":[0.905142,0.07473359,0.005572887,0.008169832,0.005481313,0.0009004379],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00330423,0.0004841021,0.4250666,0.001051578,0.0007065483,0.001249432,0.008327714,0.1834424,0.01431963,0.01798137,0.01295643,0.33111],"study_design_scores_gemma":[0.00009834359,0.0005370692,0.1051245,0.0001492221,0.0002110398,0.0004297291,0.001245881,0.8562815,0.02039365,0.009236204,0.006137346,0.0001554378],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9545249,0.0008252443,0.037202,0.0007657211,0.00009782935,0.0001737057,0.001206149,0.001288328,0.003916019],"genre_scores_gemma":[0.990346,0.00009077598,0.007004183,0.0001174157,0.0000309489,0.00007359077,0.001220483,0.00008647181,0.001030133],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01818955,"threshold_uncertainty_score":0.09619671,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0443225690858431,"score_gpt":0.3945897658567422,"score_spread":0.3502671967708991,"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."}}