{"id":"W4401952885","doi":"10.1038/d41586-024-02527-x","title":"LLMs produce racist output when prompted in African American English","year":2024,"lang":"en","type":"article","venue":"Nature","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"Ubisoft (Canada)","funders":"","keywords":"Covert; Harm; Racism; Race (biology); Political science; Criminology; Sociology; Psychology; Gender studies; Social psychology; 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":[0.0007389579,0.0006283784,0.0003782225,0.0003834566,0.0007423129,0.001119717,0.0003255873,0.000942686,0.02572038],"category_scores_gemma":[0.01023064,0.0002080421,0.0001881353,0.000330283,0.0003330012,0.0008332914,0.0009067787,0.000705247,0.009046433],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004608429,"about_ca_system_score_gemma":0.0004402412,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00560901,"about_ca_topic_score_gemma":0.007721694,"domain_scores_codex":[0.9992773,0.0003077242,0.0000378293,0.0001274601,0.0001656279,0.00008410633],"domain_scores_gemma":[0.9890352,0.008546139,0.0004081342,0.0005540515,0.001132637,0.0003237229],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.007228666,0.0007793983,0.06412737,0.002340686,0.0001442074,0.007344712,0.03404633,0.003882887,0.4172983,0.0100789,0.1516684,0.3010601],"study_design_scores_gemma":[0.0005105483,0.001512869,0.2379667,0.0008054159,0.0004310579,0.004407702,0.03539509,0.1233274,0.3412853,0.00952426,0.2444965,0.0003372392],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8498598,0.0001598112,0.01779212,0.001691913,0.000564659,0.0001495339,0.005071176,0.01567599,0.1090349],"genre_scores_gemma":[0.9579694,0.00008695557,0.01393319,0.0005414418,0.00007760786,0.00007732602,0.003327075,0.002053962,0.02193303],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02572038,"threshold_uncertainty_score":0.0860433,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.006657641511243997,"score_gpt":0.2659464306549486,"score_spread":0.2592887891437046,"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."}}