{"id":"W4403143453","doi":"10.2196/52746","title":"Performance of ChatGPT on Nursing Licensure Examinations in the United States and China: Cross-Sectional Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Licensure; China; Medical education; Medicine; Psychology; Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004322272,0.0004131674,0.0005206123,0.001843902,0.000901047,0.0008273949,0.0007684555,0.0007086509,0.001650052],"category_scores_gemma":[0.01194458,0.0004160596,0.0007362643,0.001704602,0.0008889371,0.001313838,0.001627167,0.0008004393,0.0005297334],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001657739,"about_ca_system_score_gemma":0.001451506,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0372349,"about_ca_topic_score_gemma":0.04181606,"domain_scores_codex":[0.9975817,0.0005876524,0.0003932762,0.0004508594,0.0006625516,0.0003239345],"domain_scores_gemma":[0.990428,0.002008448,0.002939124,0.0007091474,0.002647202,0.001268176],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002238054,0.00005172125,0.9974346,0.00001249262,0.00001677957,0.00004115743,0.0008850573,0.00002993876,0.00005481316,0.0000103956,0.0001047429,0.001336011],"study_design_scores_gemma":[0.000001272213,0.00008303641,0.9984528,0.000006981809,0.000009514185,0.00004769703,0.001019111,0.0001899398,0.00005139887,0.000005684382,0.0001274947,0.000005140454],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9995,0.00004142766,0.00003712281,0.0000263633,0.000003443685,0.00001670515,0.0001408197,0.000003063549,0.0002310231],"genre_scores_gemma":[0.9991761,0.00005207091,0.00006764825,0.00004222103,0.000004734489,0.0000307511,0.0003333309,0.000003143097,0.0002900056],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0372349,"threshold_uncertainty_score":0.07403636,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07565248950629676,"score_gpt":0.4845452747962911,"score_spread":0.4088927852899943,"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."}}