{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007483715,0.00006858743,0.00008651554,0.0002840027,0.0000923339,0.00003163913,0.00005837659,0.00008435768,0.0001274439],"category_scores_gemma":[0.0003750135,0.00004744605,0.00001717982,0.0006111047,0.00012399,0.00007971167,0.000004851756,0.0003299159,0.00001137243],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001043085,"about_ca_system_score_gemma":0.000666261,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007025072,"about_ca_topic_score_gemma":0.00004147608,"domain_scores_codex":[0.9988934,0.00008327675,0.0003157517,0.0001664574,0.0004316409,0.0001094859],"domain_scores_gemma":[0.9993926,0.0002483034,0.00003746133,0.0001256687,0.0001162537,0.00007968152],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006586398,0.00230426,0.8116285,0.0002427653,0.000009488115,0.000001907923,0.05807485,0.0000125444,0.00001256994,0.0007499548,0.0009213772,0.1259759],"study_design_scores_gemma":[0.00004500307,0.0005182972,0.9771388,0.0005356979,0.00001028085,0.00002615901,0.01258799,0.008324975,0.00005982084,0.0002137665,0.0004989683,0.00004020548],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9904374,0.0001799321,0.000005913873,0.007656509,0.0007198,0.000575244,0.000001311429,0.00002112874,0.0004026924],"genre_scores_gemma":[0.9986027,0.0001648511,0.00001251897,0.0004539767,0.0003848033,0.0001500754,0.00009779423,0.00000735257,0.0001259647],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1655103,"threshold_uncertainty_score":0.1934794,"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."}}