{"id":"W4389577763","doi":"10.2196/48514","title":"Performance of ChatGPT on the Chinese Postgraduate Examination for Clinical Medicine: Survey Study","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical education; Medicine; Family medicine; Traditional Chinese medicine; Traditional medicine; Psychology; Alternative medicine; Pathology","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":[0.008976622,0.0003720027,0.0006405923,0.002116173,0.0007071017,0.0009342809,0.0006811935,0.0006178101,0.001897558],"category_scores_gemma":[0.03924443,0.0002760894,0.000819143,0.001602782,0.0008616277,0.001292888,0.001549083,0.0009597679,0.0006332618],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001763579,"about_ca_system_score_gemma":0.002438102,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01508295,"about_ca_topic_score_gemma":0.01507277,"domain_scores_codex":[0.995401,0.00165113,0.000688343,0.0005491307,0.001113373,0.0005969438],"domain_scores_gemma":[0.9733291,0.01045568,0.004811996,0.001666575,0.005550692,0.004185968],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001671977,0.0004990876,0.973639,0.00010349,0.00004978156,0.0001791419,0.00635233,0.000200918,0.0004621449,0.00006794061,0.0008988806,0.01738004],"study_design_scores_gemma":[0.00001533169,0.0005089053,0.9935431,0.00002397155,0.00002215689,0.0001338455,0.003102149,0.001652602,0.0002927094,0.00003240024,0.0006538004,0.00001899718],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9990774,0.00003953868,0.00008979309,0.00008345764,0.000005428334,0.00008278684,0.0001576938,0.000009036723,0.0004547553],"genre_scores_gemma":[0.9987848,0.00006944827,0.0003417098,0.00006353373,0.000009433167,0.00009480647,0.0003164498,0.000004108332,0.000315689],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01508295,"threshold_uncertainty_score":0.04747349,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2771935890227568,"score_gpt":0.5664957622394303,"score_spread":0.2893021732166735,"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."}}