{"id":"W4410659767","doi":"10.2196/68409","title":"Clinical Performance and Communication Skills of ChatGPT Versus Physicians in Emergency Medicine: Simulated Patient Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Medicine; MEDLINE; Medical emergency; Medical physics; Computer science; World Wide Web","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":[0.00086913,0.0001125678,0.00039963,0.0001942703,0.00006288549,0.000002804428,0.0001400794,0.0001789786,0.0001472179],"category_scores_gemma":[0.0009751797,0.00009098586,0.00003887505,0.0005614725,0.0002395615,0.0001340916,0.00008221788,0.0004683926,0.000009516097],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005429499,"about_ca_system_score_gemma":0.0002045534,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001796215,"about_ca_topic_score_gemma":0.00009326702,"domain_scores_codex":[0.9971833,0.00009512806,0.001993424,0.00009616108,0.0004509102,0.0001810588],"domain_scores_gemma":[0.9984823,0.0004428058,0.0003103469,0.0003907545,0.0002453637,0.0001284557],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003363296,0.001925179,0.4390661,0.0005742644,0.0000581383,0.000001100498,0.04962563,0.00002127066,0.000001963047,0.0001253396,0.002171568,0.5060931],"study_design_scores_gemma":[0.003509487,0.007725049,0.720873,0.004775909,0.0002180112,0.000004081299,0.1565392,0.09882847,0.0002534478,0.000543232,0.006364838,0.0003652737],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.99351,0.0001486462,0.00004723644,0.001278792,0.0007166473,0.0008738391,8.603456e-7,0.00002362452,0.00340036],"genre_scores_gemma":[0.9972178,0.002009979,0.0001285382,0.0004410101,0.00006166863,0.00003178806,0.00002291974,0.000005461888,0.00008087428],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5057278,"threshold_uncertainty_score":0.3710296,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08406932550693881,"score_gpt":0.4844262238204639,"score_spread":0.4003568983135251,"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."}}