{"id":"W4402675712","doi":"10.3138/cjgim.2024.0011","title":"Direct comparison of GPT-4 and human physicians in MKSAP-19 multiple-choice questions","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of General Internal Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University; Institute for Clinical Evaluative Sciences; Ottawa Hospital; University of Ottawa","funders":"","keywords":"Medicine; Family medicine; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.02345895,0.0005199559,0.0007559104,0.00347674,0.000328551,0.001410017,0.0007844043,0.001104941,0.002834777],"category_scores_gemma":[0.1245749,0.0002947039,0.0007354117,0.002151334,0.001092968,0.001536481,0.002050153,0.0008452921,0.0007953206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001152828,"about_ca_system_score_gemma":0.0005941216,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001429612,"about_ca_topic_score_gemma":0.001723978,"domain_scores_codex":[0.9694511,0.01563177,0.002722433,0.002635559,0.008722392,0.0008368478],"domain_scores_gemma":[0.8768556,0.09193153,0.01445448,0.003114491,0.0118985,0.001745396],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002772964,0.000533918,0.9244393,0.0002981657,0.000540839,0.0001289156,0.004783439,0.002352279,0.001864451,0.0005253307,0.002574734,0.05918571],"study_design_scores_gemma":[0.0002494688,0.001941658,0.9829995,0.00008364802,0.0001214202,0.0003480063,0.001503335,0.005527669,0.002731821,0.000678038,0.003770689,0.00004473071],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9914532,0.0002137201,0.003166,0.0001907747,0.00004245407,0.0003473446,0.000783952,0.00005830953,0.003744269],"genre_scores_gemma":[0.9945544,0.00008768994,0.003330197,0.000140962,0.00003217625,0.0004130799,0.0007971529,0.00002356697,0.0006209068],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02345895,"threshold_uncertainty_score":0.1240643,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1406056818086464,"score_gpt":0.4564786048824369,"score_spread":0.3158729230737904,"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."}}