{"id":"W4367175039","doi":"10.1001/jamaophthalmol.2023.1144","title":"Performance of an Artificial Intelligence Chatbot in Ophthalmic Knowledge Assessment","year":2023,"lang":"en","type":"article","venue":"JAMA Ophthalmology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":311,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Michael's Hospital; University of Toronto; Western University","funders":"","keywords":"Medicine; Chatbot; Certification; Multiple choice; Board certification; Medical education; Curriculum; Family medicine; Significant difference; Artificial intelligence; Psychology; Internal medicine; Computer science; Continuing medical education; Pedagogy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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":[0.02741367,0.0006899037,0.0007223045,0.002476972,0.000923522,0.002188303,0.001395484,0.001625929,0.003807969],"category_scores_gemma":[0.130786,0.0004692703,0.0007441872,0.001412273,0.000911814,0.002090376,0.003216096,0.001080947,0.002186378],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002543994,"about_ca_system_score_gemma":0.00295837,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004143122,"about_ca_topic_score_gemma":0.004736081,"domain_scores_codex":[0.9731973,0.01628651,0.002773913,0.0026385,0.003680646,0.001423023],"domain_scores_gemma":[0.8292708,0.1220218,0.01964117,0.006444492,0.01415638,0.008465407],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002482375,0.003198399,0.7885876,0.0006987971,0.0001739151,0.0006623268,0.01354663,0.002916289,0.003032881,0.0002570232,0.005410628,0.1790331],"study_design_scores_gemma":[0.0002780041,0.007238096,0.9282871,0.0005159396,0.0001451367,0.000844127,0.009282025,0.03750546,0.006149137,0.0007272292,0.008832403,0.0001953743],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.992504,0.000136567,0.001991599,0.0004564077,0.00004911624,0.0005845782,0.0004615039,0.0003963633,0.00341985],"genre_scores_gemma":[0.9901045,0.0001001129,0.007267378,0.0002595723,0.00004278462,0.0004290988,0.0005799999,0.00003657441,0.001180057],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02741367,"threshold_uncertainty_score":0.1449791,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2395228869931877,"score_gpt":0.4914119746102475,"score_spread":0.2518890876170599,"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."}}