{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009493362,0.0001522842,0.0003782721,0.0005379011,0.00006820836,0.00001081119,0.0001867659,0.000254896,0.0004368111],"category_scores_gemma":[0.0002193225,0.000149667,0.00006163237,0.001082705,0.0001604886,0.0001841185,0.00006143528,0.0003748628,0.000377959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001438185,"about_ca_system_score_gemma":0.0004498337,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004112898,"about_ca_topic_score_gemma":0.00002524813,"domain_scores_codex":[0.9980804,0.0001512913,0.0007721756,0.0003693238,0.0001838503,0.0004430151],"domain_scores_gemma":[0.9988545,0.0002569446,0.0001334473,0.0004006981,0.0002092554,0.0001452036],"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":[0.0007527189,0.001817106,0.4397899,0.0007250937,0.00003295473,0.0009966689,0.005463854,0.000756806,0.01467082,0.002368376,0.0001053643,0.5325204],"study_design_scores_gemma":[0.00009410125,0.00422556,0.7937794,0.0003173252,0.00003051255,0.002557591,0.007060629,0.09832621,0.08608796,0.007100855,0.00008100641,0.0003389038],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9906488,0.00005996772,0.00001324436,0.001017621,0.000866191,0.0004054128,0.000003825921,0.00005252822,0.006932373],"genre_scores_gemma":[0.9985287,0.00006945391,0.0005600561,0.00004721707,0.0003369477,0.00008594082,0.00006696526,0.00002303004,0.0002816822],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5321814,"threshold_uncertainty_score":0.6103242,"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."}}