{"id":"W4401632453","doi":"10.1016/j.acra.2024.08.009","title":"Evaluating Artificial Intelligence Competency in Education: Performance of ChatGPT-4 in the American Registry of Radiologic Technologists (ARRT) Radiography Certification Exam","year":2024,"lang":"en","type":"article","venue":"Academic Radiology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Juravinski Hospital; Western University; University of British Columbia; McMaster University Medical Centre; Hamilton Health Sciences; University of Toronto; Trillium Health Centre","funders":"","keywords":"Certification; Medical physics; Radiography; Medicine; Medical education; Radiology; Management","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":[0.007306498,0.0007367051,0.0006076278,0.001331817,0.0003693758,0.001277628,0.001206155,0.001101414,0.002916668],"category_scores_gemma":[0.036537,0.0002227217,0.0006682217,0.000625268,0.0004613575,0.001181001,0.002657339,0.001067907,0.001535197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001259719,"about_ca_system_score_gemma":0.001392446,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005158355,"about_ca_topic_score_gemma":0.007025128,"domain_scores_codex":[0.9948384,0.00264981,0.0004337321,0.0009206639,0.0008022706,0.0003551333],"domain_scores_gemma":[0.9714715,0.01606024,0.003099888,0.001575491,0.004471435,0.003321424],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003075012,0.003547362,0.7111188,0.0006123717,0.000300323,0.0002793187,0.002462999,0.009907077,0.004636652,0.0002702746,0.01196143,0.2518284],"study_design_scores_gemma":[0.0003710552,0.00536759,0.8727978,0.0001820026,0.000174328,0.0005094879,0.001688205,0.1001729,0.01007124,0.000721536,0.0078141,0.0001297224],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.994176,0.0001205394,0.002054139,0.0002340529,0.00004887244,0.0003414087,0.0008467258,0.0005220684,0.001656082],"genre_scores_gemma":[0.9853258,0.00008871419,0.009186435,0.0001931285,0.00004134905,0.0004214699,0.003303136,0.00003851987,0.001401482],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007306498,"threshold_uncertainty_score":0.03864092,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.189072774425262,"score_gpt":0.4709775486186016,"score_spread":0.2819047741933396,"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."}}