{"id":"W4406721761","doi":"10.2196/67197","title":"Impact of Attached File Formats on the Performance of ChatGPT-4 on the Japanese National Nursing Examination: Evaluation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Nursing","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Japan Society for the Promotion of Science","keywords":"Nursing; Nurse education; Nursing practice; Computer science; Medical education; Medicine","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":[0.001333164,0.00009988881,0.0001487437,0.0001780627,0.000226571,0.00001316036,0.00009743127,0.00005660403,0.0004428459],"category_scores_gemma":[0.0005302799,0.00005840931,0.00007111453,0.0004737924,0.0001126514,0.00009767499,0.00000483694,0.0001827046,0.00001084269],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006429709,"about_ca_system_score_gemma":0.0003803895,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004497528,"about_ca_topic_score_gemma":0.000002547742,"domain_scores_codex":[0.9984992,0.0002010639,0.0003735306,0.0001354109,0.0006490103,0.000141758],"domain_scores_gemma":[0.9981179,0.0007444299,0.0001918709,0.0002610309,0.0006589766,0.00002577545],"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":[0.001254481,0.006212836,0.04097005,0.0001911599,0.0001737324,3.7374e-7,0.1696185,0.002096256,0.001956539,0.002239189,0.02338801,0.7518988],"study_design_scores_gemma":[0.0001583728,0.001890935,0.912566,0.002521088,0.00007210037,0.000003253688,0.02764399,0.04559705,0.008872299,0.0005873864,0.00001515089,0.00007233868],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9846641,0.00003513802,0.00000566956,0.003677404,0.0003113915,0.001515771,0.000007657683,0.00001156963,0.009771283],"genre_scores_gemma":[0.9993858,0.000002706741,0.000008721659,0.0001366778,0.000128095,0.0001459128,0.00003762665,0.000007149628,0.0001472841],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.871596,"threshold_uncertainty_score":0.4848852,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2172577786211252,"score_gpt":0.5266515622792063,"score_spread":0.3093937836580811,"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."}}