{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0168631,0.0005446064,0.0007833201,0.001355225,0.0008715641,0.001544265,0.0009291353,0.0007591838,0.00325801],"category_scores_gemma":[0.118384,0.0002363509,0.001385141,0.001066606,0.0009190825,0.001907592,0.00175718,0.0008393759,0.0007501724],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001612597,"about_ca_system_score_gemma":0.002070006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004940849,"about_ca_topic_score_gemma":0.006228593,"domain_scores_codex":[0.9826331,0.008360824,0.002601023,0.000949116,0.004655226,0.000800639],"domain_scores_gemma":[0.9029067,0.04641293,0.01919493,0.004524504,0.0218083,0.005152642],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01229352,0.004337254,0.831668,0.0008976135,0.0004268029,0.0004045434,0.004247659,0.0005085317,0.001551991,0.0001401667,0.002323101,0.1412008],"study_design_scores_gemma":[0.0003299733,0.01606822,0.9693522,0.0005081255,0.0008290319,0.0005492083,0.004256947,0.002225673,0.003478953,0.0001162643,0.002169364,0.0001161409],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9968506,0.0003690995,0.0004892234,0.0001680794,0.00008318994,0.0002944921,0.00027245,0.00003212373,0.001440704],"genre_scores_gemma":[0.9962661,0.0002641871,0.001453747,0.0001542813,0.00007769922,0.0004622045,0.0004504604,0.00003830547,0.0008329854],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0168631,"threshold_uncertainty_score":0.08918172,"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."}}