{"id":"W4400079977","doi":"10.2196/59133","title":"Educational Utility of Clinical Vignettes Generated in Japanese by ChatGPT-4: Mixed Methods Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Thematic analysis; Terminology; Medical education; Family medicine; Psychology; Medicine; Quality (philosophy); Qualitative research","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003949249,0.0001541136,0.0004331827,0.0002479216,0.00004515942,0.00002223117,0.0001455414,0.0003045844,0.003305974],"category_scores_gemma":[0.005096321,0.0001317112,0.0001047438,0.0008644277,0.0001784851,0.0001237067,0.00002831736,0.0006039097,0.00007619627],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001006177,"about_ca_system_score_gemma":0.006270018,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001754296,"about_ca_topic_score_gemma":0.0001561751,"domain_scores_codex":[0.9964046,0.0007783118,0.001480232,0.0004956893,0.0006004134,0.0002407049],"domain_scores_gemma":[0.9975594,0.001188351,0.0001331762,0.000353444,0.0003336056,0.000432036],"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":[0.0001081232,0.007963673,0.1310681,0.0003492207,0.00004393521,0.000002484874,0.01028825,1.690599e-7,0.0003705327,0.0001556111,0.07960136,0.7700485],"study_design_scores_gemma":[0.0004162106,0.001346484,0.8912073,0.001364888,0.0001711142,0.00005594322,0.04455879,0.01687917,0.004115392,0.003184476,0.03624463,0.0004556294],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9695988,0.002141882,0.0001038651,0.02169419,0.00468419,0.00102276,0.000003839598,0.00004683927,0.0007035958],"genre_scores_gemma":[0.9949433,0.0001249845,0.0009810528,0.0009749003,0.001350079,0.0004602115,0.0002243708,0.00002023979,0.000920874],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7695929,"threshold_uncertainty_score":0.9993635,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.173919822095302,"score_gpt":0.6064311012383689,"score_spread":0.4325112791430668,"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."}}