{"id":"W4408559650","doi":"10.2196/63353","title":"Assessing ChatGPT’s Capability as a New Age Standardized Patient: Qualitative Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Qualitative research; Medicine; Psychology; Gerontology; Computer science; World Wide Web; Sociology; Social science","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001132341,0.0001554304,0.0003331827,0.0002039228,0.000186411,0.00007657927,0.0001078467,0.0001884203,0.001252189],"category_scores_gemma":[0.008607413,0.0001354692,0.00008031239,0.0006277521,0.0001282727,0.0002202031,0.00003328736,0.0004288114,0.00008411198],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005261726,"about_ca_system_score_gemma":0.01663896,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004382593,"about_ca_topic_score_gemma":0.0002454586,"domain_scores_codex":[0.9973932,0.0004335007,0.0007016757,0.0004108092,0.0007962244,0.0002646031],"domain_scores_gemma":[0.9979063,0.0004113371,0.0001417603,0.0004128401,0.0004438507,0.0006839731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002324421,0.003440323,0.005772661,0.0001661412,0.00004720824,0.000007728378,0.3046609,1.707989e-7,0.00005271895,0.001004516,0.02642021,0.658195],"study_design_scores_gemma":[0.0004772734,0.0009725114,0.01905888,0.001000628,0.000126546,0.00001318577,0.9214661,0.00006223196,0.00120008,0.02185445,0.03352502,0.0002431464],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9560364,0.0002439116,0.000556219,0.03301812,0.001952859,0.001748532,9.280158e-7,0.00008500188,0.006358073],"genre_scores_gemma":[0.9909893,0.00001635456,0.00039484,0.005921508,0.0004602334,0.0003621967,0.00005754419,0.00001254946,0.001785523],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6579518,"threshold_uncertainty_score":0.9997435,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1212212797584189,"score_gpt":0.5691855747991221,"score_spread":0.4479642950407032,"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."}}