{"id":"W4415656798","doi":"10.1080/10872981.2025.2574647","title":"Large language models for generating key-feature questions in medical education","year":2025,"lang":"en","type":"article","venue":"Medical Education Online","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Uniwersytet Jagielloński Collegium Medicum","keywords":"Checklist; Metric (unit); Workflow; Quality (philosophy); Descriptive statistics; Inclusion (mineral); Minor (academic); MEDLINE","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.0307798,0.001193699,0.0006074949,0.002232664,0.0007507166,0.003299487,0.002356285,0.001389187,0.004918812],"category_scores_gemma":[0.15384,0.0007400111,0.001433224,0.001530424,0.0009627761,0.004851852,0.003469098,0.001691607,0.001867281],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003286398,"about_ca_system_score_gemma":0.003831952,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006128113,"about_ca_topic_score_gemma":0.006983013,"domain_scores_codex":[0.9685404,0.02442265,0.001819066,0.002171992,0.002629046,0.0004168996],"domain_scores_gemma":[0.767279,0.2094382,0.00656809,0.009011774,0.006826451,0.0008765642],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003585642,0.002877051,0.06002875,0.004415945,0.0003632709,0.001118554,0.04383834,0.178856,0.01433743,0.03439189,0.01312956,0.6430576],"study_design_scores_gemma":[0.0009397782,0.002145589,0.01503009,0.001878244,0.0003649781,0.0008121763,0.008471086,0.8284761,0.01713304,0.06688245,0.05745662,0.0004097903],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1713056,0.0004273551,0.8057282,0.001130347,0.0001047232,0.005002369,0.002371782,0.007639211,0.006290365],"genre_scores_gemma":[0.4623322,0.0002008073,0.5282787,0.000359237,0.00002892871,0.003949864,0.003037032,0.0003917902,0.001421399],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0307798,"threshold_uncertainty_score":0.1627811,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0558199528456924,"score_gpt":0.490967376924955,"score_spread":0.4351474240792626,"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."}}