{"id":"W4417199531","doi":"10.36834/cmej.80873","title":"AI that teaches: an evidence-based GPT model to improve medical student understanding of pulmonary function tests","year":2025,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Function (biology); Pulmonary medicine; Statement (logic); MEDLINE; Test (biology)","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":[0.006802239,0.0009432412,0.0004425947,0.002319364,0.0005028455,0.00279411,0.001948219,0.001690217,0.0105137],"category_scores_gemma":[0.03145764,0.0004968946,0.001678729,0.001084452,0.001230933,0.003112102,0.003019416,0.001991261,0.002030456],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001194707,"about_ca_system_score_gemma":0.003268285,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003225761,"about_ca_topic_score_gemma":0.00486259,"domain_scores_codex":[0.9961294,0.002152086,0.0002815622,0.0005498056,0.0007804725,0.0001066556],"domain_scores_gemma":[0.9796764,0.01592225,0.0008808354,0.00154512,0.001608005,0.0003674243],"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":[0.0009710981,0.001091964,0.02208594,0.003066152,0.0003565775,0.001210606,0.009277208,0.1008595,0.01035341,0.08342953,0.01053079,0.7567672],"study_design_scores_gemma":[0.0003642644,0.00117366,0.008393828,0.001867412,0.0006929798,0.001459958,0.00193052,0.7009162,0.01371636,0.1800091,0.08932066,0.0001550652],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03128583,0.0009136489,0.9360719,0.005270071,0.0001422044,0.001335985,0.001159564,0.003708585,0.02011232],"genre_scores_gemma":[0.3176084,0.000718885,0.6722467,0.0008640967,0.00006082086,0.001272997,0.001285855,0.0002331534,0.005709157],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0105137,"threshold_uncertainty_score":0.03597414,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07498799951974351,"score_gpt":0.4156260024806485,"score_spread":0.340638002960905,"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."}}