{"id":"W4413232282","doi":"10.36834/cmej.80056","title":"OSCEai dermatology: augmenting dermatologic medical education with Large Language Model GPT-4","year":2025,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Medicine and Dermatology Studies History","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; University of Toronto","funders":"","keywords":"Dermatology; Computer science; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001451141,0.0008338332,0.0002215327,0.0007398461,0.0004027532,0.001967574,0.001237547,0.001356572,0.04448927],"category_scores_gemma":[0.007357682,0.0002614873,0.0008481951,0.0002510709,0.0004970775,0.001976406,0.004404542,0.001459375,0.009632142],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004119254,"about_ca_system_score_gemma":0.0008164074,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004099319,"about_ca_topic_score_gemma":0.001634112,"domain_scores_codex":[0.9991805,0.0003801972,0.00003974093,0.00006472824,0.0002141344,0.0001205978],"domain_scores_gemma":[0.9957585,0.00278885,0.0001775585,0.0002930908,0.0002342896,0.0007476789],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0006091546,0.00164326,0.006729244,0.003062845,0.00007292333,0.003198155,0.007876325,0.002373815,0.02005536,0.006911628,0.2805423,0.666925],"study_design_scores_gemma":[0.0004030297,0.001574676,0.01239921,0.001878679,0.0001315309,0.008754378,0.003803609,0.007891363,0.01176152,0.01286279,0.938237,0.0003022589],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2513073,0.003872738,0.2752346,0.02347289,0.00526281,0.003158112,0.007266977,0.05318287,0.3772416],"genre_scores_gemma":[0.4763722,0.004222237,0.372059,0.009092627,0.001334023,0.002706676,0.004741232,0.004339044,0.1251331],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.04448927,"threshold_uncertainty_score":0.1488314,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.006550674397338951,"score_gpt":0.3008652126604034,"score_spread":0.2943145382630645,"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."}}