{"id":"W4412426567","doi":"10.2196/74085","title":"ChatGPT-4’s Level of Dermatological Knowledge Based on Board Examination Review Questions and Bloom’s Taxonomy","year":2025,"lang":"en","type":"article","venue":"JMIR Dermatology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Medicine; Dermatology; Computer science; World Wide Web","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02089212,0.0006243027,0.0006421941,0.004218644,0.000606769,0.00182553,0.001527924,0.0007927948,0.01233288],"category_scores_gemma":[0.1656422,0.0002886081,0.0007791942,0.001727369,0.0007990638,0.002489941,0.003115725,0.001313007,0.002394264],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002269171,"about_ca_system_score_gemma":0.003042056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003806465,"about_ca_topic_score_gemma":0.005566445,"domain_scores_codex":[0.9843013,0.009059406,0.002389479,0.0008327708,0.002785143,0.0006319848],"domain_scores_gemma":[0.7029498,0.2325955,0.0232533,0.008783315,0.02695002,0.005468139],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001614371,0.0006806119,0.7069511,0.001768921,0.0001316095,0.0004025891,0.007628037,0.002292828,0.002727004,0.001121967,0.02676572,0.2479153],"study_design_scores_gemma":[0.0002407859,0.002280256,0.9041427,0.002019288,0.0002783277,0.002597202,0.007630685,0.02439145,0.009957023,0.004876676,0.04134837,0.0002372331],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9296137,0.0011104,0.02591523,0.002729285,0.0003101068,0.002851167,0.008362015,0.003164975,0.0259432],"genre_scores_gemma":[0.953718,0.000337283,0.03641384,0.0004110393,0.00009314118,0.002070728,0.003239446,0.0001600571,0.003556478],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9791079,"threshold_uncertainty_score":0.1104894,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.276929403774977,"score_gpt":0.4607815919129726,"score_spread":0.1838521881379956,"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."}}