{"id":"W4410477368","doi":"10.1542/pedsos.2025-000485","title":"PedMedQA: Comparing Large Language Model Accuracy in Pediatric and Adult Medicine","year":2025,"lang":"en","type":"article","venue":"Pediatrics Open Science","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mila - Quebec Artificial Intelligence Institute; Montreal Children's Hospital; McGill University; Polytechnique Montréal; Centre for Advancing Health Outcomes; McGill University Health Centre","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research","keywords":"Computer science; Natural language processing","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["open_science"],"consensus_categories":[],"category_scores_codex":[0.003625696,0.0001971915,0.0003404827,0.001065233,0.000475809,0.0005329618,0.005650669,0.0000668422,0.000005903301],"category_scores_gemma":[0.00463726,0.0001770489,0.00001977809,0.005620003,0.0001582028,0.001699737,0.005048186,0.0005288293,0.00001253255],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001312739,"about_ca_system_score_gemma":0.0008815489,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00127123,"about_ca_topic_score_gemma":0.0004714965,"domain_scores_codex":[0.9969587,0.0001101723,0.0004955388,0.001002757,0.0007131618,0.0007196917],"domain_scores_gemma":[0.9977807,0.0005768114,0.0001958367,0.0009219915,0.0002868785,0.0002377585],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000006766766,0.00005874604,0.8696132,0.0001793857,7.985045e-7,0.00002111555,0.004820509,0.001830163,0.00002411637,0.1163751,0.0009977988,0.00607225],"study_design_scores_gemma":[0.0008448794,0.00003470292,0.1587258,0.00002600961,0.000008724847,0.000005248476,0.0002366688,0.8360909,0.000009072294,0.00369852,0.0001016672,0.0002178259],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7412149,0.002951512,0.2075566,0.01461941,0.000888861,0.001230778,0.000005546081,0.0002367752,0.03129561],"genre_scores_gemma":[0.9735702,0.0005486489,0.0240856,0.00133271,0.0001224687,0.00003102915,0.000001913282,0.000007211669,0.0003002246],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8342608,"threshold_uncertainty_score":0.9997292,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02644698040996564,"score_gpt":0.3821723891951043,"score_spread":0.3557254087851387,"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."}}