{"id":"W7136785886","doi":"10.4103/apc.apc_301_25","title":"Comparing closed and open large language models on pediatric cardiology board exam performance","year":2025,"lang":"en","type":"article","venue":"Annals of Pediatric Cardiology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hospital for Sick Children","funders":"","keywords":"Subspecialty; Clinical cardiology; Editorial board; Pediatric Radiology; MEDLINE; Clinical Practice","routes":{"ca_aff":true,"ca_fund":false,"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":[],"consensus_categories":[],"category_scores_codex":[0.0009171467,0.0001838046,0.0009052008,0.0004183172,0.0001307159,0.00001386492,0.0002314489,0.0002601196,0.00001175095],"category_scores_gemma":[0.0002508509,0.0001704553,0.0001349691,0.0004468022,0.00006988725,0.0001164556,0.0002084703,0.0003282246,0.00002305655],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003107752,"about_ca_system_score_gemma":0.0002982084,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003747978,"about_ca_topic_score_gemma":0.0000194245,"domain_scores_codex":[0.9982147,0.0001943691,0.0005647559,0.0004231489,0.0001345312,0.0004684485],"domain_scores_gemma":[0.9987087,0.0003189264,0.0001575304,0.0004287349,0.0002567508,0.0001293618],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000390254,0.0000770979,0.9679133,0.0004204382,0.0001254908,0.00003095305,0.0008729892,0.0009179635,0.00003981301,0.00164633,0.01799694,0.009568479],"study_design_scores_gemma":[0.0007611469,0.001554958,0.9854804,0.00005325728,0.0006966256,0.00006359611,0.002335029,0.002629753,0.001005781,0.002271893,0.002710972,0.0004365401],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9798288,0.00427023,0.0003511579,0.001107183,0.0007047861,0.0005654934,0.00001492108,0.00003649818,0.01312088],"genre_scores_gemma":[0.9827749,0.01383144,0.0000756938,0.001134152,0.001805852,0.00005307587,0.00003789496,0.00001446723,0.0002725664],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01756719,"threshold_uncertainty_score":0.6950966,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2416587721565181,"score_gpt":0.4488845448533944,"score_spread":0.2072257726968763,"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."}}