{"id":"W4408068147","doi":"10.1007/s00467-025-06729-x","title":"Evaluating large language models in pediatric nephrology","year":2025,"lang":"en","type":"editorial","venue":"Pediatric Nephrology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"Children's Hospital of Western Ontario; Western University","funders":"","keywords":"Nephrology; Medicine; Internal medicine; Intensive care medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01776961,0.001996142,0.002629101,0.002624727,0.001433214,0.005253579,0.002995424,0.01026477,0.01075608],"category_scores_gemma":[0.08983666,0.0009946399,0.00224631,0.000973573,0.00188624,0.003194707,0.00107695,0.01172368,0.004017265],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003258196,"about_ca_system_score_gemma":0.004648085,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005738433,"about_ca_topic_score_gemma":0.01280037,"domain_scores_codex":[0.9922385,0.003420973,0.0009759523,0.0005725467,0.002576206,0.0002158568],"domain_scores_gemma":[0.8932489,0.08543734,0.001593337,0.000981015,0.01550636,0.003233103],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001292605,0.00002059907,0.00008066137,0.0004506477,0.0001122428,0.0001147151,0.00001447234,0.0002250567,0.00005584128,0.0003574295,0.9862634,0.01217566],"study_design_scores_gemma":[0.0009712389,0.0002259908,0.001474747,0.003648218,0.001186893,0.00061306,0.0001540253,0.007471749,0.0007158708,0.01136424,0.972056,0.0001179283],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"editorial","genre_scores_codex":[0.0003202718,0.01700045,0.001681044,0.103979,0.8739846,0.00007052704,0.0004560908,0.0002941471,0.002214025],"genre_scores_gemma":[0.004460603,0.01382537,0.002254031,0.04068325,0.9275229,0.0001241575,0.0003994216,0.0002700681,0.0104602],"genre_candidate":"editorial","genre_consensus":"editorial","teacher_disagreement_score":0.01776961,"threshold_uncertainty_score":0.09397578,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07570928681527708,"score_gpt":0.440631592931464,"score_spread":0.3649223061161869,"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."}}