{"id":"W4392005596","doi":"10.1016/j.urology.2023.11.042","title":"Accuracy and Readability of Kidney Stone Patient Information Materials Generated by a Large Language Model Compared to Official Urologic Organizations","year":2024,"lang":"en","type":"letter","venue":"Urology","topic":"Health Literacy and Information Accessibility","field":"Health Professions","cited_by":18,"is_retracted":false,"has_abstract":false,"ca_institutions":"Vancouver General Hospital; University of British Columbia","funders":"","keywords":"Medicine; Readability; Medical physics; Radiology; Linguistics","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":[0.01057371,0.0001597959,0.0002569626,0.001733952,0.0007827914,0.002006175,0.0004578243,0.00115317,0.007997832],"category_scores_gemma":[0.1092539,0.0002140016,0.000588441,0.001246791,0.0005543618,0.00130386,0.0009977088,0.001066375,0.002222667],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001949044,"about_ca_system_score_gemma":0.001529654,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03460583,"about_ca_topic_score_gemma":0.02848953,"domain_scores_codex":[0.9920217,0.003445106,0.001355622,0.0005699102,0.002156813,0.0004509266],"domain_scores_gemma":[0.8168058,0.1341175,0.0124103,0.004158308,0.03066379,0.001844203],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002308988,0.0007896368,0.8756241,0.0001952394,0.00009324746,0.0008157086,0.01285085,0.0005622469,0.001800464,0.0006418028,0.036644,0.0676738],"study_design_scores_gemma":[0.000110416,0.0007184693,0.9691154,0.0003041137,0.0001009646,0.0006479554,0.01038632,0.002616897,0.002601484,0.0004272751,0.01289294,0.00007776359],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9618021,0.00021276,0.0005068076,0.01476157,0.0002946054,0.00006253456,0.002782277,0.0000833902,0.01949401],"genre_scores_gemma":[0.9872423,0.0001588248,0.0006652411,0.003226506,0.0001218652,0.00004982017,0.001117208,0.00007001503,0.00734836],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03460583,"threshold_uncertainty_score":0.06880879,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02282794518583499,"score_gpt":0.3693426068630186,"score_spread":0.3465146616771836,"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."}}