{"id":"W4396732548","doi":"10.1016/j.soard.2024.04.014","title":"Performance of artificial intelligence in bariatric surgery: comparative analysis of ChatGPT-4, Bing, and Bard in the American Society for Metabolic and Bariatric Surgery textbook of bariatric surgery questions","year":2024,"lang":"en","type":"review","venue":"Surgery for Obesity and Related Diseases","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":38,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; McMaster University","funders":"","keywords":"Medicine; Surgery; General surgery","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.003526039,0.0004839987,0.004879044,0.003605052,0.0001791187,0.0000360179,0.0001053382,0.0003550437,0.00001380147],"category_scores_gemma":[0.001377985,0.0003566306,0.002218131,0.006712888,0.0006030732,0.0001674821,0.00005333019,0.000414161,9.553227e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000746334,"about_ca_system_score_gemma":0.001803932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00135946,"about_ca_topic_score_gemma":0.0001374013,"domain_scores_codex":[0.9951629,0.0004982262,0.002852634,0.0006776001,0.0003375272,0.0004711266],"domain_scores_gemma":[0.9782696,0.01937525,0.001453757,0.0003720599,0.0003334458,0.0001958367],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0005097258,0.001100555,0.3350572,0.04600973,0.003026182,0.00000850091,0.002044149,0.00007575013,0.000001252383,0.0005647438,0.0009707242,0.6106315],"study_design_scores_gemma":[0.0002218723,0.0008496937,0.6794824,0.04380065,0.194338,0.0001325021,0.01198613,0.04024161,0.0001308727,0.004773241,0.02025624,0.003786731],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.3988554,0.5992817,0.00003902371,0.00008498311,0.000273119,0.001105351,0.0003429633,0.00001470951,0.00000277009],"genre_scores_gemma":[0.4148792,0.584445,0.00002982213,0.00001516274,0.00008110797,0.0001728495,0.0003496366,0.00001929518,0.000007960886],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6068448,"threshold_uncertainty_score":0.9998885,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1251716679673075,"score_gpt":0.4009056248275413,"score_spread":0.2757339568602338,"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."}}