{"id":"W4404517343","doi":"10.1139/cjpp-2024-0026","title":"Evaluating the concordance of ChatGPT and physician recommendations for bariatric surgery","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of Physiology and Pharmacology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Concordance; Medicine; Accreditation; Sleeve gastrectomy; Surgery; Medical physics; Gastric bypass; Medical education; Weight loss; Obesity; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03226524,0.0003666324,0.0005654624,0.001915179,0.0005861011,0.001709878,0.0008106314,0.001004415,0.002356827],"category_scores_gemma":[0.2342042,0.0002714187,0.0005155656,0.0007001077,0.001311819,0.001470996,0.002009368,0.001084587,0.0007866331],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001250425,"about_ca_system_score_gemma":0.001425737,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003626097,"about_ca_topic_score_gemma":0.004622593,"domain_scores_codex":[0.9608461,0.02495084,0.003899032,0.002451294,0.007046338,0.0008064475],"domain_scores_gemma":[0.696002,0.2407317,0.02586522,0.01052712,0.02236145,0.004512508],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001172003,0.0005374895,0.8527564,0.0004382951,0.0002371274,0.0006512977,0.03034399,0.003742913,0.003835088,0.0006732322,0.002459449,0.1031527],"study_design_scores_gemma":[0.000137158,0.002308693,0.9176426,0.0004785221,0.0001334239,0.001933538,0.02388318,0.03636059,0.0056636,0.0023085,0.0089402,0.0002099223],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.983713,0.0001911704,0.00790232,0.0007446761,0.00006774596,0.0002312657,0.0002528192,0.0001727778,0.006724163],"genre_scores_gemma":[0.9925499,0.000120784,0.006041428,0.0002629793,0.00003226771,0.0001068442,0.0003034393,0.0000243439,0.000557972],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03226524,"threshold_uncertainty_score":0.1706369,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2352769277264199,"score_gpt":0.4925647393862427,"score_spread":0.2572878116598228,"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."}}