{"id":"W7117292103","doi":"10.1002/ksa.70255","title":"ChatGPT models provide higher‐quality but lower‐readability responses than Google Gemini regarding anterior shoulder instability, with no added benefit of the orthopaedic expert plugin","year":2025,"lang":"en","type":"article","venue":"Knee Surgery Sports Traumatology Arthroscopy","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; McGill University; McGill University Health Centre","funders":"","keywords":"Orthopedic surgery; Plug-in; MEDLINE; Orthopedic Procedures; Shoulder 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003494557,0.001302837,0.0009125076,0.001540965,0.0001892224,0.001291871,0.0009133536,0.0009853964,0.01127677],"category_scores_gemma":[0.02326914,0.0002737736,0.001174161,0.0008879907,0.000382447,0.001336943,0.00161068,0.0004386186,0.004094773],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005669708,"about_ca_system_score_gemma":0.0006388442,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002563681,"about_ca_topic_score_gemma":0.005595332,"domain_scores_codex":[0.9975061,0.0009254704,0.0003125143,0.000414702,0.0007099418,0.0001313844],"domain_scores_gemma":[0.9752007,0.01627674,0.002255484,0.002560733,0.002850371,0.0008558687],"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":[0.01970427,0.00152308,0.3398087,0.004623488,0.001022145,0.001209984,0.004102976,0.009825833,0.03305061,0.0007469638,0.06476156,0.5196204],"study_design_scores_gemma":[0.001559622,0.007238592,0.8496478,0.0009100267,0.001482352,0.003311623,0.002289052,0.0494188,0.03093802,0.002342444,0.05027197,0.0005896824],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9278814,0.0008299974,0.02470975,0.0005327901,0.0002246318,0.001613177,0.01293868,0.01677595,0.01449365],"genre_scores_gemma":[0.9440358,0.0004938827,0.02782814,0.0004368387,0.000143944,0.001454543,0.01416157,0.001181347,0.01026392],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01127677,"threshold_uncertainty_score":0.03772449,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09287484815574842,"score_gpt":0.3791135514729462,"score_spread":0.2862387033171978,"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."}}