{"id":"W4399915777","doi":"10.1016/j.arthro.2024.06.017","title":"ChatGPT Provides Satisfactory but Occasionally Inaccurate Answers to Common Patient Hip Arthroscopy Questions","year":2024,"lang":"en","type":"article","venue":"Arthroscopy The Journal of Arthroscopic and Related Surgery","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":27,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia Hospital; University of British Columbia","funders":"","keywords":"Hip arthroscopy; Medicine; Arthroscopy; Physical therapy; Medical physics; Psychology; 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":[0.017956,0.001025148,0.0009580831,0.002084515,0.0006363255,0.00123889,0.0008610483,0.001407397,0.006408298],"category_scores_gemma":[0.1161275,0.0003694544,0.001075448,0.001287821,0.0007937219,0.001281235,0.003078296,0.0008007907,0.002963256],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008908584,"about_ca_system_score_gemma":0.001366577,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001369725,"about_ca_topic_score_gemma":0.002934561,"domain_scores_codex":[0.9772879,0.01362642,0.002594276,0.001502868,0.004061257,0.0009271833],"domain_scores_gemma":[0.8211233,0.1331371,0.01897786,0.006582757,0.01751009,0.00266888],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003529309,0.0007756036,0.6799686,0.003129495,0.0003138465,0.001568474,0.01443749,0.001434937,0.00921422,0.0003464722,0.02575704,0.2595245],"study_design_scores_gemma":[0.0005285505,0.004141468,0.9073923,0.002175331,0.000474835,0.01152518,0.01193658,0.01084935,0.01320171,0.001218576,0.0362107,0.0003453097],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9684918,0.001366633,0.01211775,0.001564252,0.0003561242,0.001073178,0.001812702,0.001251909,0.01196554],"genre_scores_gemma":[0.9691766,0.0008306515,0.02305463,0.001219598,0.0002243618,0.001298164,0.001454559,0.0002178159,0.002523624],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.017956,"threshold_uncertainty_score":0.09496158,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04986549109542006,"score_gpt":0.3608093150574811,"score_spread":0.3109438239620611,"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."}}