{"id":"W4385236544","doi":"10.2196/48433","title":"Examining Real-World Medication Consultations and Drug-Herb Interactions: ChatGPT Performance Evaluation","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":46,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"China Medical University; China Medical University Hospital","keywords":"Pharmacist; Medicine; Pharmacy; Family medicine; Set (abstract data type); Computer science","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":[0.02595504,0.0009651465,0.001047566,0.002961415,0.0005792734,0.001030296,0.0009232105,0.0008480307,0.001217204],"category_scores_gemma":[0.09379699,0.0003045773,0.001140821,0.001811877,0.0006833868,0.0009938296,0.002125534,0.000567104,0.0005458614],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009373932,"about_ca_system_score_gemma":0.001164545,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002470885,"about_ca_topic_score_gemma":0.003879128,"domain_scores_codex":[0.9690613,0.02080267,0.003482949,0.002099839,0.00400142,0.0005518995],"domain_scores_gemma":[0.8283029,0.1282744,0.01696223,0.006780129,0.01644629,0.003234169],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003836129,0.001004576,0.8232138,0.00130894,0.0005651482,0.0006132295,0.008238671,0.00308075,0.005902342,0.0001170036,0.001622297,0.1504971],"study_design_scores_gemma":[0.0002766511,0.005549224,0.9431522,0.0001683129,0.0003910369,0.001296451,0.003283969,0.03366619,0.009387684,0.0002639956,0.002434106,0.0001302711],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9875171,0.0002715217,0.00803548,0.0001194959,0.00004610455,0.001447554,0.0007452424,0.0003962027,0.00142123],"genre_scores_gemma":[0.9770451,0.0001879356,0.01958721,0.0001136596,0.00006625754,0.001457599,0.0009777496,0.00004305194,0.000521407],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02595504,"threshold_uncertainty_score":0.137265,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1717367803622567,"score_gpt":0.5053216186434543,"score_spread":0.3335848382811976,"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."}}