{"id":"W4401424596","doi":"10.2196/56426","title":"Assessing ChatGPT as a Medical Consultation Assistant for Chronic Hepatitis B: Cross-Language Study of English and Chinese","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Consistency (knowledge bases); Chronic hepatitis; China; Health care; Medical education; Medicine; Variance (accounting); Psychology; Family medicine; Computer science; Artificial intelligence; Political 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":[0.007676288,0.0004651332,0.000579651,0.001154999,0.002082,0.001451673,0.0007593148,0.0005961097,0.002023687],"category_scores_gemma":[0.01645034,0.0004613054,0.0005088239,0.001157009,0.001549193,0.001410956,0.001764645,0.0007859542,0.0003384644],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001948144,"about_ca_system_score_gemma":0.002878397,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04620446,"about_ca_topic_score_gemma":0.06274312,"domain_scores_codex":[0.9968665,0.001258599,0.000372664,0.000373146,0.0007235511,0.0004054813],"domain_scores_gemma":[0.9894161,0.00509195,0.002005712,0.0004501114,0.00185837,0.001177763],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002062276,0.0003104619,0.6529549,0.0003971288,0.00006829581,0.001482164,0.3298022,0.00005209351,0.001926936,0.0002091421,0.0006269609,0.0119634],"study_design_scores_gemma":[0.00002767257,0.000505479,0.7691705,0.0001825596,0.00009205113,0.001165614,0.2244546,0.0008728747,0.0006219716,0.00009713835,0.002737194,0.00007230746],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9993393,0.00005662627,0.0000502507,0.000079176,0.000004342964,0.00003330214,0.0000298494,0.000001497611,0.0004057098],"genre_scores_gemma":[0.9987261,0.0001691713,0.0002568753,0.0002070525,0.00000728786,0.00006606674,0.00007306346,0.000006580065,0.0004879248],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04620446,"threshold_uncertainty_score":0.09187102,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0501563872407287,"score_gpt":0.4894948880892685,"score_spread":0.4393385008485398,"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."}}