{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001038314,0.0001423261,0.0003313801,0.0001581205,0.0001022226,0.0001327132,0.0000945066,0.0002596681,0.0003839051],"category_scores_gemma":[0.007638821,0.0001051958,0.00005614317,0.0003287471,0.0002112019,0.0004379899,0.00004307769,0.0003934495,0.00001383872],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001274401,"about_ca_system_score_gemma":0.001501787,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003856305,"about_ca_topic_score_gemma":0.0003002594,"domain_scores_codex":[0.9974982,0.00003734119,0.001070599,0.0001375561,0.00101456,0.0002418167],"domain_scores_gemma":[0.9978464,0.0009432785,0.000142478,0.0001752295,0.0004931627,0.0003994756],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000131459,0.0007334587,0.04228307,0.006551456,0.0001441241,0.00007119992,0.4378968,0.000002450855,0.00004238242,0.0005077247,0.0009901153,0.5106458],"study_design_scores_gemma":[0.003835086,0.00950524,0.07542878,0.009835603,0.000441765,0.0004239929,0.3816929,0.4977351,0.002159745,0.001180152,0.01680345,0.0009581631],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9947463,0.0006413918,0.00124828,0.0005228673,0.0009903933,0.001140289,0.000008806064,0.0001007017,0.0006009944],"genre_scores_gemma":[0.9977816,0.000206164,0.0002886454,0.0004278341,0.0009265406,0.0002058879,0.0000761418,0.00001686349,0.00007027973],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5096877,"threshold_uncertainty_score":0.9144934,"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."}}