{"id":"W4411724140","doi":"10.2196/69485","title":"Performance of Large Language Models in the Non-English Context: Qualitative Study of Models Trained on Different Languages in Chinese Medical Examinations","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Context (archaeology); Natural language processing; Language model; Artificial intelligence; English language; Word error rate; Benchmark (surveying); Linguistics; Psychology; Mathematics education","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":[0.001829353,0.0001336275,0.0004316887,0.0003466292,0.00003720527,0.000005686504,0.0002622367,0.0001821535,0.00006544674],"category_scores_gemma":[0.001229456,0.00008066866,0.00004775814,0.0005969029,0.0001147121,0.0001742555,0.00004416564,0.0005779271,0.000001653348],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006858446,"about_ca_system_score_gemma":0.0003149605,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000498775,"about_ca_topic_score_gemma":0.001429316,"domain_scores_codex":[0.9972441,0.0002002685,0.001203077,0.00009154742,0.001044037,0.0002169548],"domain_scores_gemma":[0.9981953,0.001072626,0.0001693923,0.0002820763,0.0001791765,0.0001014555],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001261577,0.002677144,0.005872575,0.000613642,0.00002101695,0.000006130805,0.9735792,0.0001317557,0.000001910172,0.001926174,0.0001347233,0.01490954],"study_design_scores_gemma":[0.0006582462,0.0005342336,0.009414516,0.0007844004,0.00001100822,0.000001049284,0.7112913,0.2769129,0.00006482181,0.0002702628,0.000003402289,0.00005390788],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.993816,0.00004714736,0.0003651953,0.000760545,0.0001017076,0.001040646,0.000009973825,0.00001307612,0.003845666],"genre_scores_gemma":[0.998659,0.00005495466,0.00002490225,0.00095159,0.00004788833,0.0001759094,0.00003789251,0.000005489433,0.00004240939],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2767811,"threshold_uncertainty_score":0.3289573,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06632131130418917,"score_gpt":0.4693720188904845,"score_spread":0.4030507075862953,"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."}}