{"id":"W4393145420","doi":"10.2196/49964","title":"Performance of ChatGPT on the India Undergraduate Community Medicine Examination: Cross-Sectional Study","year":2024,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Bachelor; Syllabus; Medical education; Likert scale; Psychology; Subject (documents); Multidisciplinary approach; Cross-sectional study; Mathematics education; Medicine; Family medicine; Computer science; Sociology; Library science; Social science; Pathology; 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":[0.002112669,0.0002656051,0.0003867615,0.001767243,0.0007520506,0.0008924412,0.0006863771,0.0007451497,0.002048861],"category_scores_gemma":[0.00739482,0.0004650965,0.0004794172,0.001199318,0.0006102467,0.001046171,0.001160321,0.0009444323,0.0008101283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005827164,"about_ca_system_score_gemma":0.000491056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003547066,"about_ca_topic_score_gemma":0.004783453,"domain_scores_codex":[0.998224,0.0004323921,0.0001949074,0.0002809409,0.0005777107,0.0002900204],"domain_scores_gemma":[0.9929489,0.001856755,0.00236339,0.0003699018,0.001227664,0.001233396],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000519395,0.0002896168,0.9970451,0.00001371272,0.00001643694,0.0001212505,0.0009885922,0.00001790526,0.000193109,0.00001166918,0.00006990923,0.001180858],"study_design_scores_gemma":[0.000003379424,0.0007110033,0.9967673,0.000007448227,0.00001356587,0.0003815986,0.001740084,0.00009790487,0.0001101098,0.000006663571,0.0001539746,0.00000697193],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9994797,0.00002647519,0.00002369873,0.00001676359,0.000002185591,0.00002013965,0.00005507735,0.000002108934,0.0003738385],"genre_scores_gemma":[0.9994392,0.00004222417,0.00006505068,0.00003146842,0.000004909348,0.00001993209,0.0001103862,0.000002510015,0.0002844268],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003547066,"threshold_uncertainty_score":0.01117301,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4084678988529589,"score_gpt":0.572301560329234,"score_spread":0.163833661476275,"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."}}