{"id":"W4367048218","doi":"10.2196/47737","title":"Performance of ChatGPT on UK Standardized Admission Tests: Insights From the BMAT, TMUA, LNAT, and TSA Examinations","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":102,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Context (archaeology); Standardized test; Psychology; Computer science; Medical education; Medicine; Mathematics education; History","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":[0.01414661,0.0006125188,0.0006854949,0.002402799,0.0005006197,0.001922972,0.0008331501,0.0007480764,0.003351285],"category_scores_gemma":[0.1032353,0.0002731791,0.0005870747,0.002350095,0.0009856856,0.001508814,0.003915776,0.0007827145,0.001974662],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002738853,"about_ca_system_score_gemma":0.001704837,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02793699,"about_ca_topic_score_gemma":0.02579164,"domain_scores_codex":[0.9869682,0.006364012,0.00165674,0.001140592,0.003215737,0.0006546972],"domain_scores_gemma":[0.9266256,0.04052813,0.0110175,0.005006494,0.01369776,0.003124533],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007714714,0.0002912496,0.9152436,0.0002123761,0.0001123845,0.0003472691,0.009593914,0.00229173,0.0006972516,0.0003218725,0.003265653,0.06685127],"study_design_scores_gemma":[0.00003160592,0.0009324009,0.9859681,0.000129985,0.00003408444,0.0003600661,0.003288173,0.004736527,0.001235778,0.0002246642,0.003009357,0.00004908979],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9959007,0.0001904424,0.0007519409,0.000231093,0.00002576633,0.00005513244,0.0007483118,0.00008014245,0.002016475],"genre_scores_gemma":[0.9972452,0.00009375712,0.0007240641,0.00005123185,0.00001162975,0.0000489312,0.0008615864,0.00002570881,0.0009379536],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02793699,"threshold_uncertainty_score":0.07481533,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06834017598834076,"score_gpt":0.4325670551429926,"score_spread":0.3642268791546519,"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."}}