{"id":"W4386346510","doi":"10.1002/ijgo.15083","title":"Performance of <scp>ChatGPT</scp> in French language Parcours d'Accès Spécifique Santé test and in <scp>OBGYN</scp>","year":2023,"lang":"en","type":"article","venue":"International Journal of Gynecology & Obstetrics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Jewish General Hospital","funders":"","keywords":"Medicine; Obstetrics and gynaecology; Test (biology); Biostatistics; Entrance exam; Gynecology; Internal medicine; Genetics; Pregnancy; Epidemiology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002997482,0.0005070302,0.0004637844,0.001757762,0.0004532648,0.0008992141,0.0005435885,0.0009631861,0.003587504],"category_scores_gemma":[0.01080342,0.0001896235,0.0006270846,0.0009922714,0.0005448171,0.0005787982,0.0007975363,0.0004948539,0.0008905148],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001671027,"about_ca_system_score_gemma":0.001531844,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0349305,"about_ca_topic_score_gemma":0.02923311,"domain_scores_codex":[0.9974048,0.000677765,0.0002066412,0.0003917761,0.0008152858,0.0005037683],"domain_scores_gemma":[0.9919572,0.001739176,0.002759556,0.0002288573,0.001664557,0.001650538],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009906341,0.0001059437,0.9936148,0.00003350245,0.00003856244,0.00009836179,0.000643021,0.00007272456,0.0002331222,0.00002024955,0.0005047825,0.004535742],"study_design_scores_gemma":[0.000003096959,0.0002185743,0.9988765,0.000008867662,0.000006918885,0.0001251982,0.0002982839,0.0001107557,0.0000766647,0.000005045303,0.0002659675,0.000004073641],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9982619,0.0001789391,0.00007965777,0.00009572305,0.00001234884,0.00002575373,0.0002788683,0.00001665487,0.001050207],"genre_scores_gemma":[0.9987497,0.0000665074,0.0001427424,0.00004752013,0.00001131973,0.00002370171,0.0003387579,0.000003704719,0.0006159884],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0349305,"threshold_uncertainty_score":0.06945437,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04465519028056544,"score_gpt":0.3686152954693981,"score_spread":0.3239601051888327,"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."}}