{"id":"W4389633038","doi":"10.5114/pjr.2023.133330","title":"Correspondence to “Will ChatGPT pass the Polish specialty exam in radiology and diagnostic imaging?”","year":2023,"lang":"en","type":"article","venue":"Polish Journal of Radiology","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Specialty; Medicine; Radiology; Medical imaging; Medical physics; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002142057,0.000189827,0.0006899525,0.0006654901,0.00009487705,0.00003320534,0.0003558151,0.0001070968,0.0000999285],"category_scores_gemma":[0.0115309,0.0001299847,0.0001040888,0.0005787539,0.0004431603,0.000126184,0.0001274092,0.00105391,0.00003356172],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001168153,"about_ca_system_score_gemma":0.0001918438,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001705941,"about_ca_topic_score_gemma":0.00002623458,"domain_scores_codex":[0.9978029,0.000434807,0.0006383393,0.000261124,0.000248944,0.0006139604],"domain_scores_gemma":[0.9964206,0.002553753,0.000260594,0.0002449616,0.0001220854,0.0003980405],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004498577,0.00007389724,0.7372323,0.00004485499,0.0001033264,0.003864078,0.00331643,0.0004490259,0.001890347,0.001843308,0.2135906,0.03714203],"study_design_scores_gemma":[0.002041132,0.0004448097,0.8882393,0.0001868805,0.00007788956,0.02646949,0.0004616718,0.001818849,0.00002796443,0.001848375,0.07821371,0.0001699319],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9266222,0.002781708,0.0004483407,0.06741226,0.001859201,0.0002421813,0.000006038403,0.00003051327,0.0005975456],"genre_scores_gemma":[0.9847399,0.002850939,0.0003501055,0.009136239,0.002519498,0.000009306589,0.000005987048,0.00003310565,0.0003549221],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.151007,"threshold_uncertainty_score":0.9967954,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01023175335057074,"score_gpt":0.3013738248586221,"score_spread":0.2911420715080513,"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."}}