{"id":"W4389896239","doi":"10.1097/gox.0000000000005448","title":"Can a Machine Ace the Test? Assessing GPT-4.0’s Precision in Plastic Surgery Board Examinations","year":2023,"lang":"en","type":"article","venue":"Plastic & Reconstructive Surgery Global Open","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"CLARITY; Computer science; Artificial intelligence; Completeness (order theory); Scale (ratio); Medical physics; Machine learning; Plastic surgery; Test (biology); Medicine; Surgery; Mathematics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001808878,0.0002799126,0.0006701828,0.0004180669,0.0003693874,0.0002553189,0.0002742347,0.0001616646,0.0002400431],"category_scores_gemma":[0.04201695,0.0002202257,0.0001367967,0.00264021,0.0002646922,0.0004587591,0.0002042918,0.0003967631,0.0002284615],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006784889,"about_ca_system_score_gemma":0.002013452,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005470091,"about_ca_topic_score_gemma":0.003822523,"domain_scores_codex":[0.997016,0.0003294581,0.0009766907,0.0005814689,0.0004463945,0.0006500207],"domain_scores_gemma":[0.9437451,0.05489197,0.0003159348,0.0004104486,0.0003797407,0.0002568108],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001164229,0.0001185707,0.8694999,0.00005679858,0.00003455528,0.00007325097,0.0002765254,0.00008201831,0.00004457054,0.0001282407,0.006132605,0.1234366],"study_design_scores_gemma":[0.00008685763,0.0000571566,0.9854504,0.001056874,0.00007005258,0.0002162279,0.003593559,0.004242191,0.0002748546,0.003969165,0.0006979221,0.0002846722],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9894239,0.0001057705,0.0002646314,0.002730725,0.004916474,0.0008261374,0.0001388069,0.0001047141,0.001488793],"genre_scores_gemma":[0.9987195,0.0001289226,0.0002745304,0.0001679832,0.0002399806,0.0001822257,0.0001268011,0.00002737953,0.0001326564],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1231519,"threshold_uncertainty_score":0.9660525,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1464212235719897,"score_gpt":0.4064276775336855,"score_spread":0.2600064539616958,"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."}}