{"id":"W4401298008","doi":"10.1016/j.jacr.2024.07.011","title":"The Accuracy of Artificial Intelligence ChatGPT in Oncology Examination Questions","year":2024,"lang":"en","type":"article","venue":"Journal of the American College of Radiology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University; University of Toronto","funders":"","keywords":"Medical physics; Computer science; Medicine","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":[0.0136713,0.0006785208,0.0006584665,0.001543704,0.0003240652,0.001580542,0.001575256,0.001465818,0.001869929],"category_scores_gemma":[0.1344054,0.0002653122,0.0006069439,0.000849433,0.0006788701,0.001796364,0.001889587,0.0009992855,0.001248495],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009935342,"about_ca_system_score_gemma":0.0007289564,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003660564,"about_ca_topic_score_gemma":0.003335709,"domain_scores_codex":[0.9842074,0.009140417,0.001298496,0.002114059,0.002679816,0.0005598722],"domain_scores_gemma":[0.8746712,0.1022612,0.007497472,0.004886668,0.008619849,0.002063622],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004525224,0.0009983234,0.5779769,0.0008753839,0.0002763689,0.0004972899,0.004809944,0.007630386,0.01086281,0.0002825493,0.003118635,0.3881462],"study_design_scores_gemma":[0.0001836176,0.004870693,0.8475569,0.0003810063,0.0003553042,0.001859025,0.003141752,0.1070961,0.0277842,0.001559866,0.005007533,0.000203971],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9819661,0.0004369751,0.01179339,0.0003023197,0.00007028215,0.0002925556,0.0004192803,0.0008209857,0.003898239],"genre_scores_gemma":[0.9896579,0.0001353009,0.008170553,0.0001817361,0.00003399088,0.0001045977,0.0006675576,0.00004937764,0.00099897],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0136713,"threshold_uncertainty_score":0.07230157,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07966516608303623,"score_gpt":0.4325808900391962,"score_spread":0.35291572395616,"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."}}