{"id":"W4376640706","doi":"10.1148/radiol.230987","title":"GPT-4 in Radiology: Improvements in Advanced Reasoning","year":2023,"lang":"en","type":"article","venue":"Radiology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":130,"is_retracted":false,"has_abstract":true,"ca_institutions":"Women's College Hospital; Mount Sinai Hospital","funders":"","keywords":"Medicine; Medical physics; MEDLINE; Radiology; Nuclear 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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004989445,0.00008288299,0.0002638929,0.0003388607,0.0000291332,0.000002053001,0.00006315694,0.0001623126,0.00006072048],"category_scores_gemma":[0.0005754755,0.00008011462,0.00002758892,0.0005441367,0.00005602227,0.00004438739,0.00001870507,0.0002536464,0.0001967733],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001910214,"about_ca_system_score_gemma":0.000133856,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006371702,"about_ca_topic_score_gemma":0.0003712306,"domain_scores_codex":[0.998832,0.00009321033,0.0003611212,0.0002583323,0.00005195873,0.0004033143],"domain_scores_gemma":[0.9994932,0.0001866748,0.00004865636,0.0001772675,0.00002757331,0.00006658753],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00021277,0.00006039796,0.8098046,0.00004431048,0.000009126604,0.0001305315,0.002181471,0.0002310279,0.009224268,0.0007410134,0.001123598,0.1762369],"study_design_scores_gemma":[0.0005452061,0.0007922053,0.9731209,0.0001831487,0.00001003653,0.0002009053,0.003582939,0.005663431,0.004170597,0.005467816,0.006046547,0.000216306],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9949483,0.000366984,0.00002771277,0.002712753,0.000740373,0.000325621,9.821442e-7,0.00006003707,0.0008172524],"genre_scores_gemma":[0.9976909,0.000531024,0.000293561,0.0007564701,0.00018758,0.00009360797,0.00005275678,0.00001178876,0.0003823315],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1760206,"threshold_uncertainty_score":0.3266979,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09472364195722582,"score_gpt":0.4326889900903728,"score_spread":0.3379653481331469,"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."}}