{"id":"W4394845024","doi":"10.1056/aioa2300151","title":"Comparative Evaluation of LLMs in Clinical Oncology","year":2024,"lang":"en","type":"article","venue":"NEJM AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":97,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Sunnybrook Health Science Centre","funders":"National Cancer Institute; NIH Clinical Center; National Institutes of Health","keywords":"Clinical Oncology; Internal medicine; Oncology; Medicine; Cancer","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":[0.01255485,0.001593615,0.0007480932,0.002172422,0.0003997573,0.001851281,0.002630424,0.002035202,0.003774432],"category_scores_gemma":[0.05643551,0.000497699,0.001364866,0.001190238,0.0007873148,0.002963173,0.002304304,0.002149108,0.00139112],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003506851,"about_ca_system_score_gemma":0.001775065,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007098821,"about_ca_topic_score_gemma":0.007419812,"domain_scores_codex":[0.993604,0.003788892,0.0006048668,0.001060136,0.0007602062,0.0001818551],"domain_scores_gemma":[0.9526858,0.03948967,0.001313501,0.00280566,0.002743608,0.0009617656],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.006211509,0.002510506,0.07603617,0.002719878,0.001473498,0.0003260394,0.001269774,0.5099086,0.004945602,0.002977244,0.02066452,0.3709567],"study_design_scores_gemma":[0.0004409087,0.002541513,0.01604968,0.0003027366,0.0003255999,0.0003113084,0.0004181269,0.9592884,0.008375345,0.003914181,0.007926029,0.0001060441],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8820701,0.005648586,0.06475421,0.003494826,0.0004817096,0.0008965242,0.006717852,0.02305355,0.01288263],"genre_scores_gemma":[0.9567631,0.0007582796,0.03243149,0.0004624245,0.00006852769,0.0003201512,0.00767375,0.0003284362,0.001193814],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01255485,"threshold_uncertainty_score":0.06639719,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7465636406134145,"score_gpt":0.698507249034476,"score_spread":0.04805639157893848,"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."}}