{"id":"W4388205631","doi":"10.1200/op.2023.19.11_suppl.511","title":"Large language models encode medical oncology knowledge: Performance on the ASCO and ESMO examination questions.","year":2023,"lang":"en","type":"article","venue":"JCO Oncology Practice","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; University Health Network; Princess Margaret Cancer Centre","funders":"","keywords":"Medicine; Confidence interval; Internal medicine; Oncology; Family 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":[0.01327521,0.001589559,0.0009145866,0.001458457,0.0005193737,0.001867148,0.001401516,0.002627828,0.004440392],"category_scores_gemma":[0.08504023,0.0004554286,0.001064937,0.0006221442,0.0005042502,0.003180518,0.002802029,0.002247237,0.001994436],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0015404,"about_ca_system_score_gemma":0.001410128,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008545728,"about_ca_topic_score_gemma":0.007696575,"domain_scores_codex":[0.9944639,0.003840469,0.0002949557,0.0008513349,0.0003672534,0.0001820911],"domain_scores_gemma":[0.8707359,0.1220775,0.001690238,0.001893046,0.001938637,0.001664613],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.02378996,0.005160837,0.1581817,0.002963114,0.00166154,0.001248382,0.004114901,0.1247885,0.01424944,0.001669135,0.03765429,0.6245182],"study_design_scores_gemma":[0.0009084936,0.002848137,0.0406711,0.0003648606,0.0004752525,0.0004701818,0.001229692,0.9329461,0.008841652,0.0046719,0.006389004,0.0001837075],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9522303,0.001925356,0.02409506,0.00224174,0.0003524191,0.0006022095,0.004649699,0.007550082,0.006353118],"genre_scores_gemma":[0.9623967,0.0002620625,0.02653552,0.000674354,0.00009081398,0.0003656416,0.007227001,0.0001771192,0.002270674],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01327521,"threshold_uncertainty_score":0.07020682,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1754834882070465,"score_gpt":0.5117673540927489,"score_spread":0.3362838658857024,"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."}}