{"id":"W6922631198","doi":"10.13026/ckbf-pt70","title":"Asclepius-R : Clinical Large Language Model Built On MIMIC-III Discharge Summaries","year":2024,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Usability; Resource (disambiguation); Language model; Benchmark (surveying); Data modeling","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":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001157763,0.001231526,0.001625469,0.0004618372,0.0002734748,0.0003178276,0.001632709,0.0008878483,0.001188991],"category_scores_gemma":[0.0004013173,0.001043626,0.0009114564,0.0005866785,0.0002664893,0.0002767088,0.001574974,0.002962534,0.2443864],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001896833,"about_ca_system_score_gemma":0.00034439,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004183264,"about_ca_topic_score_gemma":0.001085589,"domain_scores_codex":[0.9936165,0.0004286251,0.001286731,0.001969521,0.001411458,0.001287204],"domain_scores_gemma":[0.9952141,0.000447135,0.0005196224,0.003244728,0.0001036863,0.0004707299],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005310111,0.001496836,0.00000527157,0.0004264791,0.0003581795,0.0001312774,0.0001486889,0.0001237796,0.0002057114,0.0002755928,0.9961381,0.0001590734],"study_design_scores_gemma":[0.001195944,0.0002068611,0.00005111862,0.0006459214,0.0005566545,0.000003490427,0.000077468,0.006248581,0.0000611645,0.001010925,0.988752,0.001189881],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.002424475,0.001038618,0.000006685797,0.0001957055,0.001759351,0.0007800133,0.9928059,0.0006088432,0.0003804302],"genre_scores_gemma":[0.00300737,0.0005938714,0.00007466115,0.001047521,0.002451061,0.0001693835,0.9904616,0.0004408184,0.001753689],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.2431974,"threshold_uncertainty_score":0.999724,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04527118146809302,"score_gpt":0.3859986484987003,"score_spread":0.3407274670306073,"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."}}