{"id":"W3163530677","doi":"10.1177/23742895211013533","title":"Pathology Resident Evaluation During the Pandemic: Testing and Implementation of a Comprehensive Online Pathology Exam","year":2021,"lang":"en","type":"article","venue":"Academic Pathology","topic":"COVID-19 and healthcare impacts","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mount Sinai Hospital; University of Toronto","funders":"","keywords":"Medical education; Social media; Multiple choice; Coronavirus disease 2019 (COVID-19); Medicine; Psychology; Pathology; Computer science; Internal 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":[0.001191436,0.0001925837,0.0005231344,0.0001699463,0.0001514646,0.000004178595,0.0001013002,0.0004065572,0.00007848878],"category_scores_gemma":[0.00228831,0.0001557792,0.00005641963,0.0003197424,0.0002629998,0.00007052098,0.0002085177,0.0008779196,0.000004553602],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001730786,"about_ca_system_score_gemma":0.0007346445,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001525748,"about_ca_topic_score_gemma":0.0002747157,"domain_scores_codex":[0.9970625,0.0009269476,0.0007527125,0.0005048798,0.000300032,0.0004529207],"domain_scores_gemma":[0.9972789,0.00119971,0.0003903504,0.0003753144,0.0006275013,0.0001282224],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00008615537,0.0000263673,0.5152603,0.0002188525,0.0000237915,0.0006619978,0.002698321,0.000009657277,0.440008,0.00007137855,0.00005422965,0.04088093],"study_design_scores_gemma":[0.002647311,0.0003370865,0.9731255,0.00009905278,0.0002139386,0.01142747,0.004397298,0.0001176732,0.006250743,0.001033937,0.0002320145,0.0001179863],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9877028,0.005310472,0.00006946547,0.005895477,0.0002006028,0.0006920747,0.00004661882,0.0000497529,0.00003272265],"genre_scores_gemma":[0.9915482,0.002651503,0.0007546912,0.00457947,0.0002567553,0.00005725451,0.00008786497,0.00002289741,0.00004132357],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4578652,"threshold_uncertainty_score":0.6352493,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1910317460636685,"score_gpt":0.4817064099533009,"score_spread":0.2906746638896324,"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."}}