{"id":"W7116894095","doi":"10.1038/s41746-025-02256-z","title":"The impact of GenAI on applicant behaviour, performance, and interview reliability during virtual interviews for medical school admissions","year":2025,"lang":"en","type":"article","venue":"npj Digital Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Harm; Reliability (semiconductor); Guard (computer science); Test (biology); Control (management); Medical school; Treatment and control groups","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":[0.000830386,0.0001619373,0.0003935199,0.0001185287,0.0001713827,0.0000175999,0.0001568096,0.0001123707,0.0002169088],"category_scores_gemma":[0.007942513,0.00008424463,0.0001242577,0.0002351399,0.0003035764,0.00007640233,0.00007071682,0.000311686,0.00000646735],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001638689,"about_ca_system_score_gemma":0.000737932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001337081,"about_ca_topic_score_gemma":0.00001467933,"domain_scores_codex":[0.9983786,0.00003557863,0.0007687006,0.0002636962,0.0002959935,0.0002573892],"domain_scores_gemma":[0.9979091,0.0006290214,0.0001232189,0.0003966648,0.0002346395,0.0007073226],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001042918,0.0004168078,0.0753187,0.001124717,0.00008049999,0.000002652366,0.000722073,0.000002184131,0.0002409743,0.0003697565,0.009200237,0.9114785],"study_design_scores_gemma":[0.00307631,0.02658733,0.8698919,0.03342725,0.0005497888,0.0002044235,0.01186529,0.006450662,0.01133305,0.004425578,0.03146334,0.0007250087],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9829023,0.0006946527,0.0002085724,0.01254881,0.0003199045,0.0009379156,0.00001206598,0.00002780299,0.002347922],"genre_scores_gemma":[0.9970338,0.001048249,0.00001259531,0.0005985347,0.0002408932,0.000138836,0.0000248111,0.00001172317,0.0008905568],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9107535,"threshold_uncertainty_score":0.9508502,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07737484030008086,"score_gpt":0.4451734372443007,"score_spread":0.3677985969442198,"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."}}