{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04502945,0.0003969311,0.0003958872,0.0006568979,0.001157636,0.001443651,0.0007101308,0.0005089153,0.001897126],"category_scores_gemma":[0.1471459,0.000350617,0.0004722108,0.0003901755,0.001981881,0.0007001911,0.00235284,0.0009692154,0.0004267616],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001120695,"about_ca_system_score_gemma":0.001399354,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001599936,"about_ca_topic_score_gemma":0.004463415,"domain_scores_codex":[0.9422731,0.0492528,0.001663133,0.001521818,0.00360733,0.001681805],"domain_scores_gemma":[0.7495986,0.2060793,0.01930065,0.01242252,0.005646219,0.006952702],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01510677,0.004772472,0.7293344,0.0003503436,0.0001978645,0.0004925189,0.08771468,0.001266713,0.02234712,0.0006788907,0.0006468741,0.1370915],"study_design_scores_gemma":[0.0001821854,0.02120605,0.9461467,0.00009603553,0.0001355936,0.0003884207,0.01733707,0.002561193,0.009889108,0.0004330356,0.001524498,0.0001000839],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9984418,0.00001983918,0.0004599763,0.00006136736,0.000008707328,0.00005578297,0.00000992784,0.00001247022,0.0009301636],"genre_scores_gemma":[0.9981944,0.00001997022,0.001130001,0.00004622411,0.00001070217,0.0001186462,0.00001866971,0.000007321504,0.0004540939],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9549705,"threshold_uncertainty_score":0.2381413,"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."}}