{"id":"W4410926553","doi":"10.1111/ijsa.70015","title":"Can ChatGPT Outperform Humans in Faking a Personality Assessment While Avoiding Detection?","year":2025,"lang":"en","type":"article","venue":"International Journal of Selection and Assessment","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; Wilfrid Laurier University","funders":"","keywords":"Psychology; Personality; Applied psychology; Social psychology","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":[0.01993822,0.0005395755,0.000779254,0.0007524886,0.0007175871,0.002868914,0.0007147493,0.0009585301,0.004015502],"category_scores_gemma":[0.07266456,0.0002832694,0.0003572752,0.0003632873,0.001228084,0.002111695,0.00195367,0.0007924564,0.001468719],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004501866,"about_ca_system_score_gemma":0.000577494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006653065,"about_ca_topic_score_gemma":0.001227626,"domain_scores_codex":[0.9876111,0.008963634,0.0005060866,0.001023966,0.00152241,0.0003727708],"domain_scores_gemma":[0.9146556,0.05650451,0.01036332,0.01078689,0.003912274,0.003777501],"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":[0.01167566,0.003738996,0.3965635,0.001512903,0.0005979449,0.000623421,0.01868069,0.006656949,0.03201516,0.00333534,0.009753921,0.5148456],"study_design_scores_gemma":[0.001470111,0.02765264,0.7794062,0.0007406945,0.0008007542,0.002064906,0.01576785,0.1081468,0.02653147,0.01336379,0.02336094,0.0006938057],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.987667,0.0001413653,0.006137378,0.0004841336,0.00008381707,0.0001456491,0.00009472996,0.0002721081,0.004973716],"genre_scores_gemma":[0.9929884,0.00005082969,0.005561807,0.0001961886,0.0000288012,0.00007840918,0.00006323059,0.00002286054,0.001009579],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01993822,"threshold_uncertainty_score":0.1054447,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09526578189988998,"score_gpt":0.4751924974480309,"score_spread":0.3799267155481409,"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."}}