{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008247478,0.00009073596,0.000181,0.0005069723,0.0001446464,0.00007553888,0.00006937173,0.00006206261,0.0001359946],"category_scores_gemma":[0.00007614919,0.00008320768,0.0000674928,0.0002249815,0.00002275872,0.0001722963,0.00001932515,0.0005758313,6.574642e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001130813,"about_ca_system_score_gemma":0.0006478429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006372113,"about_ca_topic_score_gemma":0.001008827,"domain_scores_codex":[0.9987428,0.00007000018,0.0005537707,0.0001384268,0.0003642371,0.0001307398],"domain_scores_gemma":[0.9989514,0.00009127599,0.0002217174,0.00004648331,0.0006193361,0.0000698252],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001922578,0.0005436462,0.787756,0.00007742899,0.0002408431,0.00002108233,0.002029946,0.0001665543,0.01374529,0.003622041,0.0001699349,0.1914349],"study_design_scores_gemma":[0.0007505793,0.000697155,0.9380602,0.0009562686,0.00009742114,0.0003273304,0.01022937,0.01498042,0.02135952,0.003644149,0.008700321,0.0001972334],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9700122,0.00007684759,0.01078518,0.0140045,0.001472884,0.0001501215,0.000001374177,0.00001127302,0.003485547],"genre_scores_gemma":[0.9975246,0.0002285173,0.0009072422,0.0007427217,0.000387593,0.0000116228,0.000003767722,0.000005535805,0.0001884586],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1912377,"threshold_uncertainty_score":0.3393111,"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."}}