{"id":"W4414309860","doi":"10.2139/ssrn.5464597","title":"Estimating the Harmlessness-Accuracy Trade-off in AI: Evidence from LLM-based Hiring Audits","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Auction Theory and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Audit; Product (mathematics); Production (economics); Dimension (graph theory); Test (biology); Key (lock); Matching (statistics)","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":[0.04817861,0.0005801456,0.001297258,0.001725317,0.001383235,0.003700269,0.001958055,0.003593656,0.006209967],"category_scores_gemma":[0.3079529,0.0005827711,0.0005498755,0.002107192,0.004014015,0.006649291,0.002647364,0.004604228,0.001331585],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001739816,"about_ca_system_score_gemma":0.001994554,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006067121,"about_ca_topic_score_gemma":0.004481485,"domain_scores_codex":[0.9746269,0.01651755,0.001396512,0.001416534,0.004847471,0.001195045],"domain_scores_gemma":[0.4025978,0.4623566,0.07162466,0.0444236,0.01611986,0.002877525],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01472352,0.00430525,0.5966974,0.001698299,0.001198551,0.0003729835,0.003197907,0.05412989,0.002435761,0.05109478,0.01107982,0.2590659],"study_design_scores_gemma":[0.001845021,0.004204215,0.5111894,0.0008245851,0.0007383364,0.0005586692,0.003870209,0.203398,0.007223391,0.2560277,0.009794699,0.0003256724],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9568704,0.00199605,0.0148949,0.004080151,0.00011382,0.0001441471,0.0003870659,0.0002013345,0.02131206],"genre_scores_gemma":[0.9967797,0.0001868049,0.001739409,0.0002123298,0.0000354634,0.00001967149,0.0001116363,0.00001365542,0.0009013899],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04817861,"threshold_uncertainty_score":0.2547958,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07550406263049957,"score_gpt":0.4042784635809353,"score_spread":0.3287744009504357,"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."}}