{"id":"W4321022177","doi":"10.2139/ssrn.4361607","title":"Auditing Large Language Models: A Three-Layered Approach","year":2023,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Software Engineering Research","field":"Computer Science","cited_by":72,"is_retracted":false,"has_abstract":false,"ca_institutions":"Institute on Governance","funders":"","keywords":"Audit; Corporate governance; Blueprint; Engineering ethics; Business; Engineering; Accounting; Finance","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":[0.02683224,0.001344866,0.002488328,0.00529032,0.002581239,0.01395421,0.006731809,0.003442156,0.00397595],"category_scores_gemma":[0.09875196,0.002815079,0.003466998,0.003470001,0.003518296,0.0247138,0.01781343,0.005704376,0.001366954],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002443963,"about_ca_system_score_gemma":0.007712827,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008264806,"about_ca_topic_score_gemma":0.01319832,"domain_scores_codex":[0.9582675,0.01325764,0.004430926,0.002800606,0.01824847,0.0029949],"domain_scores_gemma":[0.8725809,0.04062129,0.007950001,0.05730852,0.01880482,0.00273448],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00137074,0.0009264445,0.02795139,0.0007432103,0.0008356987,0.001119069,0.003722419,0.1296915,0.02382413,0.1874487,0.008218477,0.6141482],"study_design_scores_gemma":[0.00004989683,0.0001667619,0.001652804,0.0001749417,0.0002499512,0.0004064265,0.000642406,0.8046535,0.01537495,0.171246,0.005221384,0.0001609518],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01696861,0.0002076508,0.9755893,0.000864027,0.00004373653,0.0003039115,0.0001420446,0.003731214,0.002149463],"genre_scores_gemma":[0.3384138,0.0003000778,0.6556831,0.000311401,0.00009402836,0.0002492439,0.0005779369,0.000871652,0.003498757],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02683224,"threshold_uncertainty_score":0.1419041,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01955062911801558,"score_gpt":0.2626686702789555,"score_spread":0.2431180411609399,"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."}}