{"id":"W4402670259","doi":"10.18653/v1/2024.findings-acl.927","title":"From Representational Harms to Quality-of-Service Harms: A Case Study on Llama 2 Safety Safeguards","year":2024,"lang":"en","type":"article","venue":"","topic":"Risk Perception and Management","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Compute Canada; Canadian Institute for Advanced Research","keywords":"Quality (philosophy); Risk analysis (engineering); Computer science; Business; Computer security","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007087387,0.0008092869,0.0005574259,0.0008503197,0.001366425,0.001709844,0.001873855,0.003183449,0.004304368],"category_scores_gemma":[0.04034526,0.0003142435,0.0008373851,0.0007928238,0.002278462,0.002971896,0.002587056,0.00329715,0.0008342897],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002794918,"about_ca_system_score_gemma":0.001477232,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008390393,"about_ca_topic_score_gemma":0.0095093,"domain_scores_codex":[0.9935641,0.004750953,0.0002253061,0.0003433116,0.0008113374,0.0003049938],"domain_scores_gemma":[0.9456659,0.04535276,0.002188101,0.003670064,0.002122371,0.001000665],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"qualitative","study_design_scores_codex":[0.005498839,0.007804437,0.1802194,0.002948756,0.0004162731,0.01704699,0.04061591,0.4442844,0.02211566,0.0685038,0.04239197,0.1681537],"study_design_scores_gemma":[0.0009564067,0.004606068,0.03606531,0.0005488519,0.0002643916,0.00356349,0.01979594,0.7795855,0.02529602,0.05457864,0.07439225,0.0003470782],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9561852,0.0004483594,0.0286792,0.003860696,0.00008005011,0.0004276249,0.001106327,0.0009528738,0.008259827],"genre_scores_gemma":[0.9742408,0.0002002091,0.02041418,0.0007463137,0.00003196989,0.0003174556,0.0006317376,0.0001591597,0.003258123],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008390393,"threshold_uncertainty_score":0.03748208,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1441733007352848,"score_gpt":0.4942732040697287,"score_spread":0.350099903334444,"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."}}