{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001186717,0.00009530054,0.0001489048,0.0001157288,0.0002334313,0.0001454947,0.0001907,0.00003833826,0.007277665],"category_scores_gemma":[0.00009771527,0.0000819266,0.00007160735,0.0005262211,0.00003939545,0.0001535447,0.00009578437,0.00008218595,0.0009069472],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001179235,"about_ca_system_score_gemma":0.00009004857,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2044112,"about_ca_topic_score_gemma":0.06844977,"domain_scores_codex":[0.9980648,0.0003628061,0.0003469824,0.0003775753,0.0006828436,0.0001650435],"domain_scores_gemma":[0.9991968,0.0002832144,0.00003621453,0.0002544236,0.00009851534,0.0001308149],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002330739,0.001025652,0.005012932,0.00004935197,0.0003085874,0.0007038457,0.7421212,0.0003789689,0.0004616259,0.1073225,0.09153835,0.05084393],"study_design_scores_gemma":[0.0005435664,0.0001253353,0.03842725,0.00003303686,0.00004204799,0.000002023936,0.738408,0.0001483942,0.00003255938,0.00177185,0.2202293,0.000236616],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8599768,0.00001495868,0.00180397,0.02425533,0.0007449002,0.0008094567,0.00005700633,0.0001924177,0.1121452],"genre_scores_gemma":[0.9837589,0.00001635548,0.0005607031,0.002483337,0.0003579838,0.00004278689,0.00001312547,0.000009372361,0.01275747],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1359614,"threshold_uncertainty_score":0.999871,"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."}}