{"id":"W4411630053","doi":"10.18653/v1/2024.eacl-long.175","title":"Gradient-Based Language Model Red Teaming","year":2024,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Atomic Energy of Canada Limited","keywords":"Computer science; Natural language processing; Artificial intelligence","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":[],"consensus_categories":[],"category_scores_codex":[0.0001330183,0.00006585156,0.00005683538,0.00008317499,0.00003433155,0.0001543225,0.0003671284,0.00002542553,0.00001879698],"category_scores_gemma":[0.000009868317,0.00005476083,0.00004152656,0.0001643036,0.000006903444,0.0002109127,0.00008431539,0.00007686748,0.00005493614],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003044846,"about_ca_system_score_gemma":0.00005569306,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000252251,"about_ca_topic_score_gemma":0.000007773265,"domain_scores_codex":[0.9993072,0.00001084088,0.0000992647,0.0002789465,0.0001395798,0.0001641364],"domain_scores_gemma":[0.9995255,0.0000298159,0.000007717652,0.000379837,0.00001072812,0.00004642174],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000001635458,0.00003761702,0.00005901053,0.00009159422,0.0000179658,0.0001665957,0.003798805,0.1349548,0.01705988,0.5863773,0.004788297,0.2526465],"study_design_scores_gemma":[0.00005028433,0.000005575409,0.000003935962,0.00002011197,0.000001591272,0.000002657504,0.0000253219,0.9954932,0.002281215,0.001688147,0.0003522141,0.00007571649],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02348416,0.0001779261,0.9624971,0.001239242,0.000235169,0.00004554177,4.603976e-7,0.0007020321,0.0116184],"genre_scores_gemma":[0.7834367,5.897274e-7,0.2135791,0.0003455736,0.00003331166,0.000004397889,7.083493e-7,0.000005390767,0.002594234],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8605385,"threshold_uncertainty_score":0.2233082,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01901999588727844,"score_gpt":0.2566363238250168,"score_spread":0.2376163279377383,"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."}}