{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002605609,0.001766872,0.001233911,0.000632717,0.0006307305,0.001161923,0.002365798,0.001803844,0.006916833],"category_scores_gemma":[0.009398554,0.0006925542,0.001189593,0.0003804849,0.001416457,0.002438862,0.002435318,0.00349982,0.003432948],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001179797,"about_ca_system_score_gemma":0.001265451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003395814,"about_ca_topic_score_gemma":0.00577427,"domain_scores_codex":[0.9981387,0.0008232837,0.00006744552,0.0005223566,0.0002831187,0.0001651046],"domain_scores_gemma":[0.9958341,0.002634098,0.0002344958,0.0005967732,0.0004892846,0.0002111839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004217116,0.0001931471,0.001433793,0.0003090276,0.000108679,0.00031477,0.0005520588,0.679168,0.01282222,0.02420994,0.01782724,0.2626394],"study_design_scores_gemma":[0.00001745711,0.00003809363,0.0000486334,0.000008364118,0.000007872498,0.00002868175,0.00001581228,0.9884186,0.001938353,0.008437186,0.001030826,0.00001009457],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01773841,0.0003427764,0.9678177,0.0005388707,0.0001583413,0.0001181178,0.0001686293,0.009456862,0.003660321],"genre_scores_gemma":[0.6525941,0.0002401235,0.329065,0.001338887,0.0001662454,0.0004013753,0.0007939941,0.002264476,0.01313573],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006916833,"threshold_uncertainty_score":0.02313906,"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."}}