{"id":"W4389524193","doi":"10.18653/v1/2023.findings-emnlp.796","title":"Using In-Context Learning to Improve Dialogue Safety","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"SAFER; Computer science; Context (archaeology); Human–computer interaction; Artificial intelligence; Machine learning; Deep learning; Risk analysis (engineering); Computer security","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003627033,0.00006700551,0.00009715512,0.0001492922,0.00006348742,0.00006439128,0.000361091,0.00003263745,0.00001139418],"category_scores_gemma":[0.00008880303,0.00006408979,0.00002550605,0.0005556844,0.000004902998,0.0002071087,0.000405497,0.0001173344,0.0002438013],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005719729,"about_ca_system_score_gemma":0.00003884954,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005311269,"about_ca_topic_score_gemma":0.000197997,"domain_scores_codex":[0.9991116,0.00003513413,0.0001681697,0.0002987294,0.0001196793,0.0002667206],"domain_scores_gemma":[0.9995661,0.00006613244,0.00002009973,0.0002649185,0.00002039466,0.00006234124],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000007037059,0.00001569724,0.005793152,0.00001451156,0.000007569297,0.00005759064,0.008576456,0.2595459,0.01885377,0.06376166,0.000147063,0.6432196],"study_design_scores_gemma":[0.0001360185,0.00001669814,0.001312527,0.00001131729,4.693158e-7,0.000001528895,0.0001896824,0.9952321,0.0006755478,0.0007731402,0.001547529,0.0001034304],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2410761,0.00000645703,0.7545137,0.0006734667,0.0003794432,0.0001001573,2.022215e-7,0.0002566241,0.002993803],"genre_scores_gemma":[0.9668633,0.000001698038,0.03147168,0.0004190557,0.00005980162,0.000002373799,4.821252e-7,0.000005933091,0.001175642],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7356862,"threshold_uncertainty_score":0.3133654,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04731199200695141,"score_gpt":0.2840346666260831,"score_spread":0.2367226746191317,"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."}}