{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004086003,0.001838675,0.001146218,0.0009193561,0.001019813,0.001572914,0.002093447,0.001647491,0.00507326],"category_scores_gemma":[0.01807677,0.000475862,0.0008495783,0.0003008443,0.0009327462,0.003546788,0.002853321,0.002967819,0.001716472],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001019137,"about_ca_system_score_gemma":0.001701229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003250404,"about_ca_topic_score_gemma":0.004166539,"domain_scores_codex":[0.9965288,0.001858251,0.0001698904,0.0007280143,0.0005287516,0.0001863301],"domain_scores_gemma":[0.9912528,0.005376813,0.0005773013,0.001191625,0.00122981,0.0003717543],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001231861,0.0013846,0.007430932,0.0007257198,0.000225646,0.0003463591,0.002857509,0.2393109,0.06076342,0.009751892,0.007474457,0.6684967],"study_design_scores_gemma":[0.00005779067,0.0003728815,0.0006217996,0.00004284819,0.00008087186,0.0001065461,0.0002984392,0.9587466,0.02348002,0.01153331,0.004616381,0.00004251272],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06053326,0.0005527431,0.9232625,0.0006197128,0.0001562966,0.0002084898,0.0001434697,0.01047152,0.004051848],"genre_scores_gemma":[0.7488092,0.0002324112,0.2445303,0.000461615,0.0001413319,0.0002343969,0.0004862214,0.0006516352,0.004452908],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.00507326,"threshold_uncertainty_score":0.02160913,"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."}}