{"id":"W4401719093","doi":"10.1109/rew61692.2024.00011","title":"Using GPT-4 Turbo to Automatically Identify Defeaters in Assurance Cases","year":2024,"lang":"en","type":"article","venue":"","topic":"Safety Systems Engineering in Autonomy","field":"Engineering","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa; York University","funders":"","keywords":"Turbo; Computer science; Engineering","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":[0.0001622564,0.0001744772,0.0001928436,0.0002821409,0.00001263808,0.0001193612,0.0001355074,0.00006874443,0.00006889416],"category_scores_gemma":[0.00005342099,0.0001759872,0.00005160398,0.0004816992,0.000008496843,0.0001707189,0.00002819432,0.000129087,0.00036349],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002902607,"about_ca_system_score_gemma":0.0000263025,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000878038,"about_ca_topic_score_gemma":0.00004699357,"domain_scores_codex":[0.9990236,0.00001241282,0.0003197076,0.0002012289,0.0001368516,0.0003061641],"domain_scores_gemma":[0.9995174,0.0001630329,0.000005156544,0.000215804,0.00001154191,0.00008701212],"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":[0.000001254715,0.000004848767,0.000567458,0.0006816146,0.00004892773,0.0003409906,0.0004434001,0.9798388,0.0146591,0.0007072572,0.001187259,0.001519122],"study_design_scores_gemma":[0.00008650395,0.000008779759,0.008103345,0.0007188729,0.00001053226,0.0001635783,0.00004568461,0.981891,0.003539099,0.00003436409,0.005064544,0.000333746],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7903875,0.0007911495,0.1950466,0.0001205723,0.002512491,0.0003891664,0.000008587207,0.003201941,0.00754204],"genre_scores_gemma":[0.98047,0.00000336746,0.01911246,0.00002882267,0.00007978192,0.00003073697,0.000001242684,0.00006550344,0.0002080889],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1900825,"threshold_uncertainty_score":0.7176548,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02702744729768758,"score_gpt":0.2878635042349496,"score_spread":0.260836056937262,"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."}}