{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003517679,0.002022065,0.0008273366,0.002636871,0.0008215479,0.00346742,0.003108467,0.001994929,0.0110555],"category_scores_gemma":[0.02066433,0.001388037,0.003483505,0.0008871931,0.002010307,0.004280533,0.00391579,0.003662055,0.005141494],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001510602,"about_ca_system_score_gemma":0.002845905,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006098316,"about_ca_topic_score_gemma":0.007542707,"domain_scores_codex":[0.996931,0.0008609614,0.00043564,0.0005561844,0.001062181,0.0001539715],"domain_scores_gemma":[0.98565,0.009649619,0.0008901025,0.001615858,0.001893121,0.0003013607],"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":[0.001169944,0.0006415405,0.01172941,0.004250813,0.0003787527,0.003663577,0.006702018,0.1741276,0.05126468,0.1511209,0.07788806,0.5170628],"study_design_scores_gemma":[0.0002332951,0.0002353261,0.0009055535,0.000367644,0.0001390106,0.001191999,0.0005078966,0.8181358,0.03100651,0.0590599,0.08807132,0.0001457878],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01013896,0.00009644506,0.9141117,0.0004113104,0.000107351,0.000656208,0.00231538,0.06862764,0.003534984],"genre_scores_gemma":[0.07669366,0.0001917175,0.9061503,0.0003875759,0.00004121628,0.0005788285,0.007320806,0.005183239,0.00345254],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0110555,"threshold_uncertainty_score":0.03698432,"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."}}