{"id":"W4399577082","doi":"10.1145/3650105.3652291","title":"Assessing the Impact of GPT-4 Turbo in Generating Defeaters for Assurance Cases","year":2024,"lang":"en","type":"article","venue":"","topic":"Safety Systems Engineering in Autonomy","field":"Engineering","cited_by":8,"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":[0.0002503491,0.0001272229,0.0001571328,0.0000918863,0.00001884157,0.0001036485,0.00009719319,0.00004322318,0.00001112427],"category_scores_gemma":[0.00005928249,0.00008510618,0.0001082368,0.0002086274,0.00001143982,0.0001915027,0.000009629533,0.00009232324,0.00000305921],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001549204,"about_ca_system_score_gemma":0.00003917533,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001613618,"about_ca_topic_score_gemma":0.00001739176,"domain_scores_codex":[0.9993604,0.00001293983,0.0002581383,0.0001104976,0.00006205015,0.000195916],"domain_scores_gemma":[0.9992626,0.0005364011,0.00001323088,0.000152487,0.00001497159,0.00002031078],"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":[5.874822e-7,0.000002112829,0.0009628732,0.0002645701,0.00005458068,0.000007006719,0.0002473726,0.9761304,0.01851942,0.00008195933,0.0004818217,0.003247352],"study_design_scores_gemma":[0.00008270827,0.0000124849,0.005418998,0.0002261002,0.000006299275,0.00003188483,0.00008822959,0.9884277,0.005301852,0.000009669659,0.0002790096,0.0001150613],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8744243,0.001910322,0.1208244,0.00002418632,0.0006336595,0.0003054758,0.0000116565,0.0004025719,0.001463363],"genre_scores_gemma":[0.9936015,0.000006059913,0.006130681,0.000003142961,0.0001081286,0.00005760007,0.000002853863,0.00004243741,0.00004761148],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1191771,"threshold_uncertainty_score":0.347053,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02299245954635233,"score_gpt":0.3056516118762312,"score_spread":0.2826591523298789,"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."}}