{"id":"W7042821887","doi":"","title":"Predictive Power of Large Language Models in Bug Severity Detection: An Explorative Study","year":2024,"lang":"fr","type":"other","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Polytechnique Montréal","keywords":"Context (archaeology); Work (physics); Population; Limiting","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01326118,0.001484515,0.0009114119,0.002331048,0.0005794573,0.00286514,0.001446055,0.001051807,0.00240553],"category_scores_gemma":[0.06002424,0.0006730093,0.002474057,0.001751425,0.0006404116,0.003224913,0.001284629,0.002132232,0.000918425],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001241419,"about_ca_system_score_gemma":0.001199955,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01472975,"about_ca_topic_score_gemma":0.01039566,"domain_scores_codex":[0.9949496,0.003692864,0.0002041005,0.000695185,0.0003171214,0.0001410628],"domain_scores_gemma":[0.8526367,0.1390353,0.002700303,0.002897318,0.002183852,0.0005464797],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00497339,0.00408384,0.5103049,0.001070814,0.002637201,0.001244093,0.005421311,0.2491668,0.00606745,0.002403297,0.006389388,0.2062377],"study_design_scores_gemma":[0.0001462974,0.001100646,0.0582272,0.0001779014,0.0007090382,0.0003630594,0.001526915,0.9309477,0.001793938,0.00320793,0.001698102,0.0001013025],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9739236,0.0009242959,0.02114362,0.0005347577,0.00005242711,0.0002036567,0.001370957,0.0006731424,0.001173546],"genre_scores_gemma":[0.9864588,0.0002099291,0.01094561,0.00006185181,0.00002265637,0.0001441999,0.001589786,0.00008469795,0.0004823687],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01472975,"threshold_uncertainty_score":0.07013267,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01275837360977909,"score_gpt":0.2630264129820638,"score_spread":0.2502680393722846,"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."}}