{"id":"W4399151229","doi":"10.48550/arxiv.2405.17503","title":"Code Repair with LLMs gives an Exploration-Exploitation Tradeoff","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Scientific Computing and Data Management","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research; Cisco Systems; National Science Foundation","keywords":"Code (set theory); Computer science; Business; Computer security; Programming language","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002475334,0.0003198937,0.0003462311,0.0008529593,0.0002867382,0.0009936499,0.001716503,0.0001512912,0.0001865789],"category_scores_gemma":[0.0003072448,0.0002710935,0.0002243664,0.001743129,0.0002195584,0.001172703,0.001693689,0.0004639316,0.0007844333],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001327078,"about_ca_system_score_gemma":0.0002028215,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001288367,"about_ca_topic_score_gemma":0.0005867005,"domain_scores_codex":[0.9957598,0.0003182803,0.000450415,0.002550624,0.0006071238,0.0003137503],"domain_scores_gemma":[0.9961591,0.000373973,0.0003387973,0.002549714,0.0003604961,0.0002179346],"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":[0.0001562518,0.0002612813,0.00186979,0.00008706494,0.0001940728,0.0007420861,0.003950818,0.8279955,0.00001179754,0.1222505,0.03683534,0.005645548],"study_design_scores_gemma":[0.000308828,0.0001604937,0.001227042,0.0001431596,0.0001650747,0.00000362137,0.009516438,0.6664101,0.00002326873,0.3093231,0.01219644,0.0005224575],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7300372,0.00006359092,0.2566198,0.0006422595,0.002381384,0.0005362396,0.0001679463,0.0009353667,0.008616284],"genre_scores_gemma":[0.9827243,0.00002418594,0.001690984,0.00006815235,0.0001332801,0.000002214978,0.0001326528,0.00002510169,0.01519913],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2549288,"threshold_uncertainty_score":0.9999936,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3719729471813407,"score_gpt":0.2970075874439759,"score_spread":0.07496535973736479,"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."}}