{"id":"W4407031081","doi":"10.32388/8g8tb2","title":"Enhancing Code LLMs with Reinforcement Learning in Code Generation: A Survey","year":2025,"lang":"en","type":"preprint","venue":"Qeios","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Reinforcement learning; Computer science; Compiler; Code generation; Code (set theory); Resource allocation; Resource (disambiguation); Artificial intelligence; Programming language; Computer security","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001576237,0.0002528881,0.0003298583,0.0003585892,0.00008297701,0.000297023,0.001115409,0.0001791829,0.00001733033],"category_scores_gemma":[0.001027013,0.0002479625,0.00004310292,0.0005758396,0.00002231064,0.0001382111,0.001497011,0.001169961,0.00002466939],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003503658,"about_ca_system_score_gemma":0.0008730775,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008388289,"about_ca_topic_score_gemma":0.002809267,"domain_scores_codex":[0.9976482,0.0002109386,0.0003751545,0.000722502,0.000616578,0.0004265857],"domain_scores_gemma":[0.9980554,0.0006466586,0.0001050646,0.0009029962,0.0002004848,0.00008942459],"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":[0.000008829697,0.00001789342,0.02135191,0.000152047,0.0000353209,0.00004031357,0.0008261747,0.9747446,0.0001908673,0.000346902,0.0003998433,0.001885332],"study_design_scores_gemma":[0.0005033046,0.0001195861,0.03635459,0.0007253228,0.000005422184,0.000005437927,0.00001174641,0.9540061,0.006386221,0.00003446502,0.001320134,0.0005276315],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04607059,0.0001635042,0.9519563,0.0001898193,0.0004936258,0.0004550352,0.000006423606,0.0003619154,0.0003027591],"genre_scores_gemma":[0.9645777,0.00004322378,0.03062898,0.00008394511,0.0001216061,0.0002193652,0.00009179732,0.00002206776,0.004211309],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9213274,"threshold_uncertainty_score":0.9999973,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04221355533285516,"score_gpt":0.2985258908404057,"score_spread":0.2563123355075506,"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."}}