{"paper":{"title":"Evolutionary System Prompt Learning for Reinforcement Learning in LLMs","arxiv_id":"2602.14697","primary_cat":"cs.AI","submitted_at":"2026-02-16T12:34:27+00:00","version_compiled":"v4"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v4","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Building agentic systems that can autonomously self-improve from experience is a longstanding goal of AI.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.03,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.02,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":1.0,"id":"s1","own":false,"role":"background","text":"Large language models (LLMs) today primarily self-improve via two mechanisms: self-reflection for context updates, and reinforcement learning (RL) for weight updates.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.18,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.97,"id":"s2","own":true,"role":"contribution","text":"In this work, we propose Evolutionary System Prompt Learning (E-SPL), a method for jointly improving a model's experiential context and its weights.","hedge":{"p":0.97,"label":"asserted"},"own_p":0.74,"main_claim_p":0.08,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.03,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.97,"interpretation":0.0}},{"i":3,"p":1.0,"id":"s3","own":true,"role":"method","text":"In each RL iteration, E-SPL samples trajectories under multiple system prompts in parallel, then uses them to jointly update the weights with RL and evolve the system prompts with self-reflection.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.87,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.08,"role_probabilities":{"gap":0.0,"method":1.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":0.71,"id":"s4","own":true,"role":"interpretation","text":"E-SPL encourages a natural division between procedural knowledge in weights and interpretable declarative knowledge in system prompts.","hedge":{"p":0.84,"label":"asserted"},"own_p":0.51,"main_claim_p":0.06,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.12,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.17,"interpretation":0.71}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"result","text":"Most notably, in an easy-to-hard generalization setting (AIME $\\rightarrow$ BeyondAIME), E-SPL improves RL success rate from 38.8% $\\rightarrow$ 45.1%.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.13,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"E-SPL also improves RL on AIME 2025 (56.3% $\\rightarrow$ 60.6%), HMMT 2025 (50.0% $\\rightarrow$ 52.7%), and agentic search (44.2% $\\rightarrow$ 48.6% on gpt-oss-120b).","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.06,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":7,"p":0.7,"id":"s7","own":true,"role":"interpretation","text":"Across all settings, E-SPL outperforms both RL-only and evolution-only baselines, demonstrating that weight updates and context updates are deeply synergistic and can together yield gains in generalization that neither achieves alone.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.9,"quantitative":true,"quantitative_p":0.85,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.3,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.7}},{"i":8,"p":0.7,"id":"s8","own":false,"role":"contribution","text":"Code is available at https://github.com/LunjunZhang/E-SPL.","hedge":{"p":0.75,"label":"asserted"},"own_p":0.44,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.02,"role_probabilities":{"gap":0.0,"method":0.08,"result":0.01,"background":0.02,"limitation":0.01,"future_work":0.18,"contribution":0.7,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":1.0,"text":"Across all settings, E-SPL outperforms both RL-only and evolution-only baselines, demonstrating that weight updates and context updates are deeply synergistic and can together yield gains in generalization that neither achieves alone.","location":"abstract s7","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":7134,"cost_usd_reported":0.0},"compiled_at":"2026-11-05","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.06,"states_limitation":0.06,"human_participants":0.03,"controlled_comparison":0.96,"randomized_assignment":0.02,"states_quantitative_result":0.99,"compares_with_prior_results":0.97,"mentions_code_or_data_availability":0.99},"evidential_basis":{"p":0.92,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.92,"simulation_or_computation":0.08}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.97,"text":"In this work, we propose Evolutionary System Prompt Learning (E-SPL), a method for jointly improving a model's experiential context and its weights.","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v4"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":9,"sentences_in_abstract":9},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}