Pith.

Artificial Intelligence · arXiv:2602.14697 v5 · submitted 16 Feb 2026

Evolutionary System Prompt Learning for Reinforcement Learning in LLMs

Lunjun Zhang, Ryan Chen, Bradly C. Stadie

The paper says

“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.”

from the abstract · picked out as the main claim by a classifier

Pith says

Pith has not reviewed this paper.

Get a Pith Review, $5

Reviews

Reviews by people

No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Building agentic systems that can autonomously self-improve from experience is a longstanding goal of AI. Large language models (LLMs) today primarily self-improve via two mechanisms: self-reflection for context updates, and reinforcement learning (RL) for weight updates. In this work, we propose Evolutionary System Prompt Learning (E-SPL), a method for jointly improving a model's experiential context and its weights. 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. E-SPL encourages a natural division between procedural knowledge in weights and interpretable declarative knowledge in system prompts. Most notably, in an easy-to-hard generalization setting (AIME → BeyondAIME), E-SPL improves RL success rate from 38.8% → 45.1%. E-SPL also improves RL on AIME 2025 (56.3% → 60.6%), HMMT 2025 (50.0% → 52.7%), and agentic search (44.2% → 48.6% on gpt-oss-120b). 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. Code is available at https://github.com/LunjunZhang/E-SPL.

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-05 · no language-model review

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: 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

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v4). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Building agentic systems that can autonomously self-improve from experience is a longstanding goal of AI. background 1.00 0.03 0.02 asserted 1.00
Large language models (LLMs) today primarily self-improve via two mechanisms: self-reflection for context updates, and reinforcement learning (RL) for weight updates. background 1.00 0.06 0.02 asserted 1.00
In this work, we propose Evolutionary System Prompt Learning (E-SPL), a method for jointly improving a model's experiential context and its weights. contribution 0.97 0.74 0.08 asserted 0.97
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. method 1.00 0.87 0.05 asserted 0.99
E-SPL encourages a natural division between procedural knowledge in weights and interpretable declarative knowledge in system prompts. interpretation 0.71 0.51 0.06 asserted 0.84
Most notably, in an easy-to-hard generalization setting (AIME $\rightarrow$ BeyondAIME), E-SPL improves RL success rate from 38.8% $\rightarrow$ 45.1%. result 1.00 0.95 0.13 asserted 1.00
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). result 1.00 0.95 0.06 asserted 1.00
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. interpretation 0.70 0.94 0.90 asserted 1.00
Code is available at https://github.com/LunjunZhang/E-SPL. contribution 0.70 0.44 0.02 asserted 0.75

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.92

What was examined, and what was not

Sources read

  • arXiv listing abstract v4

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-05: 1 request, 7,134 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

How the literature uses it

2 arXiv papers cite this in 2 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it

Compares with it 2 papers

Challenges and discussion

Know something about this paper? Leave a comment.

No comments yet.

Sign in to comment