Learning to Solve Hard Problems in RL for LLMs by Never Giving Up

• URL: http://arxiv.org/abs/2609.13443v1
• Date: Fri, 11 Sep 2026 19:02:08 GMT
• Title: Learning to Solve Hard Problems in RL for LLMs by Never Giving Up
• Abstract summary: We show that training LLMs with RL does not improve performance equally across a dataset.<n>We introduce Never Give Up (NGU), a simple adaptive sampling method that keeps generating samples for a problem until one is correct.
• Score: 11.62149809697467
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: We demonstrate that training LLMs with RL does not improve performance equally across a dataset. RL shows large improvements on easy problems that an LLM is already good at solving, but small improvements on hard problems. We call this the Matthew Effect in RL for LLMs, after the phenomenon of cumulative advantage from economics and network science summarized as "the rich get richer". The naive explanation is that hard problems require more compute to find a solution. We argue that modern RL methods are exacerbating the issue by wasting too much compute on easy problems and instead should dynamically reallocate how they use compute. We introduce Never Give Up (NGU), a simple adaptive sampling method that keeps generating samples for a problem until one is correct. By leveraging asynchronous RL, this naturally uses fewer samples to filter out easy problems and allocates more compute to solving harder problems. We investigate the design choices that affect NGU, such as off-policy robustness, and develop a set of best practices. On the math benchmark Deepscaler, NGU improves performance per compute, especially on harder problems. On a recent coding task, Manufactoria, standard GRPO with a per-test reward fails to fully solve problems that have a range of easy and difficult tests. NGU iteratively improves, solving harder and harder tests, until it learns to fully solve coding problems.

Related papers

• DecompRL: Solving Harder Problems by Learning Modular Code Generation [27.116622101805984]
  We introduce DecompRL, an RL algorithm that explicitly learns to decompose and implement hierarchical code structures.<n>On LiveCodeBench and CodeContests, DecompRL outperforms standard and diversity-optimized RL baselines beyond $105$ tokens per problem.
  arXiv  Detail & Related papers  (2026-07-02T16:25:10Z)
• Cog-DRIFT: Exploration on Adaptively Reformulated Instances Enables Learning from Hard Reasoning Problems [70.94268452442147]
  Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning abilities of LLMs.<n>We propose a simple yet effective solution based on task reformulation.<n>We introduce Cog-DRIFT, a framework that constructs reformulated variants and organizes them into an adaptive curriculum.
  arXiv  Detail & Related papers  (2026-04-06T15:38:38Z)
• Nudging the Boundaries of LLM Reasoning [77.26972440427285]
  Current online reinforcement learning algorithms cannot learn from problems that are "unsolvable" to the model.<n>We propose NuRL, a "nudging" method that aims to push the upper bound of LLM reasoning using self-generated hints.<n>NuRL achieves consistent improvements across 6 benchmarks and 3 models, while remaining complementary to test-time scaling.
  arXiv  Detail & Related papers  (2025-09-30T02:01:40Z)
• SciML Agents: Write the Solver, Not the Solution [69.5021018644143]
  We introduce two new datasets: a diagnostic dataset of adversarial "misleading" problems; and a large-scale benchmark of 1,000 diverse ODE tasks.<n>We evaluate open- and closed-source LLM models along two axes: (i) unguided versus guided prompting with domain-specific knowledge; and (ii) off-the-shelf versus fine-tuned variants.<n>Preliminary results indicate that careful prompting and fine-tuning can yield a specialized LLM agent capable of reliably solving simple ODE problems.
  arXiv  Detail & Related papers  (2025-09-12T02:53:57Z)
• Staying in the Sweet Spot: Responsive Reasoning Evolution via Capability-Adaptive Hint Scaffolding [59.60915947702282]
  Reinforcement learning with verifiable rewards (RLVR) has achieved remarkable success in enhancing the reasoning capabilities of large language models (LLMs)<n>Existing RLVR methods often suffer from exploration inefficiency due to mismatches between the training data's difficulty and the model's capability.<n>We propose SEELE, a novel supervision-aided RLVR framework that dynamically adjusts problem difficulty to stay within the high-efficiency region.
  arXiv  Detail & Related papers  (2025-09-08T17:36:21Z)
• Curriculum Reinforcement Learning from Easy to Hard Tasks Improves LLM Reasoning [58.62311540316617]
  We aim to improve the reasoning capabilities of language models via reinforcement learning (RL)<n>We propose to schedule tasks from easy to hard (E2H), allowing LLMs to build reasoning skills gradually.<n>E2H Reasoner significantly improves the reasoning ability of small LLMs (1.5B to 3B)
  arXiv  Detail & Related papers  (2025-06-07T02:41:54Z)
• Decomposing Elements of Problem Solving: What "Math" Does RL Teach? [22.517954679764244]
  We decompose problem solving into fundamental capabilities: Plan, Execute, and Verify.<n>We show that RL-trained models struggle with fundamentally new problems, hitting a 'coverage wall' due to insufficient planning skills.<n>Our findings provide insights into the role of RL in enhancing LLM reasoning, expose key limitations, and suggest a path toward overcoming these barriers.
  arXiv  Detail & Related papers  (2025-05-28T18:18:49Z)
• Mitigating Tail Narrowing in LLM Self-Improvement via Socratic-Guided Sampling [38.7578639980701]
  Self-improvement methods enable large language models to generate solutions themselves.<n>We find that models tend to over-sample on easy queries and under-sample on queries they have yet to master.<n>We introduce Guided Self-Improvement (GSI), a strategy aimed at improving the efficiency of sampling challenging heavy-tailed data.
  arXiv  Detail & Related papers  (2024-11-01T17:18:45Z)
• Navigating the Labyrinth: Evaluating LLMs' Ability to Reason About Search Problems [62.76627483915117]
  Large Language Models (LLMs) have recently achieved impressive performance in math and reasoning benchmarks.<n>We introduce a new benchmark, SearchBench, which contains 11 unique search problems inspired by intuitive puzzles.<n>We show that using step-by-step, language-only reasoning, even the most advanced LLMs fail to solve SearchBench.
  arXiv  Detail & Related papers  (2024-06-18T00:44:58Z)
• Distilling Algorithmic Reasoning from LLMs via Explaining Solution Programs [2.3020018305241337]
  Distilling explicit chain-of-thought reasoning paths has emerged as an effective method for improving the reasoning abilities of large language models. We propose a novel approach to distill reasoning abilities from LLMs by leveraging their capacity to explain solutions. Our experiments demonstrate that learning from explanations enables the Reasoner to more effectively guide program implementation by a Coder.
  arXiv  Detail & Related papers  (2024-04-11T22:19:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.