Question Begets Question: Self-Evolving Curriculum for Reinforcement Fine-Tuning on Competition Mathematics

• URL: http://arxiv.org/abs/2608.01522v1
• Date: Sun, 02 Aug 2026 22:19:01 GMT
• Title: Question Begets Question: Self-Evolving Curriculum for Reinforcement Fine-Tuning on Competition Mathematics
• Abstract summary: Training data is scarce, ground-truth reasoning traces are usually unavailable, and models often exhibit an apparent ceiling beyond which additional data yields no further improvement.<n>We study these difficulties in a controlled setting, fine-tuning Qwen2.5-Math-7B on competition mathematics (AIME)<n>We propose a self-evolving curriculum that, each round, evaluates the current checkpoint, seeds QbQ from the problems it can mostly get right, and trains on the resulting variants.
• Score: 4.56953073429622
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Teaching a language model a skill it has not mastered is obstructed by three recurring difficulties: training data is scarce, ground-truth reasoning traces are usually unavailable, and models often exhibit an apparent ceiling beyond which additional data yields no further improvement. We study these difficulties in a controlled setting, fine-tuning Qwen2.5-Math-7B on competition mathematics (AIME), a task on which it initially solves only 5.6\% of problems (pass@1). To address data scarcity, we introduce Question-begets-Question (QbQ), a scalable procedure in which a teacher transforms existing problems into diverse variants that probe the same underlying skills; to model the absence of oracle reasoning, we train exclusively via reinforcement learning on problem statements and final answers, never on teacher reasoning traces. Static training on such data, however, plateaus well short of the task: real-plus-synthetic augmentation and non-curriculum QbQ generated synthetic data training cap pass@1 at 12.5\% and 14.5\% respectively, despite large increases in data. Our central finding is that this ceiling is not intrinsic to the model. We propose a self-evolving curriculum that, each round, evaluates the current checkpoint, seeds QbQ from the problems it can mostly get right, and trains on the resulting variants; under an identical data budget, this breaks the ceiling and lifts pass@1 to 16.5\% with no sign of saturation after 20 rounds. Counterintuitively, we find that models improve when trained on variants of problems they can mostly get right, and that models trained this way go on to solve harder problems never seen during training.

Related papers

• Three Concrete Challenges and Two Hopes for the Safety of Unsupervised Elicitation [2.5107780917370985]
  We argue that datasets used for evaluations could cause overoptimistic evaluation results.<n>Unlike many real-world datasets, they often have no features with more salience than truthfulness.<n>We construct datasets that lack each of these properties to stress-test a range of standard unsupervised elicitation and easy-to-hard generalization techniques.
  arXiv  Detail & Related papers  (2026-02-23T22:39:40Z)
• From Abstract to Contextual: What LLMs Still Cannot Do in Mathematics [79.81905350372067]
  We study gap through contextual mathematical reasoning.<n>We introduce ContextMATH, a benchmark that repurposes AIME and MATH-500 problems into two contextual settings.<n>Open-source models decline by 13 and 34 points on SG and CS, while proprietary models drop by 13 and 20.
  arXiv  Detail & Related papers  (2026-01-30T14:56:04Z)
• Can Large Reasoning Models Improve Accuracy on Mathematical Tasks Using Flawed Thinking? [0.0]
  Chain-of-thought (CoT) prompting has become central to mathematical reasoning in large language models.<n>We investigate whether training on intentionally flawed reasoning traces can teach models to detect and recover from such errors.
  arXiv  Detail & Related papers  (2025-12-18T21:20:21Z)
• UQ: Assessing Language Models on Unsolved Questions [149.46593270027697]
  We introduce UQ, a testbed of 500 challenging, diverse questions sourced from Stack Exchange.<n>UQ is difficult and realistic by construction: unsolved questions are often hard and naturally arise when humans seek answers.<n>The top model passes UQ-validation on only 15% of questions, and preliminary human verification has already identified correct answers.
  arXiv  Detail & Related papers  (2025-08-25T01:07:59Z)
• When to retrain a machine learning model [0.0]
  A significant challenge in maintaining real-world machine learning models is responding to the continuous and unpredictable evolution of data.<n>We propose an uncertainty-based method that makes decisions by continually forecasting the evolution of model performance evaluated with a bounded metric.
  arXiv  Detail & Related papers  (2025-05-20T20:55:56Z)
• Climbing the Ladder of Reasoning: What LLMs Can-and Still Can't-Solve after SFT? [59.418994222096885]
  We conduct a detailed analysis of model performance on the AIME24 dataset.<n>We categorize questions into four tiers (Easy, Medium, Hard, and Extremely Hard)<n>We find that progression from Easy to Medium tier requires adopting an R1 reasoning style with minimal SFT-1K instances.<n>Exh-level questions present a fundamentally different challenge; they require unconventional problem-solving skills.
  arXiv  Detail & Related papers  (2025-04-16T03:39:38Z)
• MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations [90.07275414500154]
  We observe significant performance drops on MATH-P-Hard across various models.<n>We also raise concerns about a novel form of memorization where models blindly apply learned problem-solving skills.
  arXiv  Detail & Related papers  (2025-02-10T13:31:46Z)
• Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problems [47.753284211200665]
  We focus on understanding the usefulness of incorporating "error-correction" data directly into the pretraining stage. This data consists of erroneous solution steps immediately followed by their corrections. We show promising results: this type of pretrain data can help language models achieve higher reasoning accuracy.
  arXiv  Detail & Related papers  (2024-08-29T06:49:20Z)
• R-Tuning: Instructing Large Language Models to Say `I Don't Know' [66.11375475253007]
  Large language models (LLMs) have revolutionized numerous domains with their impressive performance but still face their challenges. Previous instruction tuning methods force the model to complete a sentence no matter whether the model knows the knowledge or not. We present a new approach called Refusal-Aware Instruction Tuning (R-Tuning) Experimental results demonstrate R-Tuning effectively improves a model's ability to answer known questions and refrain from answering unknown questions.
  arXiv  Detail & Related papers  (2023-11-16T08:45:44Z)
• TACRED Revisited: A Thorough Evaluation of the TACRED Relation Extraction Task [80.38130122127882]
  TACRED is one of the largest, most widely used crowdsourced datasets in Relation Extraction (RE) In this paper, we investigate the questions: Have we reached a performance ceiling or is there still room for improvement? We find that label errors account for 8% absolute F1 test error, and that more than 50% of the examples need to be relabeled.
  arXiv  Detail & Related papers  (2020-04-30T15:07:37Z)

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.