mlsanityData to 14 Oct 2026

Boosting LLM Reasoning via Spontaneous Self-Correction

arXiv 2506.06923 ↗First posted 7 Jun 2025cs.AIcited by 18

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

While large language models (LLMs) have demonstrated remarkable success on a broad range of tasks, math reasoning remains a challenging one. One of the approaches for improving math reasoning is self-correction, which designs self-improving loops to let the model correct its own mistakes. However, existing self-correction approaches treat corrections as standalone post-generation refinements, relying on extra prompt and system designs to elicit self-corrections, instead of performing real-time, spontaneous self-corrections in a single pass. To address this, we propose SPOC, a spontaneous self-correction approach that enables LLMs to generate interleaved solutions and verifications in a single inference pass, with generation dynamically terminated based on verification outcomes, thereby effectively scaling inference time compute. SPOC considers a multi-agent perspective by assigning dual roles -- solution proposer and verifier -- to the same model. We adopt a simple yet effective approach to generate synthetic data for fine-tuning, enabling the model to develop capabilities for self-verification and multi-agent collaboration. We further improve its solution proposal and verification accuracy through online reinforcement learning. Experiments on mathematical reasoning benchmarks show that SPOC significantly improves performance. Notably, SPOC boosts the accuracy of Llama-3.1-8B and 70B Instruct models, achieving gains of 8.8% and 11.6% on MATH500, 10.0% and 20.0% on AMC23, and 3.3% and 6.7% on AIME24, respectively.

Key facts. Cited by 18 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 32 of its 44 references that are in the map. First posted on arXiv 7 Jun 2025.

Builds on

32 of 44 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025Self-rewarding correction for mathematical reasoning26 Feb
  2. 2024Training Language Models to Self-Correct via Reinforcement Learning19 Sep
  3. 2024Critic-CoT: Boosting the reasoning abilities of large language model via Chain-of-thoughts Critic29 Aug
  4. 2024AutoPRM: Automating Procedural Supervision for Multi-Step Reasoning via Controllable Question Decomposition18 Feb
  5. 2023Large Language Models Cannot Self-Correct Reasoning Yet3 Oct
  6. 2023REFINER: Reasoning Feedback on Intermediate Representations4 Apr

Other references · 26

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  4. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  5. 2023Let's Verify Step by Stepcited by 3,789
  6. 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,224
  7. 2023AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversationcited by 1,988
  8. 2024Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvementcited by 1,090
  9. 2023Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debatecited by 1,011
  10. 2017Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithmcited by 734
  11. 2023RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignmentcited by 686
  12. 2024Generative Verifiers: Reward Modeling as Next-Token Predictioncited by 480
14 more
  1. 2022Self-critiquing models for assisting human evaluatorscited by 380
  2. 2023Iterative Preference Learning from Human Feedback: Bridging Theory and Practice for RLHF under KL-Constraintcited by 350
  3. 2022Generating Sequences by Learning to Self-Correctcited by 277
  4. 2024Recursive Introspection: Teaching Language Model Agents How to Self-Improvecited by 165
  5. 2024Orca-Math: Unlocking the potential of SLMs in Grade School Mathcited by 151
  6. 2025Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thoughtcited by 109
  7. 2024Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMscited by 108
  8. 2023RL4F: Generating Natural Language Feedback with Reinforcement Learning for Repairing Model Outputscited by 105
  9. 2024GLoRe: When, Where, and How to Improve LLM Reasoning via Global and Local Refinementscited by 103
  10. 2025DeepSeek-R1 Thoughtology: Let's think about LLM Reasoningcited by 92
  11. 2024MALT: Improving Reasoning with Multi-Agent LLM Trainingcited by 76
  12. 2024Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problemscited by 50
  13. 2024The Perfect Blend: Redefining RLHF with Mixture of Judgescited by 43
  14. 2025S2R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learningcited by 39

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 14.

18 papers.

Sort:

By area:

  1. 2025Failure Makes the Agent Stronger: Enhancing Accuracy through Structured Reflection for Reliable Tool Interactionscited by 12
  2. 2026Token-Level LLM Collaboration via FusionRoutecited by 9
  3. 2025Population-Evolve: a Parallel Sampling and Evolutionary Method for LLM Math Reasoningcited by 6
  4. 2026Re2: Unlocking LLM Reasoning via Reinforcement Learning with Re-solvingcited by 6
  5. 2026ThinkTwice: Jointly Optimizing Large Language Models for Reasoning and Self-Refinementcited by 5
  6. 2025Putting on the Thinking Hats: A Survey on Chain of Thought Fine-tuning from the Perspective of Human Reasoning Mechanismcited by 4
  7. 2026CoT-Seg: Rethinking Segmentation with Chain-of-Thought Reasoning and Self-Correctioncited by 3
  8. 2026The Self-Correction Illusion: Role Relabeling Gates Explicit Error Flagging in Large Language Modelscited by 2
  9. 2025Learning Deliberately, Acting Intuitively: Unlocking Test-Time Reasoning in Multimodal LLMscited by 1
  10. 2026Beyond In-Domain Detection: SpikeScore for Cross-Domain Hallucination Detectioncited by 1
  11. 2026VeryTrace: Verifying Reasoning Traces through Compilable Formalism and Structured Verificationcited by 1
  12. 2026Mixture of Debaters: Learn to Debate at Architectural Level in Multi-Agent Reasoningcited by 1
6 more
  1. 2026SVR-R1: Bootstrapping Multi-modal Reasoning with Self-verification in Reinforcement Learningcited by 1
  2. 2026The Two-Stage Decision-Sampling Hypothesis: Understanding the Emergence of Self-Reflection in RL-Trained LLMscited by 0
  3. 2026Learning When to Act: Interval-Aware Reinforcement Learning with Predictive Temporal Structurecited by 0
  4. 2026ReSum: Synergizing LLM Reasoning and Summarization with Reinforcement Learningcited by 0
  5. 2026AMTFV: Agentic Mathematical Tool-Flow Verification for LLM Self-Correctioncited by 0
  6. 2026Reinforcing Step-level Reasoning for Effective Self-Correction in LLMscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024S^3cMath: Spontaneous Step-level Self-correction Makes Large Language Models Better Mathematical Reasoners
  2. 2024Embedding Self-Correction as an Inherent Ability in Large Language Models for Enhanced Mathematical Reasoning
  3. 2025Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMs
  4. 2024Small Language Models Need Strong Verifiers to Self-Correct Reasoning
  5. 2025Incentivizing LLMs to Self-Verify Their Answers
  6. 2025A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning
  7. 2025From Solving to Verifying: A Unified Objective for Robust Reasoning in LLMs
  8. 2026SAGE: Multi-Agent Self-Evolution for LLM Reasoning