mlsanityData to 7 Oct 2026

LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints

arXiv 2410.06458 ↗First posted 9 Oct 2024cs.CL cs.AI cs.LGcited by 34

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

Abstract (arXiv)

Instruction following is a key capability for LLMs. However, recent studies have shown that LLMs often struggle with instructions containing multiple constraints (e.g. a request to create a social media post "in a funny tone" with "no hashtag"). Despite this, most evaluations focus solely on synthetic data. To address this, we introduce RealInstruct, the first benchmark designed to evaluate LLMs' ability to follow real-world multi-constrained instructions by leveraging queries real users asked AI assistants. We also investigate model-based evaluation as a cost-effective alternative to human annotation for this task. Our findings reveal that even the proprietary GPT-4 model fails to meet at least one constraint on over 21% of instructions, highlighting the limitations of state-of-the-art models. To address the performance gap between open-source and proprietary models, we propose the Decompose, Critique and Refine (DeCRIM) self-correction pipeline, which enhances LLMs' ability to follow constraints. DeCRIM works by decomposing the original instruction into a list of constraints and using a Critic model to decide when and where the LLM's response needs refinement. Our results show that DeCRIM improves Mistral's performance by 7.3% on RealInstruct and 8.0% on IFEval even with weak feedback. Moreover, we demonstrate that with strong feedback, open-source LLMs with DeCRIM can outperform GPT-4 on both benchmarks.

Key facts. Cited by 34 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 79 of its 101 references that are in the map. First posted on arXiv 9 Oct 2024.

Builds on

79 of 101 references.

Closest ancestors · 7

Same direction, newest first

  1. 2024Learning From Correctness Without Prompting Makes LLM Efficient Reasoner28 Mar
  2. 2023Large Language Models Cannot Self-Correct Reasoning Yet3 Oct
  3. 2023SCREWS: A Modular Framework for Reasoning with Revisions20 Sep
  4. 2023Recursion of Thought: A Divide-and-Conquer Approach to Multi-Context Reasoning with Language Models12 Jun
  5. 2023CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing19 May
  6. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May
  7. 2023Self-Refine: Iterative Refinement with Self-Feedback30 Mar

Other references · 72

  1. 2020Language Models are Few-Shot Learnerscited by 32,199
  2. 2023GPT-4 Technical Reportcited by 18,132
  3. 2022Training language models to follow instructions with human feedbackcited by 15,712
  4. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  5. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  6. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  7. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  8. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  9. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  10. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,462
  11. 2019HellaSwag: Can a Machine Really Finish Your Sentence?cited by 4,018
  12. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
38 more
  1. 2023Mistral 7Bcited by 3,065
  2. 2022Scaling Instruction-Finetuned Language Modelscited by 2,997
  3. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,674
  4. 2023Sparks of Artificial General Intelligence: Early experiments with GPT-4cited by 2,579
  5. 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,205
  6. 2022Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language modelscited by 1,970
  7. 2023Improving Factuality and Reasoning in Language Models through Multiagent Debatecited by 1,739
  8. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,696
  9. 2021Multitask Prompted Training Enables Zero-Shot Task Generalizationcited by 1,462
  10. 2023FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generationcited by 1,259
  11. 2024Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preferencecited by 1,151
  12. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,143
  13. 2022Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Taskscited by 858
  14. 2022Galactica: A Large Language Model for Sciencecited by 736
  15. 2021Cross-Task Generalization via Natural Language Crowdsourcing Instructionscited by 696
  16. 2024LLM Evaluators Recognize and Favor Their Own Generationscited by 651
  17. 2023Prometheus: Inducing Fine-grained Evaluation Capability in Language Modelscited by 558
  18. 2023Zephyr: Direct Distillation of LM Alignmentcited by 498
  19. 2024Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Modelscited by 431
  20. 2023TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answeringcited by 416
  21. 2022Self-critiquing models for assisting human evaluatorscited by 372
  22. 2024Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Modelscited by 301
  23. 2023State of What Art? A Call for Multi-Prompt LLM Evaluationcited by 299
  24. 2023Evaluating Large Language Models at Evaluating Instruction Followingcited by 290
  25. 2022Teaching Small Language Models to Reasoncited by 279
  26. 2022Generating Sequences by Learning to Self-Correctcited by 271
  27. 2024Debating with More Persuasive LLMs Leads to More Truthful Answerscited by 255
  28. 2024When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction of LLMscited by 230
  29. 2023A Stitch in Time Saves Nine: Detecting and Mitigating Hallucinations of LLMs by Validating Low-Confidence Generationcited by 202
  30. 2022Re3: Generating Longer Stories With Recursive Reprompting and Revisioncited by 193
  31. 2023Implicit Chain of Thought Reasoning via Knowledge Distillationcited by 165
  32. 2024Distilling System 2 into System 1cited by 152
  33. 2024Benchmarking Complex Instruction-Following with Multiple Constraints Compositioncited by 147
  34. 2024InFoBench: Evaluating Instruction Following Ability in Large Language Modelscited by 128
  35. 2023Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselvescited by 123
  36. 2023Table Meets LLM: Can Large Language Models Understand Structured Table Data? A Benchmark and Empirical Studycited by 116
  37. 2023FollowBench: A Multi-level Fine-grained Constraints Following Benchmark for Large Language Modelscited by 112
  38. 2023Branch-Solve-Merge Improves Large Language Model Evaluation and Generationcited by 104

50 most cited of 72 All 72 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 20; 2026: 14.

34 papers.

Sort:

By area:

  1. 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?cited by 215
  2. 2025SETS: Leveraging Self-Verification and Self-Correction for Improved Test-Time Scalingcited by 54
  3. 2025Thinking Machines: A Survey of LLM based Reasoning Strategiescited by 25
  4. 2026RubricEval: A Rubric-Level Meta-Evaluation Benchmark for LLM Judges in Instruction Followingcited by 19
  5. 2025Incentivizing Reasoning for Advanced Instruction-Following of Large Language Modelscited by 17
  6. 2025Lost at the Beginning of Reasoningcited by 15
  7. 2025Can Language Models Follow Multiple Turns of Entangled Instructions?cited by 14
  8. 2025FactSelfCheck: Fact-Level Black-Box Hallucination Detection for LLMscited by 14
  9. 2025Accelerating Earth Science Discovery via Multi-Agent LLM Systemscited by 13
  10. 2025MuSC: Improving Complex Instruction Following with Multi-granularity Self-Contrastive Trainingcited by 12
  11. 2025ReflectEvo: Improving Meta Introspection of Small LLMs by Learning Self-Reflectioncited by 11
  12. 2025RECAST: Expanding the Boundaries of LLMs' Complex Instruction Following with Multi-Constraint Datacited by 11
22 more
  1. 2025WildIFEval: Instruction Following in the Wildcited by 10
  2. 2025Revisiting Compositional Generalization Capability of Large Language Models Considering Instruction Following Abilitycited by 10
  3. 2025Are Checklists Really Useful for Automatic Evaluation of Generative Tasks?cited by 9
  4. 2026Revision or Re-Solving? Decomposing Second-Pass Gains in Multi-LLM Pipelinescited by 7
  5. 2025Revisiting the Reliability of Language Models in Instruction-Followingcited by 6
  6. 2026Retrieval-Augmented LLM Agents: Learning to Learn from Experiencecited by 6
  7. 2025Evaluating the Efficacy of LLM-Based Reasoning for Multiobjective HPC Job Schedulingcited by 3
  8. 2025Disambiguation-Centric Finetuning Makes Enterprise Tool-Calling LLMs More Realistic and Less Riskycited by 2
  9. 2025Self-Review Framework for Enhancing Instruction Following Capability of LLMcited by 1
  10. 2025CLARIFID: Improving Radiology Report Generation by Reinforcing Clinically Accurate Impressions and Enforcing Detailed Findingscited by 1
  11. 2026Enhancing LLM Instruction Following: An Evaluation-Driven Multi-Agentic Workflow for Prompt Instructions Optimizationcited by 1
  12. 2026SEIF: Self-Evolving Reinforcement Learning for Instruction Followingcited by 1
  13. 2026Same Evidence, Different Answers: Canonical-Context On-Policy Distillation for Multi-Turn Language Modelscited by 1
  14. 2025Critique-Guided Distillation for Robust Reasoning via Refinementcited by 0
  15. 2026Joint Consistency: A Unified Test-Time Aggregation Framework via Energy Minimizationcited by 0
  16. 2026When Cloud Agents Meet Device Agents: Lessons from Hybrid Multi-Agent Systemscited by 0
  17. 2026Bridging Auxiliary Constraints to Resolve Instruction Following in Large Reasoning Modelscited by 0
  18. 2026Instruction Stacking Collapse: A Benchmark and the Capability-Dependent Value of Prompt Compilationcited by 0
  19. 2026UNSPECIFIC: General Constraint Synthesis for Breaking Copy-and-Paste Shortcut in LLM Instruction Followingcited by 0
  20. 2026Large Language Models Can Follow Instructions, But Not Many at Once: Phase Transitions in Compositional Constraint Satisfactioncited by 0
  21. 2026OBJECTION! Lawyer Agents Mitigate Guilty Bias in Legal Judgment Predictioncited by 0
  22. 2026Exact Feedback Is Not Control: Evaluating Text-based Closed-Loop Revision in LLMscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Divide-Verify-Refine: Can LLMs Self-Align with Complex Instructions?
  2. 2024Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMs
  3. 2025Meeseeks: A Feedback-Driven, Iterative Self-Correction Benchmark evaluating LLMs' Instruction Following Capability
  4. 2025Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMs
  5. 2024Self-Judge: Selective Instruction Following with Alignment Self-Evaluation
  6. 2025Training with Pseudo-Code for Instruction Following
  7. 2026Reinforcing Step-level Reasoning for Effective Self-Correction in LLMs
  8. 2024Human-Instruction-Free LLM Self-Alignment with Limited Samples