mlsanityData to 14 Oct 2026

Two Failures of Self-Consistency in the Multi-Step Reasoning of LLMs

arXiv 2305.14279 ↗First posted 23 May 2023cs.CLcited by 36

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

Abstract (arXiv)

Large language models (LLMs) have achieved widespread success on a variety of in-context few-shot tasks, but this success is typically evaluated via correctness rather than consistency. We argue that self-consistency is an important criteria for valid multi-step reasoning in tasks where the solution is composed of the answers to multiple sub-steps. We propose two types of self-consistency that are particularly important for multi-step reasoning -- hypothetical consistency (a model's ability to predict what its output would be in a hypothetical other context) and compositional consistency (consistency of a model's final outputs when intermediate sub-steps are replaced with the model's outputs for those steps). We demonstrate that multiple variants of the GPT-3/-4 models exhibit poor consistency rates across both types of consistency on a variety of tasks.

Key facts. Cited by 36 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 12 of its 23 references that are in the map. First posted on arXiv 23 May 2023.

Builds on

12 of 23 references.

References · 12

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2021What Makes Good In-Context Examples for GPT-3?cited by 1,195
  3. 2017DailyDialog: A Manually Labelled Multi-turn Dialogue Datasetcited by 652
  4. 2021Learning To Retrieve Prompts for In-Context Learningcited by 648
  5. 2019Evaluating the Factual Consistency of Abstractive Text Summarizationcited by 584
  6. 2021Measuring and Improving Consistency in Pretrained Language Modelscited by 364
  7. 2022Prompt Consistency for Zero-Shot Task Generalizationcited by 67
  8. 2021BeliefBank: Adding Memory to a Pre-Trained Language Model for a Systematic Notion of Beliefcited by 57
  9. 2022Measuring Reliability of Large Language Models through Semantic Consistencycited by 38
  10. 2021P-Adapters: Robustly Extracting Factual Information from Language Models with Diverse Promptscited by 22
  11. 2022Factual Consistency of Multilingual Pretrained Language Modelscited by 15
  12. 2021Accurate, yet inconsistent? Consistency Analysis on Language Understanding Modelscited by 13

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 5; 2024: 10; 2025: 13; 2026: 8.

36 papers.

Sort:

By area:

  1. 2023A Survey on Large Language Model based Autonomous Agentscited by 2,242
  2. 2023ChatDB: Augmenting LLMs with Databases as Their Symbolic Memorycited by 140
  3. 2024Looking Inward: Language Models Can Learn About Themselves by Introspectioncited by 113
  4. 2023SAC3: Reliable Hallucination Detection in Black-Box Language Models via Semantic-aware Cross-check Consistencycited by 90
  5. 2023Lyfe Agents: Generative agents for low-cost real-time social interactionscited by 49
  6. 2025Does It Make Sense to Speak of Introspection in Large Language Models?cited by 25
  7. 2025Optimas: Optimizing Compound AI Systems with Globally Aligned Local Rewardscited by 24
  8. 2023Towards Evaluating AI Systems for Moral Status Using Self-Reportscited by 22
  9. 2024AutoPenBench: Benchmarking Generative Agents for Penetration Testingcited by 21
  10. 2026Self-Improvements in Modern Agentic Systems: A Surveycited by 19
  11. 2025Consistency in Language Models: Current Landscape, Challenges, and Future Directionscited by 14
  12. 2025Learning from Diverse Reasoning Paths with Routing and Collaborationcited by 12
24 more
  1. 2025Evidence for Limited Metacognition in LLMscited by 12
  2. 2026Whose Facts Win? LLM Source Preferences under Knowledge Conflictscited by 12
  3. 2025NLI under the Microscope: What Atomic Hypothesis Decomposition Revealscited by 11
  4. 2025Failure Modes in LLM Systems: A System-Level Taxonomy for Reliable AI Applicationscited by 10
  5. 2024Towards an Understanding of Stepwise Inference in Transformers: A Synthetic Graph Navigation Modelcited by 9
  6. 2024In-Context Learning May Not Elicit Trustworthy Reasoning: A-Not-B Errors in Pretrained Language Modelscited by 8
  7. 2025BioAgents: Democratizing Bioinformatics Analysis with Multi-Agent Systemscited by 8
  8. 2024Decompose and Compare Consistency: Measuring VLMs' Answer Reliability via Task-Decomposition Consistency Comparisoncited by 7
  9. 2026Learning to Explore with Parameter-Space Noise: A Deep Dive into Parameter-Space Noise for Reinforcement Learning with Verifiable Rewardscited by 6
  10. 2025Existing LLMs Are Not Self-Consistent For Simple Taskscited by 5
  11. 2024DCR-Consistency: Divide-Conquer-Reasoning for Consistency Evaluation and Improvement of Large Language Modelscited by 4
  12. 2024RORA: Robust Free-Text Rationale Evaluationcited by 4
  13. 2024Preference Consistency Matters: Enhancing Preference Learning in Language Models with Automated Self-Curation of Training Corporacited by 3
  14. 2025Found in Translation: Measuring Multilingual LLM Consistency as Simple as Translate then Evaluatecited by 3
  15. 2026Self-Blinding and Counterfactual Self-Simulation Mitigate Biases and Sycophancy in Large Language Modelscited by 3
  16. 2026TextResNet: Decoupling and Routing Optimization Signals in Compound AI Systems via Deep Residual Tuningcited by 3
  17. 2024Examining the Behavior of LLM Architectures Within the Framework of Standardized National Exams in Brazilcited by 2
  18. 2024Sensitivity of Generative VLMs to Semantically and Lexically Altered Promptscited by 2
  19. 2025Can LLMs Assist Computer Education? an Empirical Case Study of DeepSeekcited by 1
  20. 2025PepThink-R1: LLM for Interpretable Cyclic Peptide Optimization with CoT SFT and Reinforcement Learningcited by 1
  21. 2026LendNova: Towards Automated Credit Risk Assessment with Language Modelscited by 1
  22. 2026Seeing the Reasoning: How LLM Rationales Influence User Trust and Decision-Making in Factual Verification Taskscited by 1
  23. 2026Empirical Bayes Conformal Prediction for Vision and Language Modelscited by 1
  24. 2025Explaining Hitori Puzzles: Neurosymbolic Proof Staging for Sequential Decisionscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Semantic Self-Consistency: Enhancing Language Model Reasoning via Semantic Weighting
  2. 2025Self-Consistency Is Losing Its Edge: Diminishing Returns and Rising Costs in Modern LLMs
  3. 2024Lachesis: Predicting LLM Inference Accuracy using Structural Properties of Reasoning Paths
  4. 2026Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning
  5. 2026Opening the Black Box: A Survey on the Mechanisms of Multi-Step Reasoning in Large Language Models
  6. 2025A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning
  7. 2024Internal Consistency and Self-Feedback in Large Language Models: A Survey
  8. 2025Bridging Internal Probability and Self-Consistency for Effective and Efficient LLM Reasoning