# Two Failures of Self-Consistency in the Multi-Step Reasoning of LLMs arXiv: https://arxiv.org/abs/2305.14279 ยท HTML: https://mlsanity.com/paper/2305.14279/ ## 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. ## Placed in - LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/) ## Builds on (most cited first, 12 of 12) - [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020) - [What Makes Good In-Context Examples for GPT-3?](https://mlsanity.com/paper/2101.06804/) (2021) - [DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset](https://mlsanity.com/paper/1710.03957/) (2017) - [Learning To Retrieve Prompts for In-Context Learning](https://mlsanity.com/paper/2112.08633/) (2021) - [Evaluating the Factual Consistency of Abstractive Text Summarization](https://mlsanity.com/paper/1910.12840/) (2019) - [Measuring and Improving Consistency in Pretrained Language Models](https://mlsanity.com/paper/2102.01017/) (2021) - [Prompt Consistency for Zero-Shot Task Generalization](https://mlsanity.com/paper/2205.00049/) (2022) - [BeliefBank: Adding Memory to a Pre-Trained Language Model for a Systematic Notion of Belief](https://mlsanity.com/paper/2109.14723/) (2021) - [Measuring Reliability of Large Language Models through Semantic Consistency](https://mlsanity.com/paper/2211.05853/) (2022) - [P-Adapters: Robustly Extracting Factual Information from Language Models with Diverse Prompts](https://mlsanity.com/paper/2110.07280/) (2021) - [Factual Consistency of Multilingual Pretrained Language Models](https://mlsanity.com/paper/2203.11552/) (2022) - [Accurate, yet inconsistent? Consistency Analysis on Language Understanding Models](https://mlsanity.com/paper/2108.06665/) (2021) ## Built on by (most cited first, 30 of 36) - [A Survey on Large Language Model based Autonomous Agents](https://mlsanity.com/paper/2308.11432/) (2023) - [ChatDB: Augmenting LLMs with Databases as Their Symbolic Memory](https://mlsanity.com/paper/2306.03901/) (2023) - [Looking Inward: Language Models Can Learn About Themselves by Introspection](https://mlsanity.com/paper/2410.13787/) (2024) - [SAC3: Reliable Hallucination Detection in Black-Box Language Models via Semantic-aware Cross-check Consistency](https://mlsanity.com/paper/2311.01740/) (2023) - [Lyfe Agents: Generative agents for low-cost real-time social interactions](https://mlsanity.com/paper/2310.02172/) (2023) - [Does It Make Sense to Speak of Introspection in Large Language Models?](https://mlsanity.com/paper/2506.05068/) (2025) - [Optimas: Optimizing Compound AI Systems with Globally Aligned Local Rewards](https://mlsanity.com/paper/2507.03041/) (2025) - [Towards Evaluating AI Systems for Moral Status Using Self-Reports](https://mlsanity.com/paper/2311.08576/) (2023) - [AutoPenBench: Benchmarking Generative Agents for Penetration Testing](https://mlsanity.com/paper/2410.03225/) (2024) - [Self-Improvements in Modern Agentic Systems: A Survey](https://mlsanity.com/paper/2607.13104/) (2026) - [Consistency in Language Models: Current Landscape, Challenges, and Future Directions](https://mlsanity.com/paper/2505.00268/) (2025) - [Learning from Diverse Reasoning Paths with Routing and Collaboration](https://mlsanity.com/paper/2508.16861/) (2025) - [Evidence for Limited Metacognition in LLMs](https://mlsanity.com/paper/2509.21545/) (2025) - [Whose Facts Win? LLM Source Preferences under Knowledge Conflicts](https://mlsanity.com/paper/2601.03746/) (2026) - [NLI under the Microscope: What Atomic Hypothesis Decomposition Reveals](https://mlsanity.com/paper/2502.08080/) (2025) - [Failure Modes in LLM Systems: A System-Level Taxonomy for Reliable AI Applications](https://mlsanity.com/paper/2511.19933/) (2025) - [Towards an Understanding of Stepwise Inference in Transformers: A Synthetic Graph Navigation Model](https://mlsanity.com/paper/2402.07757/) (2024) - [In-Context Learning May Not Elicit Trustworthy Reasoning: A-Not-B Errors in Pretrained Language Models](https://mlsanity.com/paper/2409.15454/) (2024) - [BioAgents: Democratizing Bioinformatics Analysis with Multi-Agent Systems](https://mlsanity.com/paper/2501.06314/) (2025) - [Decompose and Compare Consistency: Measuring VLMs' Answer Reliability via Task-Decomposition Consistency Comparison](https://mlsanity.com/paper/2407.07840/) (2024) - [Learning to Explore with Parameter-Space Noise: A Deep Dive into Parameter-Space Noise for Reinforcement Learning with Verifiable Rewards](https://mlsanity.com/paper/2602.02555/) (2026) - [Existing LLMs Are Not Self-Consistent For Simple Tasks](https://mlsanity.com/paper/2506.18781/) (2025) - [DCR-Consistency: Divide-Conquer-Reasoning for Consistency Evaluation and Improvement of Large Language Models](https://mlsanity.com/paper/2401.02132/) (2024) - [RORA: Robust Free-Text Rationale Evaluation](https://mlsanity.com/paper/2402.18678/) (2024) - [Preference Consistency Matters: Enhancing Preference Learning in Language Models with Automated Self-Curation of Training Corpora](https://mlsanity.com/paper/2408.12799/) (2024) - [Found in Translation: Measuring Multilingual LLM Consistency as Simple as Translate then Evaluate](https://mlsanity.com/paper/2505.21999/) (2025) - [Self-Blinding and Counterfactual Self-Simulation Mitigate Biases and Sycophancy in Large Language Models](https://mlsanity.com/paper/2601.14553/) (2026) - [TextResNet: Decoupling and Routing Optimization Signals in Compound AI Systems via Deep Residual Tuning](https://mlsanity.com/paper/2602.08306/) (2026) - [Examining the Behavior of LLM Architectures Within the Framework of Standardized National Exams in Brazil](https://mlsanity.com/paper/2408.05035/) (2024) - [Sensitivity of Generative VLMs to Semantically and Lexically Altered Prompts](https://mlsanity.com/paper/2410.13030/) (2024) ## 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.