Self-Consistency in Language Models

Follow this topic Ask about this topic

This topic studies how language models’ answers and reasoning vary across repeated generations, prompts, and related inputs, and whether consistency can improve reliability. Common approaches sample multiple reasoning paths and aggregate them with majority voting, semantic clustering, ranking, or verification; adaptive sampling and stopping rules aim to reduce the cost. Other work trains models to align their explanations and behavior or tests consistency in settings such as safety, preferences, and evaluation. A central tension is that agreement can improve performance but may also reinforce correlated errors, so consistency alone is not proof of correctness.

Latest papers

49 papers10 papers in the last four weeks, up 67% on the four weeks before. 0.1% of all new papers.

May 8, 2026

Reliable Chain-of-Thought via Prefix Consistency

Large Language Models often improve accuracy on reasoning tasks by sampling multiple Chain-of-Thought (CoT) traces and aggregating them with majority voting (MV), a test-time technique called self-consistency. When we truncate a CoT partway through and regenerate the remainder, we observe that traces with correct answers reproduce their original answer more often than traces with wrong answers. We use this difference as a reliability signal, prefix consistency, that weights each candidate answer by how often it reappears under regeneration. It requires no access to token log-probabilities or self-rating prompts. Across five reasoning models and four math and science benchmarks, prefix consistency is the best correctness predictor in most settings, and reweighting votes by it reaches Standard MV plateau accuracy at up to 21x fewer tokens (median 4.6x). Our code is available at https://github.com/naoto-iwase/prefix-consistency.

Naoto Iwase, Yuki Ichihara, Mohammad Atif Quamar +1

Nagoya University · Nara Institute of Science and Technology · Mohamed bin Zayed University of Artificial Intelligence +1

Self-Consistency Decoding Reasoning Consistency in Language Models LLM Reliability Efficient Language Model Inference Self-Consistency in Language Models LLM Reasoning+2

Apr 21, 2026

Binary Agreement for Efficient Unsupervised Confidence Calibration of Reasoning LLMs

Reasoning language models often lack the calibrated confidence needed for reliable deployment, particularly when correctness labels are unavailable and sampling budgets are limited. We introduce Binary Agreement for Self-consistency Calibration (BASC), which uses self-consistency as an unlabeled proxy for correctness without constructing costly self-consistency targets for each calibration query. Our key observation is that agreement between an anchor response and a single independent probe is a Bernoulli observation whose conditional mean is the anchor's latent self-consistency. Pooling these observations across unlabeled queries, BASC supports both recalibrating existing confidence scores and learning more discriminative scores before recalibration, using as few as two responses per calibration query and only one at deployment. The use of binary isotonic recalibration provides a distribution-free calibration guarantee with respect to latent self-consistency. Across 5 tasks and 9 reasoning models, BASC substantially improves calibration with respect to correctness.

Thomas Zollo, Jimmy Wang, Richard Zemel

Columbia University

Confidence Estimation in Language Models Language Model Calibration LLM Reliability LLM Uncertainty Estimation Self-Consistency in Language Models Large Reasoning Models

Apr 19, 2026

Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning

Self-consistency (SC) is a popular technique for improving the reasoning accuracy of large language models by aggregating multiple sampled outputs, but it comes at a high computational cost due to extensive sampling. We introduce a hybrid ensembling approach that leverages the complementary strengths of two distinct modes of reasoning: Chain-of-Thought (CoT) and Program-of-Thought (PoT). We describe a general framework for combining these two forms of reasoning in self-consistency, as well as particular strategies for both full sampling and early-stopping. We show that CoT-PoT ensembling not only improves overall accuracy, but also drastically reduces the number of samples required for SC by a factor of 9.3x. In particular, the majority of tasks (78.6%) can be addressed with only two samples, which has not been possible with any prior SC methods.

Raman Saparkhan, Majd Hawasly, Md Rizwan Parvez +1

Carnegie Mellon University Qatar · Qatar Computing Research Institute

LLM Inference Efficiency Self-Consistency Decoding CoT Reasoning Efficient Language Model Reasoning Reasoning Consistency in Language Models Language Model Ensembles+3

Apr 14, 2026

Efficient Best-Of-N with Radial Consensus Score

Solving complex problems with large language models (LLMs) often requires selecting the best answer from multiple candidate responses (best-of-

N

), yet identifying the most reliable one remains challenging, particularly when correctness does not align with majority agreement. Existing approaches based on voting or probability scoring rely on surface-level signals (e.g., frequency or token-level likelihood), often missing high-quality rare answers and ignoring the geometric structure of candidate representations, where correctness may emerge as semantic consistency rather than surface agreement. To address these limitations, we introduce Radial Consensus Score (RCS), a simple, efficient, and training-free method for best-of-

N

selection. RCS models semantic consensus by computing a weighted Frechet mean (semantic center) of answer embeddings and ranking candidates by their radial distance to this center. This enables it to recover semantically coherent correct answers despite lexical fragmentation, even when no single surface form holds a plurality under exact-match voting. Importantly, RCS offers a flexible framework that supports multiple weighting schemes, enabling the integration of arbitrary signals, such as semantic agreement or model confidence, with the position of the center adapting accordingly. Extensive experiments across five open-weight models and various benchmarks demonstrate that RCS matches or outperforms strong baselines at every sampling budget, with the largest gains when correct answers are lexically fragmented. Moreover, RCS serves as an effective drop-in replacement for majority voting in multi-agent debate and remains effective in black-box scenarios. These results suggest that geometric consensus can serve as a scalable and reliable alternative to voting-based methods for answer selection. Code is publicly available at https://github.com/manhitv/RCS.

Manh Nguyen, Sunil Gupta, Hung Le

Applied Artificial Intelligence Initiative, Deakin University, Australia

Best-of-N Sampling LLM Reranking Rank Aggregation Self-Consistency in Language Models

Apr 13, 2026

Consistency of AI-Generated Exercise Prescriptions: A Repeated Generation Study Using a Large Language Model

Background: Large language models (LLMs) have been explored as tools for generating personalized exercise prescriptions, yet the consistency of outputs under identical conditions remains insufficiently examined. Objective: This study evaluated the intra-model consistency of LLM-generated exercise prescriptions using a repeated generation design. Methods: Six clinical scenarios were used to generate exercise prescriptions using Gemini 2.5 Flash (20 outputs per scenario; total n = 120). Consistency was assessed across three dimensions: (1) semantic consistency using SBERT-based cosine similarity, (2) structural consistency based on the FITT principle using an AI-as-a-judge approach, and (3) safety expression consistency, including inclusion rates and sentence-level quantification. Results: Semantic similarity was high across scenarios (mean cosine similarity: 0.879-0.939), with greater consistency in clinically constrained cases. Frequency showed consistent patterns, whereas variability was observed in quantitative components, particularly exercise intensity. Unclassifiable intensity expressions were observed in 10-25% of resistance training outputs. Safety-related expressions were included in 100% of outputs; however, safety sentence counts varied significantly across scenarios (H=86.18, p less than 0.001), with clinical cases generating more safety expressions than healthy adult cases. Conclusions: LLM-generated exercise prescriptions demonstrated high semantic consistency but showed variability in key quantitative components. Reliability depends substantially on prompt structure, and additional structural constraints and expert validation are needed before clinical deployment.

Kihyuk Lee

Data Convergence Team, Office of Hospital Information, Seoul National University Bundang Hospital, Seongnam, Republic of Korea

Self-Consistency in Language Models Medical LLMs Clinical Language Model Evaluation

Apr 13, 2026

Frugal Knowledge Graph Construction with Local LLMs: A Zero-Shot Pipeline, Self-Consistency and Wisdom of Artificial Crowds

This paper presents an empirical study of a multi-model zero-shot pipeline for knowledge graph construction and exploitation, executed entirely through local inference on consumer-grade hardware. We propose a reproducible evaluation framework integrating two external benchmarks (DocRED, HotpotQA), WebQuestionsSP-style synthetic data, and the RAGAS evaluation framework in an automated pipeline. On 500 document-level relations, our system achieves an F1 of 0.70

\pm

0.041 in zero-shot, compared to 0.80 for supervised DREEAM. Text-to-query achieves an accuracy of 0.80

\pm

0.06 on 200 samples. Multi-hop reasoning achieves an Exact Match (EM) of 0.46

\pm

0.04 on 500 HotpotQA questions, with a RAGAS faithfulness of 0.96

\pm

0.04 on 50 samples. Beyond the pipeline, we study diversity mechanisms for difficult multi-hop reasoning. On 181 questions unsolvable at zero temperature, self-consistency (k=5, T =0.7) recovers up to 23% EM with a single Mixture-of-Experts (MoE) model, but the cross-model oracle (3 architectures x 5 samples) reaches 46.4%. We highlight an agreement paradox: strong consensus among samples signals collective hallucination rather than a reliable answer, echoing the work of Moussa{ï}d et al. on the wisdom of crowds. Extending to the full pipeline (500 questions), self-consistency (k=3) raises EM from 0.46 to 0.48

\pm

0.04. A confidence-routing cascade mechanism (Phi-4

\rightarrow

GPT-OSS, k=5) achieves an EM of 0.55

\pm

0.04, the best result obtained, with 45.4% of questions rerouted. Finally, we show that V3 prompt engineering applied to other models does not reproduce the gains observed with Gemma-4, confirming the specific prompt/model interaction. The entire system runs in

\sim

5 h on a single RTX 3090, without any training, for an estimated carbon footprint of 0.09 kg CO2 eq.

Pierre Jourlin

LIA · Avignon Université, Laboratoire d’Informatique d’Avignon (LIA)

KG Construction Knowledge Graph Question Answering Multi-Hop QA Language Model Ensembles Self-Consistency in Language Models Hallucination in Language Models

Apr 3, 2026

The Impact of LLM Self-Consistency and Reasoning Effort on Automated Scoring Accuracy and Cost

Strategic model selection and reasoning settings are more effective than ensembling for optimizing automated scoring with large language models (LLMs). We examined self-consistency (intra-model majority voting) and reasoning effort for scoring conversation-based assessment items in high school mathematics, evaluating 900 student conversations against human-scored ground truths using frontier and low-cost models from OpenAI and Google. Temperature sampling significantly improved accuracy over deterministic calls, but increasing ensemble size (j = 1 to 7) produced no significant gains. Higher reasoning effort showed a significant positive linear trend with scoring accuracy, though the benefit varied by model family. An efficiency frontier analysis identified Gemini 3.1 Pro Preview at low reasoning as the most accurate but costly configuration; GPT-5.4 Nano and Mini with no reasoning offered the best cost-performance balance.

Scott Frohn

Khan Academy, Mountain View, CA, USA

Mathematics Education Automated Evaluation Self-Consistency in Language Models LLM Evaluation

Apr 2, 2026

Do We Need Frontier Models to Verify Mathematical Proofs?

Advances in training, post-training, and inference-time methods have enabled frontier reasoning models to win gold medals in math competitions and settle challenging open problems. Gaining trust in the responses of these models requires that natural language proofs be checked for errors. LLM judges are increasingly being adopted to meet the growing demand for evaluating such proofs. While verification is considered easier than generation, what model capability does reliable verification actually require? We systematically evaluate four open-source and two frontier LLMs on datasets of human-graded natural language proofs of competition-level problems. We consider two key metrics: verifier accuracy and self-consistency (the rate of agreement across repeated judgments on the same proof). We observe that smaller open-source models are only up to ~10% behind frontier models in accuracy but they are up to ~25% more inconsistent. Furthermore, we see that verifier accuracy is sensitive to prompt choice across all models. We then demonstrate that the smaller models, in fact, do possess the mathematical capabilities to verify proofs at the level of frontier models, but they struggle to reliably elicit these capabilities with general judging prompts. Through an LLM-guided prompt search, we synthesize an ensemble of specialized prompts that overcome the specific failure modes of smaller models, boosting their performance by up to 9.1% in accuracy and 15.9% in self-consistency. These gains are realized across models and datasets, allowing models like Qwen3.5-35B to perform on par with frontier models such as Gemini 3.1 Pro for proof verification.

Aaditya Naik, Guruprerana Shabadi, Rajeev Alur +1

University of Pennsylvania, U.S.A.

LLM Input Optimization Reasoning Verification LLM Reasoning LLM Evaluation LLM Prompting Self-Consistency in Language Models

Apr 1, 2026

Beyond Symbolic Solving: Multi Chain-of-Thought Voting for Geometric Reasoning in Large Language Models

Geometric Problem Solving (GPS) remains at the heart of enhancing mathematical reasoning in large language models because it requires the combination of diagrammatic understanding, symbolic manipulation and logical inference. In existing literature, researchers have chiefly focused on synchronising the diagram descriptions with text literals and solving the problem. In this vein, they have either taken a neural, symbolic or neuro-symbolic approach. But this solves only the first two of the requirements, namely diagrammatic understanding and symbolic manipulation, while leaving logical inference underdeveloped. The logical inference is often limited to one chain-of-thought (CoT). To address this weakness in hitherto existing models, this paper proposes MARS-GPS, that generates multiple parallel reasoning rollouts augmented with Python code execution for numerical verification, ranks them using token-level entropy as a confidence signal, and aggregates answers through a multi-stage voting and self-verification pipeline. Empirical results show that MARS-GPS with 8 parallel rollouts achieves 88.8% on Geometry3K, a nearly +11% improvement over the prior state-of-the-art, with accuracy scaling consistently as the number of rollouts increases from 1 to 16 (+6.0% on ablation subset). We provide our code and data in an anonymous repository: https://anonymous.4open.science/r/MARS-GPS-DE55.

Md. Abu Bakor Siddique, Shahrin Hossain, Sadman Ahmed Siam +3

Systems and Software Lab (SSL), Department of Computer Science and Engineering, Islamic University of Technology, Dhaka, Bangladesh

LLM Mathematical Reasoning Self-Consistency in Language Models

Mar 19, 2026

How Uncertainty Estimation Scales with Sampling in Reasoning Models

Uncertainty estimation is critical for deploying reasoning language models, yet remains poorly understood under extended chain-of-thought reasoning. We study parallel sampling as a fully black-box approach using verbalized confidence and self-consistency. Across three reasoning models and 17 tasks spanning mathematics, STEM, and humanities, we characterize how these signals scale. Both self-consistency and verbalized confidence scale in reasoning models, but self-consistency exhibits lower initial discrimination and lags behind verbalized confidence under moderate sampling. Most uncertainty gains, however, arise from signal combination: with just two samples, a hybrid estimator improves AUROC by up to

+12

on average and already outperforms either signal alone even when scaled to much larger budgets, after which returns diminish. These effects are domain-dependent: in mathematics, the native domain of RLVR-style post-training, reasoning models achieve higher uncertainty quality and exhibit both stronger complementarity and faster scaling than in STEM or humanities.

Maksym Del, Markus Kängsepp, Marharyta Domnich +4

Institute of Computer Science, University of Tartu

LLM Uncertainty Estimation Uncertainty Quantification Confidence Estimation in Language Models Confidence Estimation Large Reasoning Models Verbalized Confidence Estimation+1

Mar 18, 2026

CoVerRL: Breaking the Consensus Trap in Label-Free Reasoning via Generator-Verifier Co-Evolution

Label-free reinforcement learning enables large language models to improve reasoning capabilities without ground-truth supervision, typically by treating majority-voted answers as pseudo-labels. However, we identify a critical failure mode: as training maximizes self-consistency, output diversity collapses, causing the model to confidently reinforce systematic errors that evade detection. We term this the consensus trap. To escape it, we propose CoVerRL, a framework where a single model alternates between generator and verifier roles, with each capability bootstrapping the other. Majority voting provides noisy but informative supervision for training the verifier, while the improving verifier progressively filters self-consistent errors from pseudo-labels. This co-evolution creates a virtuous cycle that maintains high reward accuracy throughout training. Experiments across Qwen and Llama model families demonstrate that CoVerRL outperforms label-free baselines by 4.7-5.9% on mathematical reasoning benchmarks. Moreover, self-verification accuracy improves from around 55% to over 85%, confirming that both capabilities genuinely co-evolve.

Teng Pan, Yuchen Yan, Zixuan Wang +6

Zhejiang University · Baidu Inc.

Language Model Self-Verification RL for Language Model Reasoning LLM Reasoning Self-Training for Language Models Self-Consistency in Language Models

Mar 17, 2026

Dual Consensus: Escaping from Spurious Majority in Unsupervised RLVR via Two-Stage Vote Mechanism

Current label-free RLVR approaches for large language models (LLMs), such as TTRL and Self-reward, have demonstrated effectiveness in improving the performance of LLMs on complex reasoning tasks. However, these methods rely heavily on accurate pseudo-label estimation and converge on spurious yet popular answers, thereby trapping in a dominant mode and limiting further improvements. Building on this, we propose Dual Consensus Reinforcement Learning (DCRL), a novel self-supervised training method which is capable of generating more reliable learning signals through a two-stage consensus mechanism. The model initially acts as an anchor, producing dominant responses; then it serves as an explorer, generating diverse auxiliary signals via a temporary unlearning process. The final training target is derived from the harmonic mean of these two signal sets. Notably, the process operates entirely without external models or supervision. Across eight benchmarks and diverse domains, DCRL consistently improves Pass@1 over majority vote while yielding more stable training dynamics. These results demonstrate that DCRL establishes a scalable path toward stronger reasoning without labels.

Kaixuan Du, Meng Cao, Hang Zhang +3

School of Automation Science and Electrical Engineering, Beihang University

RL for Language Model Reasoning Self-Training for Language Models Reinforcement Learning with Verifiable Rewards Self-Consistency in Language Models

Mar 8, 2026

Toward Epistemic Stability: Engineering Consistent Procedures for Industrial LLM Hallucination Reduction

Hallucinations in large language models (LLMs) are outputs that are syntactically coherent but factually incorrect or contextually inconsistent. They are persistent obstacles in high-stakes industrial settings such as engineering design, enterprise resource planning, and IoT telemetry platforms. We present and compare five prompt engineering strategies intended to reduce the variance of model outputs and move toward repeatable, grounded results without modifying model weights or creating complex validation models. These methods include: (M1) Iterative Similarity Convergence, (M2) Decomposed Model-Agnostic Prompting, (M3) Single-Task Agent Specialization, (M4) Enhanced Data Registry, and (M5) Domain Glossary Injection. Each method is evaluated against an internal baseline using an LLM-as-Judge framework over 100 repeated runs per method (same fixed task prompt, stochastic decoding at tau = 0.7. Under this evaluation setup, M4 (Enhanced Data Registry) received ``Better'' verdicts in all 100 trials; M3 and M5 reached 80% and 77% respectively; M1 reached 75%; and M2 was net negative at 34% when compared to single shot prompting with a modern foundation model. We then developed enhanced version 2 (v2) implementations and assessed them on a 10-trial verification batch; M2 recovered from 34% to 80%, the largest gain among the four revised methods. We discuss how these strategies help overcome the non-deterministic nature of LLM results for industrial procedures, even when absolute correctness cannot be guaranteed. We provide pseudocode, verbatim prompts, and batch logs to support independent assessment.

Brian Freeman, Adam Kicklighter, Matt Erdman +1

Trane Technologies, Davidson NC, USA

LLM Hallucination Mitigation LLM Prompting LLM Reliability Self-Consistency in Language Models

Mar 2, 2026

nchellwig at SemEval-2026 Task 3: Self-Consistent Structured Generation (SCSG) for Dimensional Aspect-Based Sentiment Analysis using Large Language Models

We present Self-Consistent Structured Generation (SCSG) for Dimensional Aspect-Based Sentiment Analysis in SemEval-2026 Task 3 (Track A). SCSG enhances prediction reliability by executing a LoRA-adapted large language model multiple times per instance, retaining only tuples that achieve a majority consensus across runs. To mitigate the computational overhead of multiple forward passes, we leverage vLLM's PagedAttention mechanism for efficient key--value cache reuse. Evaluation across 6 languages and 8 language--domain combinations demonstrates that self-consistency with 15 executions yields statistically significant improvements over single-inference prompting, with our system (leveraging Gemma 3) ranking in the top seven across all settings, achieving second place on three out of four English subsets and first place on Tatar-Restaurant for DimASTE.

Nils Constantin Hellwig, Jakob Fehle, Udo Kruschwitz +1

Media Informatics Group, University of Regensburg, Regensburg, Germany · Information Science Group, University of Regensburg, Regensburg, Germany

Aspect-Based Sentiment Analysis Structured Output Generation Self-Consistency in Language Models

Feb 20, 2026

Consensus is Not Verification: Why Crowd Wisdom Strategies Fail for LLM Truthfulness

Pass@k and other methods of scaling inference compute can improve language model performance in domains with external verifiers, including mathematics and code, where incorrect candidates can be filtered reliably. This raises a natural question: can we similarly scale compute to elicit gains in truthfulness for domains without convenient verification? We show that across five benchmarks and models, surprisingly, it cannot. Even at 25x the inference cost of naive sampling, polling-style aggregation yields no consistent accuracy gains over single-sample baselines and often amplifies shared misconceptions. We find that under uncertainty, models are better at predicting what other models will say within model ensembles than at identifying what is true, revealing a separation between social prediction and truth verification. Across models and benchmarks, aggregation fails to provide a robust truth signal because language model errors are strongly correlated. The source of correlation goes beyond any individual benchmark: we show that even when conditioned on out of distribution random strings and asked to produce pseudo-random outputs, different models produce correlated outputs. Confidence-based weighting provides no benefit because self-reported confidence fails to reliably distinguish correct from incorrect answers. These results delineate a boundary for inference-time scaling: in verified domains, additional samples provide more candidates for a verifier to filter; in unverified domains, additional samples merely reinforce shared misconceptions.

Yegor Denisov-Blanch, Joshua Kazdan, Jessica Chudnovsky +4

Stanford Computer Science · Stanford Statistics · Stanford University +1

Correlated Errors in Language Models Inference-Time Scaling LLM Honesty Self-Consistency in Language Models

Feb 18, 2026

PETS: A Principled Framework Towards Optimal Trajectory Allocation for Efficient Test-Time Self-Consistency

Test-time scaling can improve model performance by aggregating stochastic reasoning trajectories. However, achieving sample-efficient test-time self-consistency under a limited budget remains an open challenge. We introduce PETS (Principled and Efficient Test-TimeSelf-Consistency), which initiates a principled study of trajectory allocation through an optimization framework. Central to our approach is the self-consistency rate, a new measure defined as agreement with the infinite-budget majority vote. This formulation makes sample-efficient test-time allocation theoretically grounded and amenable to rigorous analysis. We study both offline and online settings. In the offline regime, where all questions are known in advance, we connect trajectory allocation to crowdsourcing, a classic and well-developed area, by modeling reasoning traces as workers. This perspective allows us to leverage rich existing theory, yielding theoretical guarantees and an efficient majority-voting-based allocation algorithm. In the online streaming regime, where questions arrive sequentially and allocations must be made on the fly, we propose a novel method inspired by the offline framework. Our approach adapts budgets to question difficulty while preserving strong theoretical guarantees and computational efficiency. Experiments show that PETS consistently outperforms uniform allocation. On GPQA, PETS achieves perfect self-consistency in both settings while reducing the sampling budget by up to 75% (offline) and 55% (online) relative to uniform allocation. Code is available at https://github.com/ZDCSlab/PETS.

Zhangyi Liu, Huaizhi Qu, Xiaowei Yin +4

Stanford University · UNC at Chapel Hill · Yale University +1

Inference-Time Compute Allocation Self-Consistency in Language Models Test-Time Scaling Adaptive Sampling

Nov 15, 2025

Optimal Self-Consistency for Efficient Reasoning with Large Language Models

Self-consistency (SC) is a widely used test-time inference technique for improving performance in chain-of-thought reasoning. It consists of generating multiple responses, or ``samples", from a large language model (LLM) and selecting the most frequent answer. This procedure can naturally be viewed as a majority vote or empirical mode estimation. Despite its effectiveness, self-consistency is prohibitively expensive at scale when naively applied to datasets, and it lacks a unified theoretical understanding of sample efficiency and scaling behavior. In this paper, we provide the first comprehensive analysis of SC's scaling behavior and its variants, drawing on mode estimation and voting theory. We derive and empirically validate power law scaling for self-consistency across datasets, and analyze the sample efficiency for fixed-allocation and dynamic-allocation sampling schemes. From these insights, we introduce Blend-ASC, a novel variant of self-consistency that dynamically allocates samples to questions during inference, achieving state-of-the-art sample efficiency. Our approach uses 4.8 times fewer samples than vanilla SC on average, outperforming both fixed- and dynamic-allocation SC baselines, thereby demonstrating the superiority of our approach in terms of efficiency. In contrast to existing variants, we note that Blend-ASC is hyperparameter-free, supports batching, and can fit any budget of samples, ensuring it can be easily applied to any self-consistency application.

Austin Feng, Marius Alonso, Ambroise Odonnat +2

1Yale University, New Haven, CT, USA · 2Noah’s Ark Lab, Paris, France · 3Inria, Rennes, France +1

Self-Consistency Decoding Inference-Time Scaling Reasoning Consistency in Language Models Inference-Time Compute Allocation Efficient Language Model Inference Self-Consistency in Language Models

Oct 3, 2025

Sample, Align, Synthesize: Graph-Based Response Synthesis with ConGrs

Language models can be sampled multiple times to access the distribution underlying their responses, but existing methods cannot efficiently synthesize rich epistemic signals across different long-form responses. We introduce Consensus Graphs (ConGrs), a flexible DAG-based data structure that represents shared information, as well as semantic variation in a set of sampled LM responses to the same prompt. We construct ConGrs using a light-weight lexical sequence alignment algorithm from bioinformatics, supplemented by the targeted usage of a secondary LM judge. Further, we design task-dependent decoding methods to synthesize a single, final response from our ConGr data structure. Our experiments show that synthesizing responses from ConGrs improves factual precision on two biography generation tasks by up to 31% over an average response and reduces reliance on LM judges by more than 80% compared to other methods. We also use ConGrs for three refusal-based tasks requiring abstention on unanswerable queries and find that abstention rate is increased by up to 56%. We apply our approach to the MATH and AIME reasoning tasks and find an improvement over self-verification and majority vote baselines by up to 6 points of accuracy. We show that ConGrs provide a flexible method for capturing variation in LM responses and using the epistemic signals provided by response variation to synthesize more effective responses.

Sayan Ghosh, Shahzaib Saqib Warraich, Dhruv Tarsadiya +2

Thomas Lord Department of Computer Science, University of Southern California

Text Generation Self-Consistency in Language Models LLM Reasoning Factual Knowledge in Language Models

May 14, 2025

Atomic Consistency Preference Optimization for Long-Form Question Answering

Large Language Models (LLMs) often produce factoid hallucinations - plausible yet incorrect answers. A common mitigation strategy is model alignment, which improves factual accuracy by training on curated (factual, non-factual) pairs. However, this approach often relies on a stronger model (e.g., GPT-4) or an external knowledge base to assess factual correctness that may not always be accessible. Addressing this, we propose Atomic Consistency Preference Optimization (ACPO), a self-supervised preference-tuning method that enhances factual accuracy without external supervision. ACPO leverages atomic consistency signals (i.e., the agreement of individual facts across multiple stochastic responses) to identify high- and low-quality data pairs for model alignment. Despite being fully self-supervised, ACPO outperforms the strong supervised alignment baseline by 1.95 points averaged across Phi-3 and Llama3 on the LongFact and BioGen datasets, demonstrating its effectiveness in improving factual reliability without relying on external models or knowledge bases.

Jingfeng Chen, Raghuveer Thirukovalluru, Junlin Wang +2

LLM Hallucination Mitigation LLM Alignment Factual Knowledge in Language Models Preference Optimization Question Answering Self-Consistency in Language Models