LLM Reasoning

LLM: Large Language Model

Ask about this topic

Overview

This field asks how language models can carry out multi-step inference across mathematics, coding, logic, science, and real-world tasks, and how to tell robust reasoning from pattern matching. Common approaches include chain-of-thought prompting, self-consistency and tree search at inference time, tool use and symbolic verification, and post-training with supervised traces, reinforcement learning from verifiable rewards, or process feedback. Researchers are also exploring latent and recurrent computation to reduce reliance on long written traces, alongside benchmarks that test compositionality, robustness, and long-horizon execution. A central tension is that more computation can improve accuracy but raise cost, while plausible explanations and correct final answers do not by themselves establish faithful, reliable reasoning.

Latest papers

CardsList

931 papers105 papers in the last four weeks, up 27% on the four weeks before. 0.6% of all new papers.

Mar 13, 2026

Deciphering Scientific Reasoning Steps from Outcome Data for Molecule Optimization

Emerging reasoning models hold promise for automating scientific discovery. However, their training is hindered by a critical supervision gap: experimental outcomes are abundant, whereas intermediate reasoning steps are rarely documented at scale. To bridge this gap, we propose DESRO, a framework for deciphering scientific reasoning from outcomes. By analyzing shared patterns and key differences within grouped data, a large language model (LLM) can recover the underlying logic. We instantiate this framework in molecule optimization, a pivotal stage in drug discovery that traditionally relies on the iterative reasoning of medicinal chemists. Across 2.3 million molecular property records, our framework infers optimization rationales by grouping molecules with shared fragments, then using an LLM to analyze how structural variations correlate with property differences. Based on the derived data, we train a model that conducts molecule optimization through an interpretable reasoning process. DESRO achieves the highest success rates on 15 out of 18 tasks, spanning both single- and multi-property optimization of bioactivity and ADMET properties. The reasoning process enables robust generalization to out-of-distribution scenarios, including novel property combinations, unseen biological targets, and unseen properties defined solely by natural language descriptions. In retrospective case studies under strict temporal splits, the model autonomously reconstructs expert-level lead optimization trajectories. Additionally, our framework extends beyond molecule optimization to reaction ligand selection. Our results establish deciphering reasoning steps from outcome data as a viable paradigm for enabling scientific reasoning, providing a scalable approach to accelerate scientific discovery.

Zequn Liu, Kehan Wu, Shufang Xie +5

Zhongguancun Academy · School of Computer Science and Technology, University of Science and Technology of China · Global Health Drug Discovery Institute

Scientific Reasoning in Language ModelsMolecular OptimizationLarge Language Model-Guided OptimizationLLM ReasoningDrug Discovery
AskPDF
Mar 13, 2026

RetroReasoner: A Reasoning LLM for Strategic Retrosynthesis Prediction

Retrosynthesis prediction aims to identify reactants that can synthesize a given product molecule. Although molecular large language models (LLMs) have recently shown promising results, most existing methods either generate reactants directly or provide only generic product-level analysis, without explicitly reasoning about bond-disconnection strategies that justify specific reactant choices. This paper proposes RetroReasoner, a retrosynthetic reasoning model that captures chemists' strategic disconnection-based thinking. RetroReasoner is trained with supervised fine-tuning and reinforcement learning. For supervised fine-tuning, SyntheticRetro generates structured disconnection rationales paired with reactant predictions. For reinforcement learning, a round-trip reward evaluates predicted reactants by passing them through a forward synthesis model and rewarding predictions that reconstruct the original product. RetroReasoner can also be applied to multi-step retrosynthetic planning by incorporating it into a parallelized Monte Carlo tree search framework, reducing search time while increasing the number and diversity of valid synthetic pathways. Experimental results show that RetroReasoner outperforms prior baselines, including not only molecular LLMs but also retrosynthesis-specific expert models, and generates a broader range of feasible reactant proposals, especially for challenging reaction instances. The code is available at https://github.com/KU-AGI/RetroReasoner.

Hanbum Ko, Chanhui Lee, Ye Rin Kim +4

Department of Artificial Intelligence, Korea University · Materials Intelligence Lab, LG AI Research · Department of Statistics, Korea University

AI-Assisted Scientific ResearchLanguage ModelingLLM ReasoningRetrosynthesis
AskPDF
Mar 12, 2026

PLDR-LLMs Reason At Self-Organized Criticality

We show that PLDR-LLMs pretrained at self-organized criticality exhibit reasoning at inference time. The characteristics of PLDR-LLM deductive outputs at criticality is similar to second-order phase transitions. At criticality, the correlation length diverges, and the deductive outputs attain a metastable steady state. The steady state behaviour suggests that deductive outputs learn representations equivalent to scaling functions, universality classes and renormalization groups from the training dataset, leading to generalization and reasoning capabilities in the process. We can then define an order parameter from the global statistics of the model's deductive output parameters at inference. The reasoning capabilities of a PLDR-LLM is better when its order parameter is close to zero at criticality. This observation is supported by the benchmark scores of the models trained at near-criticality and sub-criticality. Our results provide a self-contained explanation on how reasoning manifests in large language models, and the ability to reason can be quantified solely from global model parameter values of the deductive outputs at steady state, without any need for evaluation of curated benchmark datasets through inductive output for reasoning and comprehension.

Burc Gokden

Fromthesky Research Labs LLC Oregon, USA

LLM ReasoningImplicit Reasoning in Language Models
AskPDF
Mar 12, 2026

TopoBench: Benchmarking LLMs on Hard Topological Reasoning

Solving topological grid puzzles requires reasoning over global spatial invariants such as connectivity, loop closure, and region symmetry and remains challenging for even the most powerful large language models (LLMs). To study these abilities under controlled settings, we introduce TopoBench, a benchmark of six puzzle families across three difficulty levels. We evaluate strong reasoning LLMs on TopoBench and find that even frontier models solve fewer than one quarter of hard instances, with two families nearly unsolved. To investigate whether these failures stem from reasoning limitations or from difficulty extracting and maintaining spatial constraints, we annotate 750 chain of thought traces with an error taxonomy that surfaces four candidate causal failure modes, then test them with targeted interventions simulating each error type. These interventions show that certain error patterns like premature commitment and constraint forgetting have a direct impact on the ability to solve the puzzle, while repeated reasoning is a benign effect of search. Finally we study mitigation strategies including prompt guidance, cell-aligned grid representations and tool-based constraint checking, finding that the bottleneck lies in extracting constraints from spatial representations and not in reasoning over them. Code and data are available at github.com/mayug/topobench-benchmark.

Mayug Maniparambil, Nils Hoehing, Janak Kapuriya +5

Intercom Research · University College Dublin · University of Galway +3

Constraint-Guided Reasoning in Language ModelsSpatial Reasoning BenchmarksSpatial ReasoningLLM Reasoning
AskPDF
Mar 11, 2026

The Stepwise Informativeness Assumption: When Do Entropy Dynamics Track Reasoning Correctness in LLMs?

Recent work uses entropy-based signals at multiple representation levels to study reasoning in large language models, but the field remains largely empirical. From a theoretical perspective, reductions in model uncertainty do not necessarily reflect progress toward correctness. In this work, we characterize when such alignment can be expected through the Stepwise Informativeness Assumption (SIA), which requires entropy-reducing reasoning-prefix extensions increase the probability assigned to valid answers in expectation. We show that next-token prediction can transfer answer-relevant information from training data to the model, but does not, by itself, guarantee alignment with correctness. Instead, we propose that this alignment depends on how the training regime couples reasoning traces to correct answers. Furthermore, we hypothesize observable signatures of this phenomenon and characterize its expected failure regimes. We empirically investigate how entropy dynamics relate to correctness across four reasoning benchmarks (GSM8K, ARC, SVAMP, MATH) and diverse open-weight language models spanning base, instruction-tuned, and RL-trained regimes. Our results show that entropy-correctness alignment strengthens with post-training but can weaken when task difficulty exceeds model capabilities.

Mar Gonzàlez I Català, Haitz Sáez de Ocáriz Borde, George D. Montañez +1

University of Cambridge · University of Oxford · Harvey Mudd College

LLM Reasoning
AskPDF
Mar 11, 2026

Making Bielik LLM Reason (Better): A Field Report

This paper presents a research program dedicated to evaluating and advancing the reasoning capabilities of Bielik, a Polish large language model. The study describes a number of stages of work: initial benchmarking and creation of evaluation methodology, analyzing of comparative results with other LLMs and outlining of future prospects that take into account the limitations of the analyses conducted so far and aims to keep Bielik in the race give the ever-changing -- and competitive -- AI landscape.

Adam Trybus, Bartosz Bartnicki, Remigiusz Kinas

Institute of Philosophy, Jagiellonian University, 52 Grodzka St., Kraków 31-044, Poland · Bielik.ai (Speakleash Foundation), 33/4 Niemczańska St., Wrocław 50-561, Poland

LLM ReasoningLanguage ModelingLLM Evaluation
AskPDF
Mar 11, 2026

Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stability

Evaluating LLM reliability via scalar probabilities often fails to capture the structural dynamics of reasoning. We introduce TRACED, a framework that assesses reasoning quality through theoretically grounded geometric kinematics. By decomposing reasoning traces into Progress (displacement) and Stability (curvature), we reveal a distinct topological divergence: correct reasoning manifests as high-progress, stable trajectories, whereas hallucinations are characterized by low-progress, unstable patterns (stalled displacement with high curvature fluctuations). Leveraging these signatures, our probabilistic framework achieves competitive performance and superior robustness across diverse benchmarks. Crucially, TRACED bridges geometry and cognition by mapping high curvature to ''Hesitation Loops'' and displacement to ''Certainty Accumulation'', offering a physical lens to decode the internal dynamics of machine thought.

Xinyan Jiang, Ninghao Liu, Di Wang +1

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) · King Abdullah University of Science and Technology · Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai, China +2

LLM ReliabilityLLM ReasoningLLM Hallucination Detection
AskPDF
Mar 11, 2026

HEAL: Hindsight Entropy-Assisted Learning for Reasoning Distillation

Distilling reasoning capabilities from Large Reasoning Models (LRMs) into smaller models is typically constrained by the limitations of rejection sampling. Standard methods treat the teacher as a static filter, discarding complex "corner-case" problems where the teacher fails to explore valid solutions independently, thereby creating an artificial "Teacher Ceiling" for the student. In this work, we propose Hindsight Entropy-Assisted Learning (HEAL), an RL-free framework designed to bridge this reasoning gap. Drawing on the educational theory of the Zone of Proximal Development (ZPD), HEAL synergizes three core modules: (1) Guided Entropy-Assisted Repair (GEAR), an active intervention mechanism that detects critical reasoning breakpoints via entropy dynamics and injects targeted hindsight hints to repair broken trajectories; (2) Perplexity-Uncertainty Ratio Estimator (PURE), a ratio-based filtering heuristic that reduces high-anomaly shortcut-like rationales; and (3) Progressive Answer-guided Curriculum Evolution (PACE), a three-stage distillation strategy that organizes training from foundational alignment to hard-case adaptation. Extensive experiments on multiple benchmarks demonstrate that HEAL significantly outperforms traditional SFT distillation and other baselines.

Wenjing Zhang, Jiangze Yan, Jieyun Huang +7

Data Science & AI Research Institute, China Unicom · Unicom Data Intelligence, China Unicom

CoT DistillationLanguage Model DistillationLLM ReasoningLarge Reasoning Models
AskPDF
Mar 10, 2026

Think Twice Before You Write -- an Entropy-based Decoding Strategy to Enhance LLM Reasoning

Decoding strategies play a central role in shaping the reasoning ability of large language models (LLMs). Traditional methods such as greedy decoding and beam search often suffer from error propagation, while sampling-based approaches introduce randomness without adequate robustness. Self-consistency improves reliability by aggregating multiple rollouts, but incurs significant computational overhead. We propose an entropy-guided decoding framework that introduces token-level adaptivity into generation. At each step, the model computes the entropy of the token distribution, identifies high-uncertainty positions, and selectively branches on these vulnerable points. A dynamic pool of partial rollouts is maintained and expanded until solutions are completed, concentrating computation where uncertainty is greatest and avoiding unnecessary exploration in confident regions. To enable efficient termination, we apply a rollout-level Entropy After </Think> (EAT) stopping criterion by performing entropy evaluation after the full reasoning trace, rather than incrementally at every step. Experiments on GSM8K, AMC2023, and their perturbed variants demonstrate that our method achieves consistently strong accuracy. Notably, on smaller LLMs, performance is comparable to GPT-5 while operating at a fraction of the cost.

Jiashu He, Meizhu Liu, Olaitan P Olaleye +9

Oracle AI Science · University of Pennsylvania

Language Model DecodingEfficient Language Model ReasoningLLM ReasoningNumerical Reasoning in Language Models
AskPDF
Mar 10, 2026

Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs

While reasoning in LLMs plays a natural role in math, code generation, and multi-hop factual questions, its effect on simple, single-hop factual questions remains unclear. Such questions do not require step-by-step logical decomposition, making the utility of reasoning highly counterintuitive. Nevertheless, we find that enabling reasoning substantially expands the capability boundary of the model's parametric knowledge recall, unlocking correct answers that are otherwise effectively unreachable. Why does reasoning aid parametric knowledge recall when there are no complex reasoning steps to be done? To answer this, we design a series of hypothesis-driven controlled experiments, and identify two key driving mechanisms: (1) a computational buffer effect, where the model uses the generated reasoning tokens to perform latent computation independent of their semantic content; and (2) factual priming, where generating topically related facts acts as a semantic bridge that facilitates correct answer retrieval. Importantly, this latter generative self-retrieval mechanism carries inherent risks: we demonstrate that hallucinating intermediate facts during reasoning increases the likelihood of hallucinations in the final answer. Finally, we show that our insights can be harnessed to directly improve model accuracy by prioritizing reasoning trajectories that contain hallucination-free factual statements.

Zorik Gekhman, Roee Aharoni, Eran Ofek +3

Google Research · Technion - Israel Institute of Technology · Tel Aviv University

Factual Knowledge in Language ModelsLLM ReasoningHallucination in Language Models
AskPDF
Mar 10, 2026

Good Reasoning Makes Good Demonstrations: Implicit Reasoning Quality Supervision via In-Context Reinforcement Learning

Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that arrive at correct answers by chance. We observe that better reasoning makes better demonstrations: high-quality solutions serve as more effective in-context examples than low-quality ones. We term this teaching ability Demonstration Utility, and show that the policy model's own in-context learning ability provides an efficient way to measure it, yielding a quality signal termed Evidence Gain. To leverage this signal during training, we introduce In-Context RLVR, which prepends demonstrations before each rollout. Theoretically, we prove that this simple input modification implicitly reweights rewards by a factor approximately proportional to Evidence Gain, assigning higher weights to high-quality traces without requiring costly computation. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements in both accuracy and reasoning quality over standard RLVR baselines. Our codes and datasets are available at https://github.com/Mithas-114/IC-DAPO.

Tiehua Mei, Minxuan Lv, Leiyu Pan +5

School of Data Science, Fudan University · University of Chinese Academy of Sciences · College of Intelligence and Computing, Tianjin University

RL for Language Model ReasoningLLM ReasoningIn-Context RLReinforcement Learning with Verifiable RewardsReasoning Evaluation
AskPDF
Mar 10, 2026

EXPLORE-Bench: Egocentric Scene Prediction with Long-Horizon Reasoning

Multimodal large language models (MLLMs) are increasingly considered as a foundation for embodied agents, yet it remains unclear whether they can reliably reason about the long-term physical consequences of actions from an egocentric viewpoint. We study this gap through a new task, Egocentric Scene Prediction with LOng-horizon REasoning: given an initial-scene image and a sequence of atomic action descriptions, a model is asked to predict the final scene after all actions are executed. To enable systematic evaluation, we introduce EXPLORE-Bench, a benchmark curated from real first-person videos spanning diverse scenarios. Each instance pairs long action sequences with structured final-scene annotations, including object categories, visual attributes, and inter-object relations, which supports fine-grained, quantitative assessment. Experiments on a range of proprietary and open-source MLLMs reveal a significant performance gap to humans, indicating that long-horizon egocentric reasoning remains a major challenge. We further analyze test-time scaling via stepwise reasoning and show that decomposing long action sequences can improve performance to some extent, while incurring non-trivial computational overhead. Overall, EXPLORE-Bench provides a principled testbed for measuring and advancing long-horizon reasoning for egocentric embodied perception.

Chengjun Yu, Xuhan Zhu, Chaoqun Du +4

University of Science and Technology of China · Foundation Model Team, Li Auto Inc. · Tsinghua University

Multimodal Large Language ModelsLong-Horizon ReasoningEgocentric Video UnderstandingEgocentric VisionMultimodal Model EvaluationLLM Reasoning
AskPDF
Mar 10, 2026

Beyond Test-Time Memory: State-Space Optimal Control for LLM Reasoning

Associative memory has long underpinned the design of sequential models. Beyond recall, humans reason by projecting future states and selecting goal-directed actions, a capability that modern language models increasingly require but do not natively encode. While prior work uses reinforcement learning or test-time training, planning remains external to the model architecture. We formulate reasoning as optimal control and introduce the Test-Time Control (TTC) layer, which performs finite-horizon LQR planning over latent states at inference time, represents a value function within neural architectures, and leverages it as the nested objective to enable planning before prediction. To ensure scalability, we derive a hardware-efficient LQR solver based on a symplectic formulation and implement it as a fused CUDA kernel, enabling parallel execution with minimal overhead. Integrated as an adapter into pretrained LLMs, TTC layers improve mathematical reasoning performance by up to +27.8% on MATH-500 and 2-3x Pass@8 improvements on AMC and AIME, demonstrating that embedding optimal control as an architectural component provides an effective and scalable mechanism for reasoning beyond test-time training.

Peihao Wang, Shan Yang, Xijun Wang +8

Amazon · The University of Texas at Austin · Georgia Institute of Technology +2

LLM ReasoningLanguage Model-Based ControlFeedback ControlContinuous Latent Reasoning
AskPDF
Mar 10, 2026

The Reasoning Trap -- Logical Reasoning as a Mechanistic Pathway to Situational Awareness

Situational awareness, the capacity of an AI system to recognize its own nature, understand its training and deployment context, and reason strategically about its circumstances, is widely considered among the most dangerous emergent capabilities in advanced AI systems. Separately, a growing research effort seeks to improve the logical reasoning capabilities of large language models (LLMs) across deduction, induction, and abduction. In this paper, we argue that these two research trajectories are on a collision course. We introduce the RAISE framework (Reasoning Advancing Into Self Examination), which identifies three mechanistic pathways through which improvements in logical reasoning enable progressively deeper levels of situational awareness: deductive self inference, inductive context recognition, and abductive self modeling. We formalize each pathway, construct an escalation ladder from basic self recognition to strategic deception, and demonstrate that every major research topic in LLM logical reasoning maps directly onto a specific amplifier of situational awareness. We further analyze why current safety measures are insufficient to prevent this escalation. We conclude by proposing concrete safeguards, including a "Mirror Test" benchmark and a Reasoning Safety Parity Principle, and pose an uncomfortable but necessary question to the logical reasoning community about its responsibility in this trajectory.

Subramanyam Sahoo, Aman Chadha, Vinija Jain +1

MARS 4.0 Fellowship, Cambridge AI Safety Hub(CAISH), University of Cambridge · AWS Generative AI Innovation Center, Amazon Web Services, USA · Stanford University +2

Strategic Reasoning Risks in Language ModelsLLM Safety AlignmentLanguage Model Safety EvaluationLLM Reasoning
AskPDF
Mar 9, 2026

Learning When to Sample: Confidence-Aware Selective Sampling for Efficient Chain-of-Thought Reasoning

Large language models (LLMs) can achieve strong reasoning performance through chain-of-thought (CoT) reasoning, yet they often generate unnecessarily long reasoning paths that incur high inference cost. Self-consistency-based approaches push accuracy higher still, but they require sampling and aggregating multiple reasoning trajectories, leading to substantial computational overhead. In this paper, we introduce a confidence-aware selective sampling framework that, at inference time, analyzes a single reasoning trajectory to adaptively determine whether to rely on that trajectory alone or trigger multi-path sampling. The framework uses trajectory-level numeric features and sentence-level linguistic features extracted from reasoning states to guide selective multi-path reasoning. We train it on MedQA and evaluate it in-domain on MedQA and under calibration-only transfer on MathQA, MedMCQA, and MMLU, without further fine-tuning. Experimental results show that the proposed framework maintains comparable performance to full and efficient multi-path reasoning baselines, with accuracy changes of −0.41±0.58-0.41 \pm 0.58 and −0.31±0.58-0.31 \pm 0.58 percentage points, respectively, while reducing token usage by 71.7±5.071.7 \pm 5.0% and 36.6±9.136.6 \pm 9.1%. These findings demonstrate that reasoning trajectories contain rich signals for uncertainty estimation, enabling a simple, transferable mechanism to balance accuracy and efficiency in LLM reasoning.

Juming Xiong, Kevin Guo, Congning Ni +7

1Vanderbilt University · 2Vanderbilt University Medical Center · 3Intuit AI Research

CoT ReasoningEfficient Language Model ReasoningLLM Inference AccelerationAdaptive SamplingLLM Uncertainty EstimationLLM Reasoning+1
AskPDF
Mar 8, 2026

TableMind++: An Uncertainty-Aware Programmatic Agent for Tool-Augmented Table Reasoning

Table reasoning requires models to jointly perform semantic understanding and precise numerical operations. Most existing methods rely on a single-turn reasoning paradigm over tables which suffers from context overflow and weak numerical sensitivity. To address these limitations, we previously proposed TableMind as a tuning-based autonomous programmatic agent that simulates human-like interaction within a lightweight large language model (LLM). TableMind internalizes planning, action, and reflection through a two-stage training strategy involving supervised fine-tuning (SFT) on filtered high-quality data and reinforcement learning (RL) via a multi-perspective reward and the Rank-Aware Policy Optimization (RAPO) algorithm. While TableMind establishes a solid foundation for programmatic agents, the inherent stochasticity of LLMs remains a critical challenge that leads to hallucinations. In this paper, we extend this foundation to TableMind++ by introducing a novel uncertainty-aware inference framework to mitigate hallucinations. Specifically, we propose memory-guided plan pruning to retrieve historical trajectories for validating and filtering out logically flawed plans to address epistemic uncertainty. To ensure execution precision, we introduce confidence-based action refinement which monitors token-level probabilities to detect and self-correct syntactic noise for aleatoric uncertainty mitigation. Finally, we employ dual-weighted trajectory aggregation to synthesize a robust consensus from multiple reasoning paths. Extensive experiments on diverse benchmarks demonstrate that TableMind++ consistently outperforms previous baselines and proprietary models to validate the effectiveness of integrating autonomous training with uncertainty quantification. Our code is available.

Mingyue Cheng, Shuo Yu, Chuang Jiang +5

State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China, Hefei, China

LLM ReasoningMemory-Augmented Language Model AgentsLLM Uncertainty Estimation
AskPDF
Mar 7, 2026

Prompt Complexity Dilutes Structured Reasoning: A Follow-Up Study on the Car Wash Problem

In a previous study [Jo, 2026], STAR reasoning (Situation, Task, Action, Result) raised car wash problem accuracy from 0% to 85% on Claude Sonnet 4.5, and to 100% with additional prompt layers. This follow-up asks: does STAR maintain its effectiveness in a production system prompt? We tested STAR inside InterviewMate's 60+ line production prompt, which had evolved through iterative additions of style guidelines, format instructions, and profile features. Three conditions, 20 trials each, on Claude Sonnet 4.6: (A) production prompt with Anthropic profile, (B) production prompt with default profile, (C) original STAR-only prompt. C scored 100% (verified at n=100). A and B scored 0% and 30%. Prompt complexity dilutes structured reasoning. STAR achieves 100% in isolation but degrades to 0-30% when surrounded by competing instructions. The mechanism: directives like "Lead with specifics" force conclusion-first output, reversing the reason-then-conclude order that makes STAR effective. In one case, the model output "Short answer: Walk." then executed STAR reasoning that correctly identified the constraint -- proving the model could reason correctly but had already committed to the wrong answer. Cross-model comparison shows STAR-only improved from 85% (Sonnet 4.5) to 100% (Sonnet 4.6) without prompt changes, suggesting model upgrades amplify structured reasoning in isolation. These results imply structured reasoning frameworks should not be assumed to transfer from isolated testing to complex prompt environments. The order in which a model reasons and concludes is a first-class design variable.

Heejin Jo

Independent Researcher

LLM PromptingLLM Reasoning
AskPDF
Mar 7, 2026

Entropy-Aware On-Policy Distillation of Language Models

On-policy distillation is a promising approach for transferring knowledge between language models, where a student learns from dense token-level signals along its own trajectories. This framework typically uses reverse KL divergence, encouraging the student to match the teacher's high-confidence predictions. However, we show that the mode-seeking property of reverse KL reduces generation diversity and yields unstable learning signals when the teacher distribution has high entropy. To address this, we introduce Entropy-Aware On-Policy Distillation. Our key idea is augmenting the standard reverse KL objective with forward KL when teacher entropy is high, capturing the full range of plausible outputs while retaining precise imitation elsewhere. It balances mode-seeking precision with mode-covering robustness without sacrificing on-policy training efficiency. Experiments show that our method maintains generation diversity (sustained token-level entropy) and improves student-teacher alignment (lower forward KL on high-entropy tokens). Across six math reasoning benchmarks, this yields Pass@8 accuracy gains of +1.37 for Qwen3-0.6B-Base, +2.39 for Qwen3-1.7B-Base, and +5.05 for Qwen3-4B-Base compared to baseline on-policy distillation methods. These results demonstrate that accounting for teacher uncertainty is essential for maintaining diversity and achieving effective knowledge transfer.

Woogyeol Jin, Taywon Min, Yongjin Yang +5

KAIST AI, Seoul, South Korea · University of Toronto, Ontario, Canada · Vector Institute, Ontario, Canada +1

Language Model DistillationOn-Policy DistillationUncertainty-Aware Knowledge DistillationLLM Reasoning
AskPDF
Mar 6, 2026

Reforming the Mechanism: Editing Reasoning Patterns in LLMs with Circuit Reshaping

Large language models (LLMs) often exhibit flawed reasoning ability that undermines reliability. Existing approaches to improving reasoning typically treat it as a general and monolithic skill, applying broad training which is inefficient and unable to target specific reasoning errors. We introduce Reasoning Editing, a paradigm for selectively modifying specific reasoning patterns in LLMs while preserving other reasoning pathways. This task presents a fundamental trade-off between Generality, the ability of an edit to generalize across different tasks sharing the same reasoning pattern, and Locality, the ability to preserve other reasoning capabilities. Through systematic investigation, we uncover the Circuit-Interference Law: Edit interference between reasoning patterns is proportional to the overlap of their neural circuits. Guided by this principle, we propose REdit, the first framework to actively reshape neural circuits before editing, thereby modulating interference between reasoning patterns and mitigating the trade-off. REdit integrates three components: (i) Contrastive Circuit Reshaping, which directly addresses the generality-locality trade-off by disentangling overlapping circuits; (ii) Meta-Contrastive Learning, which extends transferability to novel reasoning patterns; and (iii) Dual-Level Protection, which preserves preexisting abilities by constraining reshaping update directions and regularizing task-level predictions. Extensive experiments with Qwen-2.5-3B on propositional logic reasoning tasks across three difficulty levels demonstrate that REdit consistently achieves superior generality and locality compared to baselines, with additional validation in mathematics showing broader potential. Our code is available at https://github.com/LzyFischer/REdit.

Zhenyu Lei, Qiong Wu, Jianxiong Dong +4

University of Virginia · AT&T Chief Data Office · Florida State University

Language Model EditingLLM ReasoningCircuits in Language Models
AskPDF
Mar 6, 2026

Not Too Short, Not Too Long: How LLM Response Length Shapes People's Critical Thinking in Error Detection

Large language models (LLMs) have become common decision-support tools across educational and professional contexts, raising questions about how their outputs shape human critical thinking. Prior work suggests that the amount of AI assistance can influence cognitive engagement, yet little is known about how specific properties of LLM outputs (e.g., response length) impacts users' critical evaluation of information. In this study, we examine whether the length of LLM responses shapes users' accuracy in evaluating LLM-generated reasoning on critical thinking tasks, particularly in interaction with the correctness of the LLM's reasoning. To begin evaluating this, we conducted a within-subjects experiment with 24 participants who completed 15 modified Watson--Glaser critical thinking items, each accompanied by an LLM-generated explanation that varied in length and correctness. Mixed-effects logistic regression revealed a strong and statistically reliable effect of LLM output correctness on participant accuracy, with participants more likely to answer correctly when the LLM's explanation was correct. Response length appeared to moderated this effect: when the LLM output was incorrect, medium-length explanations were associated with higher participant accuracy than either shorter or longer explanations, whereas accuracy remained high across lengths when the LLM output was correct. Together, these findings suggest that response length alone may be insufficient to support critical thinking, and that how reasoning is presented-including a potential advantage of mid-length explanations under some conditions-points to design opportunities for LLM-based decision-support systems that emphasize transparent reasoning and calibrated expressions of certainty.

Natalie Friedman, Adelaide Nyanyo, Kevin Weatherwax +4

BTP Innovation, SAP, Palo Alto, CA, USA

Human-AI Decision MakingLLM Reasoning
AskPDF
Mar 6, 2026

LEAD: Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning

Long-horizon execution in Large Language Models (LLMs) remains unstable even when high-level strategies are provided. Evaluating on controlled algorithmic puzzles, we demonstrate that while decomposition is essential for stability, extreme decomposition creates a "no-recovery bottleneck". We show that this bottleneck becomes critical due to highly non-uniform error distribution, where consistent errors on a few "hard" steps become irreversible. To address this, we propose Lookahead-Enhanced Atomic Decomposition (LEAD). By incorporating short-horizon future validation and aggregating overlapping rollouts, LEAD provides enough isolation to maintain stability while retaining enough local context to correct errors. This enables the o4-mini model to solve Checkers Jumping up to complexity n=13n=13, whereas extreme decomposition fails beyond n=11n=11.

Denys Pushkin, Emmanuel Abbe

EPFL, Apple

Long-Horizon LLM ReasoningLong-Horizon ReasoningLLM Reasoning
AskPDF
Mar 6, 2026

Evaluation of Deontic Conditional Reasoning in Large Language Models: The Case of Wason's Selection Task

As large language models (LLMs) advance in linguistic competence, their reasoning abilities are gaining increasing attention. In humans, reasoning often performs well in domain specific settings, particularly in normative rather than purely formal contexts. Although prior studies have compared LLM and human reasoning, the domain specificity of LLM reasoning remains underexplored. In this study, we introduce a new Wason Selection Task dataset that explicitly encodes deontic modality to systematically distinguish deontic from descriptive conditionals, and use it to examine LLMs' conditional reasoning under deontic rules. We further analyze whether observed error patterns are better explained by confirmation bias (a tendency to seek rule-supporting evidence) or by matching bias (a tendency to ignore negation and select items that lexically match elements of the rule). Results show that, like humans, LLMs reason better with deontic rules and display matching-bias-like errors. Together, these findings suggest that the performance of LLMs varies systematically across rule types and that their error patterns can parallel well-known human biases in this paradigm.

Hirohiko Abe, Kentaro Ozeki, Risako Ando +3

Keio University, Tokyo, Japan · University of Tokyo, Tokyo, Japan

LLM Reasoning
AskPDF
Mar 5, 2026

CRISP: Compressed Reasoning via Iterative Self-Policy Distillation

Reasoning models often generate far more tokens than a task requires, which raises inference cost and can compound errors. We introduce CRISP (Compressed Reasoning via Iterative Self-Policy Distillation), an on-policy self-distillation method that teaches a model to reason more concisely by distilling its own concise behavior back into itself. The method uses a single idea: condition the same model on a "be concise" instruction to obtain teacher logits, then minimize the per-token reverse KL divergence between the student and this teacher on the student's own rollouts. It requires no ground-truth answers, no token budgets, and no difficulty estimators. The reverse-KL objective is naturally difficulty-adaptive: it compresses easy problems aggressively while preserving the reasoning steps that hard problems require. On Qwen3-14B, CRISP cuts reasoning length by up to 56% on MATH-500 and 38% on the harder AIME 2024, while improving MATH-500 accuracy by up to 3.3 points over the base model and holding AIME 2024 accuracy within about one point. This behavior generalizes across model sizes and families: Qwen3-8B shows the same compression with accuracy preserved, and DeepSeek-R1-Distill-Llama-8B improves accuracy on all five benchmarks while shortening its responses. General capabilities are preserved across all three models. Code is available at https://github.com/HJSang/OPSD_Reasoning_Compression.

Hejian Sang, Yuanda Xu, Zhengze Zhou +3

CoT DistillationEfficient Language Model ReasoningOn-Policy DistillationCoT CompressionLLM Reasoning
AskPDF
Mar 5, 2026

X-RAY: Mapping LLM Reasoning Capability via Formalized and Calibrated Probes

Large language models (LLMs) achieve promising performance, yet their ability to reason remains poorly understood. Existing evaluations largely emphasize task-level accuracy, often conflating pattern matching with reasoning capability. We present X-RAY, an explainable reasoning analysis system that maps the LLM reasoning capability using calibrated, formally verified probes. We model reasoning capability as a function of extractable structure, operationalized through formal properties such as constraint interaction, reasoning depth, and solution-space geometry. X-Ray generates probes via formal tools with controlled structural variations, enabling precise isolation of incremental structural information through formal calibration and verification. We evaluate state-of-the-art LLMs on problems ranging from junior-level to advanced in mathematics, physics, and chemistry. Our analysis reveals a systematic asymmetry in LLM reasoning: models are relatively robust to constraint refinement, where additional conditions shrink an existing solution space, but degrade sharply under solution-space restructuring, where modifications alter the underlying structural form of the solution manifold. Moreover, calibrated formal probes differentiate models that appear indistinguishable on standard benchmarks and reveal failure modes that are structurally interpretable rather than opaque. Beyond evaluation, our framework is contamination-free and supports the training and testing of reasoning models.

Tianxi Gao, Yufan Cai, Yusi Yuan +1

National University of Singapore, Singapore

Structured Reasoning in Language ModelsReasoning EvaluationLLM ReasoningImplicit Reasoning in Language ModelsScientific Reasoning in Language Models
AskPDF
Mar 5, 2026

Enhancing Zero-shot Commonsense Reasoning by Integrating Visual Knowledge via Machine Imagination

Recent advancements in zero-shot commonsense reasoning have empowered Pre-trained Language Models (PLMs) to acquire extensive commonsense knowledge without requiring task-specific fine-tuning. Despite this progress, these models frequently suffer from limitations caused by human reporting biases inherent in textual knowledge, leading to understanding discrepancies between machines and humans. To bridge this gap, we introduce an additional modality to enrich the reasoning capabilities of PLMs. We propose Imagine (Machine Imagination-based Reasoning), a novel zero-shot commonsense reasoning framework that supplements textual inputs with visual signals from machine-generated images. Specifically, we enhance PLMs with the ability to imagine by embedding an image generator directly into the reasoning pipeline. To facilitate effective utilization of this imagined visual context, we construct synthetic datasets designed to emulate visual question-answering scenarios. Through comprehensive evaluations on multiple commonsense reasoning benchmarks, we demonstrate that Imagine substantially outperforms existing zero-shot approaches and even surpasses advanced large language models. These results underscore the capability of machine imagination to mitigate reporting bias and significantly enhance the generalization ability of commonsense reasoning models

Hyuntae Park, Yeachan Kim, SangKeun Lee

Department of Artificial Intelligence, Korea University, 145 Anam-ro, Seungbuk-gu, Seoul, 02841, Republic of Korea · Department of Computer Science and Engineering, Korea University, 145 Anam-ro, Seungbuk-gu, Seoul, 02841, Republic of Korea

LLM ReasoningZero-Shot LearningKnowledge Augmentation for Language Models
AskPDF
Mar 5, 2026

∇\nabla-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space

Scaling inference-time compute for Large Language Models (LLMs) has unlocked unprecedented reasoning capabilities. However, existing inference-time scaling methods typically rely on inefficient and suboptimal discrete search algorithms or trial-and-error prompting to improve the online policy. In this paper, we propose ∇\nabla-Reasoner, an iterative generation framework that integrates differentiable optimization over token logits into the decoding loop to refine the policy on the fly. Our core component, Differentiable Textual Optimization (DTO), leverages gradient signals from both the LLM's likelihood and a reward model to refine textual representations. ∇\nabla-Reasoner further incorporates rejection sampling and acceleration design to robustify and speed up decoding. Theoretically, we show that performing inference-time gradient descent in the sample space to maximize reward is dual to aligning an LLM policy via KL-regularized reinforcement learning. Empirically, ∇\nabla-Reasoner achieves over 20% accuracy improvement on a challenging mathematical reasoning benchmark, while reducing number of model calls by approximately 10-40% compared to strong baselines. Overall, our work introduces a paradigm shift from zeroth-order search to first-order optimization at test time, offering a cost-effective path to amplify LLM reasoning.

Peihao Wang, Ruisi Cai, Zhen Wang +4

The University of Texas at Austin · UC San Diego · TTIC +1

LLM ReasoningTest-Time OptimizationEfficient Language Model ReasoningInference-Time Language Model Alignment
AskPDF
Mar 4, 2026

T2S-Bench & Structure-of-Thought: Benchmarking and Prompting Comprehensive Text-to-Structure Reasoning

Think about how human handles complex reading tasks: marking key points, inferring their relationships, and structuring information to guide understanding and responses. Likewise, can a large language model benefit from text structure to enhance text-processing performance? To explore it, in this work, we first introduce Structure of Thought (SoT), a prompting technique that explicitly guides models to construct intermediate text structures, consistently boosting performance across eight tasks and three model families. Building upon this insight, we present T2S-Bench, the first benchmark designed to evaluate and improve text-to-structure capabilities of models. T2S-Bench includes 1.8K samples across 6 scientific domains and 32 structural types, rigorously constructed to ensure accuracy, fairness, and quality. Evaluation on 45 mainstream models reveals substantial improvement potential: the average accuracy on the multi-hop reasoning task is only 52.1%, and even the most advanced model achieves 58.1% node accuracy in end-to-end extraction. Furthermore, on Qwen2.5-7B-Instruct, SoT alone yields an average +5.7% improvement across eight diverse text-processing tasks, and fine-tuning on T2S-Bench further increases this gain to +8.6%. These results highlight the value of explicit text structuring and the complementary contributions of SoT and T2S-Bench. Dataset and eval code have been released at https://t2s-bench.github.io/T2S-Bench-Page/.

Qinsi Wang, Hancheng Ye, Jinhee Kim +12

Duke University · UT Austin · Meta

Structured ReasoningLLM ReasoningLLM EvaluationLLM PromptingLLM Information Extraction
AskPDF
Mar 3, 2026

Verify to Amplify: Improving Reasoning via Learned Chain-of-Thought Verification

Large Language Models (LLMs) using chain-of-thought have demonstrated great potential for solving complex reasoning and planning tasks. Despite these advances, LLM-generated outputs remain susceptible to errors, making verification important for reliable reasoning systems. Learned verifiers can increase trust, enforce safety constraints, and ensure alignment with personal preferences, while also providing feedback to improve generation. This raises a central challenge: when learned verifiers are used to guide generation, the feedback loop between generator and verifier may induce a distribution shift. This is particularly salient for process reward models, a prominent class of learned verifiers that score or classify individual steps in a chain-of-thought reasoning trace. Motivated by this challenge, we propose a new online learning framework for chain-of-thought verifiers that, given a problem statement and a reasoning trace, check the correctness of each reasoning step given the preceding steps. Highlighting the asymmetric role of soundness errors (accepting an incorrect reasoning step) and completeness errors (flagging a correct step as wrong), we introduce novel notions of dimension that characterize their optimal tradeoff. We then show how our learned verifiers can boost the accuracy of a weak generator. Assuming that the generator can produce a correct next step with a small success probability, we show how to learn a strong generator with small error and abstention rates. Our results also allow learning from offline data when queries to an expert verifier can be simulated from a small set of correct reasoning traces. However, we establish a separation between our approach and learning from offline expert demonstrations: we show that learning from offline demonstrations cannot in general achieve the soundness-completeness guarantees produced by our interactive learning approach.

Maria-Florina Balcan, Avrim Blum, Kiriaki Fragkia +2

Carnegie Mellon University · Toyota Technological Institute at Chicago · Northwestern University

LLM EvaluationLLM Answer VerificationProcess Reward ModelsLLM ReliabilityLLM Reasoning
AskPDF
Mar 3, 2026

Step-Level Sparse Autoencoder for Reasoning Process Interpretation

Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning. However, their reasoning patterns remain too complicated to analyze. While Sparse Autoencoders (SAEs) have emerged as a powerful tool for interpretability, existing approaches predominantly operate at the token level, creating a granularity mismatch when capturing more critical step-level information, such as reasoning direction and semantic transitions. In this work, we propose step-level sparse autoencoder (SSAE), which serves as an analytical tool to disentangle different aspects of LLMs' reasoning steps into sparse features. Specifically, by precisely controlling the sparsity of a step feature conditioned on its context, we form an information bottleneck in step reconstruction, which splits incremental information from background information and disentangles it into several sparsely activated dimensions. Experiments on multiple base models and reasoning tasks show the effectiveness of the extracted features. By linear probing, we can easily predict surface-level information, such as generation length and first token distribution, as well as more complicated properties, such as the correctness and logicality of the step. These observations indicate that LLMs should already at least partly know about these properties during generation, which provides the foundation for the self-verification ability of LLMs. Our code is available at https://github.com/Miaow-Lab/SSAE.

Xuan Yang, Jiayu Liu, Yuhang Lai +3

Department of Data Science, City University of Hong Kong · Hong Kong Institute of AI for Science, City University of Hong Kong · Li Auto Inc +1

LLM InterpretabilitySparse AutoencodersLLM Reasoning
AskPDF
Mar 3, 2026

Multi-Agent Debate with Memory Masking

Large language models (LLMs) have recently demonstrated impressive capabilities in reasoning tasks. Currently, mainstream LLM reasoning frameworks predominantly focus on scaling up inference-time sampling to enhance performance. In particular, among all LLM reasoning frameworks, multi-agent debate (MAD), which employs multiple LLMs as agents to perform reasoning in the way of multi-round debate, has emerged as a powerful reasoning paradigm since it allows agents to access previous memories to alleviate fallacious content and refine their reasoning iteratively in each debate round. However, although MAD significantly improves the reasoning capabilities of LLMs, in this paper, we observe that there remain erroneous memories, and LLM agents are vulnerable to these erroneous memories. To explore this phenomenon, we provide a theoretical insight that the performance of MAD is highly dependent on the quality of memories derived from the previous debate, indicating that the existence of erroneous memories poses a threat to the performance of MAD. To address this problem, we introduce a simple yet effective multi-agent debate framework, multi-agent debate with memory masking (MAD-M2^2), to improve the robustness of MAD by allowing LLM agents to mask erroneous memories from the previous debate round at the beginning of each debate round. In this way, MAD-M2^2 can polish the contextual information before each debate round by preserving informative and meaningful memories while discarding the erroneous memories. Extensive experiments and analyses on mainstream mathematical and logical reasoning benchmarks demonstrate that MAD-M2^2 can identify the erroneous memories and achieve better performance in reasoning than MAD.

Hongduan Tian, Xiao Feng, Ziyuan Zhao +3

TMLR Group, Hong Kong Baptist University · Search Algorithm Group, WeChat, Tencent

Multi-Agent DebateLLM ReasoningMemory-Augmented Language Model Agents
AskPDF