Large Language Models (LLMs) face significant accuracy degradation due to insufficient reasoning ability when dealing with complex and abstract tasks. Thought structures such as Chain of Thought (CoT) and Tree of Thought (ToT) focus on enhancing the reasoning capability of LLMs. However, they suffer from inherent drawbacks such as redundancy within the same layer of the tree structure and the singularity of the paths in the chain structure. Some studies have utilized Retrieval-Augmented Generation (RAG) methods to enhance CoT and ToT in mitigating hallucinations in LLMs, yet the fundamental shortcomings of the thought structures still persist. Furthermore, when dealing with multi-entity and multi-hop information, the retrieved verification knowledge often contains large amounts of fragmented, superficial, or even erroneous data, misleading the reasoning process of LLMs. To address these issues, we propose the Matrix of Thought (MoT), a novel and efficient thought structure for LLMs. MoT explores problems in both horizontal and vertical dimensions through a "column-cell communication" mechanism, enabling LLMs to actively engage in multi-strategy and deep thinking while reducing redundancy in the thought nodes within the column cells, thereby enhancing the reasoning capability of LLMs. Additionally, through a fact-correction mechanism, it leverages the knowledge graph triples retrieved by RAG and the original text to construct knowledge units and correct erroneous answers. To validate the effectiveness of this method, we conducted extensive experiments in three tasks: 24-point game, question answering evaluation, and proposition writing.The results demonstrate that our framework outperforms state-of-the-art methods, with reasoning time only 14.4\% of that of the baseline method, proving its efficiency and accuracy. The code for framework is available at https://github.com/lyfiter/mtqa.

Crystalline structure prediction is an essential prerequisite for designing materials with targeted properties. Yet, it is still an open challenge in materials design and drug discovery. Despite recent advances in computational materials science, accurately predicting three-dimensional non-polymeric crystal structures remains elusive. In this work, we focus on the molecular assembly problem, where a set $\mathcal{S}$ of identical rigid molecules is packed to form a crystalline structure. Such a simplified formulation provides a useful approximation to the actual problem. However, while recent state-of-the-art methods have increasingly adopted sophisticated techniques, the underlying learning objective remains ill-posed. We propose a better formulation that introduces a loss function capturing key geometric molecular properties while ensuring permutation invariance over $\mathcal{S}$. Remarkably, we demonstrate that within this framework, a simple regression model already outperforms prior approaches, including flow matching techniques, on the COD-Cluster17 benchmark, a curated non-polymeric subset of the Crystallography Open Database (COD).

Existing multimodal browsing benchmarks often fail to require genuine multimodal reasoning, as many tasks can be solved with text-only heuristics without vision-in-the-loop verification. We introduce MMSearch-Plus, a 311-task benchmark that enforces multimodal understanding by requiring extraction and propagation of fine-grained visual cues through iterative image-text retrieval and cross-validation under retrieval noise. Our curation procedure seeds questions whose answers require extrapolating from spatial cues and temporal traces to out-of-image facts such as events, dates, and venues. Beyond the dataset, we provide a model-agnostic agent framework with standard browsing tools and a set-of-mark (SoM) module, which lets the agent place marks, crop subregions, and launch targeted image/text searches. SoM enables provenance-aware zoom-and-retrieve and improves robustness in multi-step reasoning. We evaluated closed- and open-source MLLMs in this framework. The strongest system achieves an end-to-end accuracy of 36.0%, and integrating SoM produces consistent gains in multiple settings, with improvements up to +3.9 points. From failure analysis, we observe recurring errors in locating relevant webpages and distinguishing between visually similar events. These results underscore the challenges of real-world multimodal search and establish MMSearch-Plus as a rigorous benchmark for advancing agentic MLLMs.

Interactive streetscape mapping tools such as Google Street View (GSV) and Meta Mapillary enable users to virtually navigate and experience real-world environments via immersive 360° imagery but remain fundamentally inaccessible to blind users. We introduce StreetReaderAI, the first-ever accessible street view tool, which combines context-aware, multimodal AI, accessible navigation controls, and conversational speech. With StreetReaderAI, blind users can virtually examine destinations, engage in open-world exploration, or virtually tour any of the over 220 billion images and 100+ countries where GSV is deployed. We iteratively designed StreetReaderAI with a mixed-visual ability team and performed an evaluation with eleven blind users. Our findings demonstrate the value of an accessible street view in supporting POI investigations and remote route planning. We close by enumerating key guidelines for future work.

Chain-of-Thought (CoT) prompting has significantly advanced the reasoning capabilities of Large Language Models (LLMs), yet the reliability of these reasoning chains remains a critical challenge. A widely held "cascading failure" hypothesis suggests that errors are most detrimental when they occur early in the reasoning process. This paper challenges that assumption through systematic error-injection experiments, revealing a counter-intuitive phenomenon we term "Late-Stage Fragility": errors introduced in the later stages of a CoT chain are significantly more likely to corrupt the final answer than identical errors made at the beginning. To address this specific vulnerability, we introduce the Adaptive Self-Correction Chain-of-Thought (ASCoT) method. ASCoT employs a modular pipeline in which an Adaptive Verification Manager (AVM) operates first, followed by the Multi-Perspective Self-Correction Engine (MSCE). The AVM leverages a Positional Impact Score function I(k) that assigns different weights based on the position within the reasoning chains, addressing the Late-Stage Fragility issue by identifying and prioritizing high-risk, late-stage steps. Once these critical steps are identified, the MSCE applies robust, dual-path correction specifically to the failure parts. Extensive experiments on benchmarks such as GSM8K and MATH demonstrate that ASCoT achieves outstanding accuracy, outperforming strong baselines, including standard CoT. Our work underscores the importance of diagnosing specific failure modes in LLM reasoning and advocates for a shift from uniform verification strategies to adaptive, vulnerability-aware correction mechanisms.

Reinforcement learning (RL) is a pivotal task for enhancing Large Language Model (LLM) reasoning. Conventional algorithms, however, typically adhere to a coarse-grained credit assignment paradigm, applying a uniform reward to all tokens in a sequence, a critical flaw in long-chain reasoning tasks. In this paper, we address this challenge and propose Dynamic Entropy Weighting, a novel mechanism that facilitates fine-grained rewards through two new algorithms: Group Token Policy Optimization (GTPO), which assigns an entropy-weighted reward to each token, and the analogous algorithm Sequence-Level GRPO (GRPO-S). Our approach is founded on the hypothesis that high policy entropy within a reasoning path is a powerful heuristic for cognitive effort at pivotal junctures, which can be repurposed into a learning signal. By repurposing policy entropy for reward shaping, we achieve true per-token credit assignment. Experimental results across challenging reasoning benchmarks validate the superiority of our approach, showing our methods significantly outperform a strong DAPO baseline and confirming our entropy-weighting mechanism as the key driver of this performance boost.

Modern language models address complex questions through chain-of-thought (CoT) reasoning (Wei et al., 2023) and retrieval augmentation (Lewis et al., 2021), yet struggle with error propagation and knowledge integration. Tree-structured reasoning methods, particularly the Probabilistic Tree-of-Thought (ProbTree)(Cao et al., 2023) framework, mitigate these issues by decomposing questions into hierarchical structures and selecting answers through confidence-weighted aggregation of parametric and retrieved knowledge (Yao et al., 2023). However, ProbTree's static implementation introduces two key limitations: (1) the reasoning tree is fixed during the initial construction phase, preventing dynamic adaptation to intermediate results, and (2) each node requires exhaustive evaluation of all possible solution strategies, creating computational inefficiency. We present a dynamic reinforcement learning (Sutton and Barto, 2018) framework that transforms tree-based reasoning into an adaptive process. Our approach incrementally constructs the reasoning tree based on real-time confidence estimates, while learning optimal policies for action selection (decomposition, retrieval, or aggregation). This maintains ProbTree's probabilistic rigor while improving both solution quality and computational efficiency through selective expansion and focused resource allocation. The work establishes a new paradigm for treestructured reasoning that balances the reliability of probabilistic frameworks with the flexibility required for real-world question answering systems. Code available at: https://github.com/ahmedehabb/From-Roots-to-Rewards-Dynamic-Tree-Reasoning-with-RL

Score-function based methods for policy learning, such as REINFORCE and PPO, have delivered strong results in game-playing and robotics, yet their high variance often undermines training stability. Using pathwise policy gradients, i.e. computing a derivative by differentiating the objective function, alleviates the variance issues. However, they require an accurate action-conditioned value function, which is notoriously hard to learn without relying on replay buffers for reusing past off-policy data. We present an on-policy algorithm that trains Q-value models purely from on-policy trajectories, unlocking the possibility of using pathwise policy updates in the context of on-policy learning. We show how to combine stochastic policies for exploration with constrained updates for stable training, and evaluate important architectural components that stabilize value function learning. The result, Relative Entropy Pathwise Policy Optimization (REPPO), is an efficient on-policy algorithm that combines the stability of pathwise policy gradients with the simplicity and minimal memory footprint of standard on-policy learning. Compared to state-of-the-art on two standard GPU-parallelized benchmarks, REPPO provides strong empirical performance at superior sample efficiency, wall-clock time, memory footprint, and hyperparameter robustness.

We propose cache steering, a lightweight method for implicit steering of language models via a one-shot intervention applied directly to the key-value cache. To validate its effectiveness, we apply cache steering to induce chain-of-thought reasoning in small language models. Our approach constructs steering vectors from reasoning traces, obtained either from teacher models (e.g., GPT-4o) or existing human annotations, that shift model behavior toward more explicit, multi-step reasoning without fine-tuning or prompt modifications. Experimental evaluations on diverse reasoning benchmarks demonstrate that cache steering improves both the qualitative structure of model reasoning and quantitative task performance. Additional experiments show that the method also scales to larger models and yields further gains on challenging datasets such as GPQA and MATH. Compared to prior activation steering techniques that require continuous interventions, our one-shot cache steering offers substantial advantages in terms of inference latency, hyperparameter stability, and ease of integration with existing inference APIs. Beyond mere reasoning induction, we show that cache steering enables controllable transfer of reasoning styles (e.g., stepwise, causal, analogical), making it a practical tool for behavior-level guidance of language models.

Large Language Models (LLMs) have achieved strong performance in domains like mathematics, factual question answering, and code generation, yet their ability to reason on these tasks in different languages remains underdeveloped. Especially for low-resource languages such as Swahili or Thai, LLMs can often misinterpret prompts or default to reasoning in English. This implicit bias toward high-resource languages undermines factual accuracy, interpretability, and trust. We propose M2A, a novel method that combines multi-scale multilingual alignment with language-consistency rewards on machine-translated questions, training models to reason directly and accurately in the target language. Furthermore, existing multilingual benchmarks only evaluate on final answers, overlooking whether reasoning occurs in the intended language. To close this gap, we introduce GeoFact-X, a geography-based multilingual factual reasoning benchmark together with reasoning traces in five languages: English, Hindi, Japanese, Swahili, and Thai. Our results show that M2A significantly enhances multilingual reasoning fidelity in both mathematical and factual reasoning tasks, highlighting that reasoning-aware multilingual reinforcement learning is crucial for robust cross-lingual generalization. https://jd730.github.io/projects/M2A_GeoFact-X