Tool-integrated reasoning (TIR) enables large language models (LLMs) to enhance their capabilities by interacting with external tools, such as code interpreters (CI). Most recent studies focus on exploring various methods to equip LLMs with the ability to use tools. However, how to further boost the reasoning ability of already tool-capable LLMs at inference time remains underexplored. Improving reasoning at inference time requires no additional training and can help LLMs better leverage tools to solve problems. We observe that, during tool-capable LLM inference, both the number and the proportion of erroneous tool calls are negatively correlated with answer correctness. Moreover, erroneous tool calls are typically resolved successfully within a few subsequent turns. If not, LLMs often struggle to resolve such errors even with many additional turns. Building on the above observations, we propose **PruneTIR**, a rather effective yet efficient framework that enhances the tool-integrated reasoning at inference time. During LLM inference, PruneTIR prunes trajectories, resamples tool calls, and suspends tool usage through three components: *Success-Triggered Pruning*, *Stuck-Triggered Pruning and Resampling*, and *Retry-Triggered Tool Suspension*. These three components enable PruneTIR to mitigate the negative impact of erroneous tool calls and prevent LLMs from getting stuck in repeated failed resolution attempts, thereby improving overall LLM performance. Extensive experimental results demonstrate the effectiveness of PruneTIR, which significantly improves Pass@1 and efficiency while reducing the working context length for tool-capable LLMs.

Knowledge editing for large language models typically treats a fact update as a replacement in which a new object overwrites the old. Many real-world facts, however, evolve through sequences: the same subject–relation pair maps to different objects across disjoint intervals, each correct within its validity window. Existing temporal-editing benchmarks probe latest-value or explicit-time recall, so they cannot determine whether an edited model retains a coherent timeline. We introduce TKE-Bench, a diagnostic benchmark that operationalizes temporal accessibility along three reasoning-oriented axes with paraphrased variants. On TKE-Bench, all evaluated editing baselines collapse on chronologically structured queries, revealing catastrophic temporal interference (CTI). We trace CTI to a geometric property implicit in locate-and-edit methods: mid-layer subject-token keys are nearly collinear across time slices of the same subject, so stacked updates write along one direction and the latest edit overwrites every earlier value. Building on this, we propose ChronoExpert, a temporal mixture-of-experts that stores each time slice in an independent lightweight expert and activates only the matching one at inference, leaving the base model otherwise untouched. Across multiple LLM backbones, ChronoExpert consistently outperforms the strongest editing baselines and stays robust as chains grow longer.

Large language models are trained on static corpora but deployed in a dynamic world, leading to systematic temporal failures—from mis-anchored expressions and inconsistent timelines to hallucinated future events, stale world knowledge, and related issues. Existing surveys on temporal knowledge graphs, retrieval-augmented generation, hallucination, and knowledge editing cover only isolated fragments of this space: they are typically task-centric and do not offer a holistic theoretical account of how frozen LLMs represent and reason about time. This survey provides a unified perspective on temporal reasoning in LLMs. We formalize temporal queries in an information-theoretic framework based on the parametric reachability of temporal premises and answers, which induces four temporal information regimes corresponding to internal reasoning, answer recency, premise anchoring, and genuine world indeterminacy. Under this lens, we delineate the landscape of temporal failure modes, consolidate methodologies for diagnosing temporal deficiencies, and synthesize mitigation approaches into a coherent design space. Together, these contributions provide a systematic roadmap toward reliable time-aware large language models.

Knowledge within large language models (LLMs) inevitably lags behind an evolving world, motivating knowledge editing methods that update facts without expensive retraining. In multi-hop knowledge editing, models must not only recall updated facts but also correctly propagate them through multi-step reasoning chains. However, most existing approaches rely on unidirectional, feed-forward pipelines, decomposing questions and retrieving edited facts in a rigid hop-wise sequence. This design is brittle: a minor retrieval error or logical mismatch at an early hop can become a silent failure that cascades to the final answer without an explicit recovery mechanism. To address this limitation, we propose Critic-Guided Multi-Agent Reasoning for Knowledge Editing (CARE), a framework for closed-loop post-edit reasoning. A Critic agent performs chain-level verification by checking both global coherence and step-wise correctness, and triggers bounded backtracking for iterative self-correction, while a Selector agent supplies high-fidelity, low-noise candidate pools from the edit store to enable effective revision. Experiments on MQuAKE-2002 and MQuAKE-hard demonstrate that CARE effectively mitigates error propagation, achieving a new state-of-the-art.

Medical report generation from medical images is a vital AI task that helps doctors with diagnosis and marks a significant step toward creating general AI-powered medical systems. However, previous methods either fail to optimize factual accuracy or heavily depend on expert preference data. To overcome these challenges, we propose MedQPA, an automatic and generalizable report evaluation technique that uses question proposing and answering to enable controllable, structured reasoning grounded in medical domain knowledge and the factual correctness of the report. Additionally, we design MedQPA-Gen, a medical report generation pipeline that maximizes the MedQPA score through prompt engineering and reinforcement learning with MedQPA as a reward signal. We demonstrate that MedQPA is an accurate evaluation metric that closely correlates with human preferences. More importantly, MedQPA-Gen achieves higher human preference scores and better performance on downstream tasks. We open-source code at this repo https://github.com/MedQPA-gen/MedQPA-gen.

Large Language Models (LLMs) have shown great potential in Knowledge Base Question Answering (KBQA) via semantic parsing. However, existing retrieval-augmented approaches typically retrieve entities and relations in isolation based solely on semantic similarity, ignoring the structural information of the Knowledge Base (KB) and the question. To address this limitation, we propose SELF-KBQA (Subgraph-Guided Executable Logical Form Generation), a novel framework that empowers LLMs to generate logical forms conditioned on structurally aligned and semantically relevant subgraphs. Specifically, we introduce a structure-aware subgraph retrieval stage that ranks candidate subgraphs by aligning them with the question’s structure, along with semantic relevance. Subsequently, we employ a token-budgeted evidence condensation strategy to distill the top-ranked subgraphs into compact contexts for the generation stage. Extensive experiments on GrailQA, WebQSP, and GraphQuestions demonstrate that SELF-KBQA achieves state-of-the-art performance.

Automatic fact verification (FV) predicts the veracity of a claim against retrieved evidence, yet modern verifiers often behave as shortcut-driven verifiers, predicting from surface regularities in the claim rather than the factual content of the evidence. Existing debiasing methods over-correct this by uniformly enforcing evidence reliance on every sample, thereby suppressing the valid information already carried by the claim itself. We propose Evidence-Grounded Remixing for Fact Verification (EGR-FV), which estimates a per-sample evidence necessity score by contrasting claim-with-evidence and claim-only predictions, and uses it to remix training batches so that debiasing is adapted to each sample’s actual reliance on evidence. On FEVER, PolitiHop, and HOVER together with their symmetric counterparts, EGR-FV achieves state-of-the-art accuracy and robustness, validating the effectiveness of EGR-FV.

Retrieval-augmented generation (RAG) improves factual question answering by grounding LLMs in external evidence, but it becomes brittle when retrieved context conflicts with the parametric memory of the model. Existing training-free methods often address this problem by globally comparing two knowledge sources or structurally aligning them, yet they still struggle with three coupled failure axes: question complexity, long misleading evidence, and off-target parametric recall. We first quantitatively analyze the individual and joint effects of these three axes, showing that their interaction makes knowledge conflict resolution harder than any single axis alone. Motivated by this finding, we propose TRIAD, a training-free framework for knowledge conflict resolution in RAG. TRIAD decomposes a question into minimal subquestions, performs subquestion-level evidence calibration to retain concise and sufficient evidence, and uses indirect clues from calibrated evidence to guide parametric knowledge elicitation. Across three benchmarks and six LLMs, TRIAD achieves state-of-the-art performance and is the only method that surpasses a setting with additional retrieval in most cases. Additional experiments further demonstrate its generalization and worst-case robustness.

Knowledge Base Question Answering (KBQA) aims to extract accurate answers from the Knowledge Base (KB). Traditional Semantic Parsing (SP)-based methods are widely used but struggle with complex queries. Recently, large language models (LLMs) have shown promise in improving KBQA performance. However, the challenge of generating error-free logical forms remains, as skeleton, topic Entity, and relation Errors still frequently occur. To address these challenges, we propose CompKBQA(Component-wise Task Decomposition for Knowledge Base Question Answering), a novel framework that optimizes the process of fine-tuning a LLM for generating logical forms by enabling the LLM to progressively learn relevant sub-tasks like skeleton generation, topic entity generation, and relevant relations generation. Additionally, we propose R³, which retrieves and incorporates KB information into the process of logical form generation. Experimental evaluations on two benchmark KBQA datasets, WebQSP and CWQ, demonstrate that CompKBQA achieves state-of-the-art performance, highlighting the importance of task decomposition and KB-aware learning.