Retrieval-augmented generation (RAG) has become a cornerstone for knowledge-intensive tasks. However, the efficacy of RAG is often bottlenecked by the “one-size-fits-all” retrieval paradigm, as different queries exhibit distinct preferences for different retrievers. While recent routing techniques attempt to select the optimal retriever dynamically, they typically operate under a ‘single and static capability’ assumption, selecting retrievers solely based on semantic relevance. This overlooks a critical distinction in RAG: a retrieved document must not only be relevant but also effectively support the generator in producing correct answers. To address this limitation, we propose R³AG, a novel routing framework that explicitly models the dynamic alignment between queries and retriever capabilities. Unlike previous approaches, R³AG decomposes retriever capability into two learnable dimensions: retrieval quality and generation utility. We employ a contrastive learning objective that leverages complementary supervision signals, i.e., document assessments and downstream answer correctness, to capture query-specific preference shifts. Extensive experiments on diverse knowledge-intensive tasks demonstrate that R³AG consistently outperforms both the best individual retrievers and state-of-the-art static routing methods.
论文检索
输入标题、作者或关键词，从 9,460 篇学术成果中精准定位
Reasoning segmentation requires models to ground complex, implicit textual queries into precise pixel-level masks. Existing approaches rely on a single segmentation token \<SEG\>, whose hidden state implicitly encodes both semantic reasoning and spatial localization, limiting the model’s ability to explicitly disentangle *what to segment* from *where to segment*. We introduce AnchorSeg, which reformulates reasoning segmentation as a structured conditional generation process over image tokens, conditioned on language grounded query banks. Instead of compressing all semantic reasoning and spatial localization into a single embedding, AnchorSeg constructs an ordered sequence of query banks: latent reasoning tokens that capture intermediate semantic states, and a segmentation anchor token that provides explicit spatial grounding. We model spatial conditioning as a factorized distribution over image tokens, where the anchor query determines localization signals while contextual queries provide semantic modulation. To bridge token-level predictions and pixel-level supervision, we propose Token–Mask Cycle Consistency (TMCC), a bidirectional training objective that enforces alignment across resolutions. By explicitly decoupling spatial grounding from semantic reasoning through structured language grounded query banks, AnchorSeg achieves state-of-the-art results on ReasonSeg test set (67.7% gIoU and 68.1% cIoU). All code and models are publicly available at https://github.com/rui-qian/AnchorSeg.
Despite recent progress, existing agent benchmarks neglect a fundamental real-world capability: hierarchical rule application, a critical requirement in fields such as law and medicine where agents must reason from broad categories down to specific exceptions to reach rule-compliant decisions.This introduces significant challenges in resolving logical dependencies and disambiguating vague boundaries.To bridge this gap, we introduce HSCodeComp, a novel benchmark derived from e-commerce, requiring agents to assign a unique 10-digit Harmonized System (HS) Code to products by aligning their fuzzy attributes with strict tariff classification rules.HSCodeComp comprises 632 realistic products across 32 categories, featuring detailed yet noisy product information (titles, attributes, and images). The HS Codes are annotated by a panel of 26 tariff experts, strictly adhering to official rules and an empirical knowledge base, both of which we jointly open-source.Through a comprehensive evaluation of 23 LLMs, VLMs, and agents on HSCodeComp, we demonstrate that: 1) a substantial performance gap remains between state-of-the-art agents and human experts (46.8% vs. 95.0%); and 2) test-time scaling fails to close this gap. Further analysis reveals that 1) excessive reasoning steps frequently induce “reasoning drift,” which degrades accuracy; and 2) agents are prone to premature decisions on high-level categories and reasoning hallucinations that lack factual grounding.
Large Reasoning Models (LRMs) have achieved remarkable success on reasoning-intensive tasks such as mathematics and programming. However, their enhanced reasoning capabilities do not necessarily translate to improved safety performance—and in some cases, may even degrade it. This raises an important research question: how should we enhance the safety of LRMs? In this paper, we present a comprehensive empirical study on how to enhance the safety of LRMs through Supervised Fine-Tuning (SFT). Our investigation begins with an unexpected observation: directly distilling safe responses from DeepSeek-R1 fails to significantly enhance safety. We analyze this phenomenon and identify five key risky patterns that contribute to it. We then demonstrate that explicitly addressing these issues during the data distillation process can lead to substantial safety improvements. Next, we explore whether a long and complex reasoning process is necessary for achieving safety. Interestingly, we find that simply using short or template-based reasoning process can attain comparable safety performance. These findings prompt a deeper reflection on the role of reasoning in ensuring safety. Finally, we conduct a comprehensive ablation study to reveal the impact of different training configurations. Overall, we hope our empirical study could provide a more holistic picture on enhancing the safety of LRMs.
Multi-hop question answering requires retrieving and integrating evidence from multiple contexts. Despite the rapid progress of current research, multi-hop reasoning remains constrained by two persistent limitations: attention decay, where the model’s focus on main question degrades as the reasoning chain grows, and error accumulation, where mistakes propagate across hops and compounds into final failure. Inspired by Theta-Gamma hierarchical oscillation which decouples global planning from local retrieval, enabling efficient attention transfer between hops and a verification and repair mechanism that interrupts the accumulation of errors in the wrong paths, we present **THOR**, a brain-inspired Theta-Gamma hierarchical oscillatory reasoning framework. Extensive comparative experiments and specific validation experiments on multi-hop QA benchmarks demonstrate that THOR improves answer accuracy and robustness while mitigating limitations, showcasing its generalization across different backbones.
Recent breakthroughs in video generation have demonstrated an emerging capability termed Chain-of-Frames (CoF) reasoning, where models resolve complex tasks through the generation of continuous frames. While these models show promise for Generative Video Reasoning (GVR), existing evaluation frameworks often rely on single-frame assessments, which can lead to outcome-hacking, where a model reaches a correct conclusion through an erroneous process. To address this, we propose a process-aware evaluation paradigm. We introduce VIPER, a comprehensive benchmark spanning 16 tasks across temporal, structural, symbolic, spatial, physics, and planning reasoning. Furthermore, we propose Process-outcome Consistency (POC@r), a new metric that utilizes VLM-as-Judge with a hierarchical rubric to evaluate both the validity of the intermediate steps and the final result. Our experiments reveal that state-of-the-art video models achieve POC@1.0 only about 20% and exhibit a significant outcome-hacking. We further explore the impact of test-time scaling and sampling robustness, highlighting a substantial gap between current video generation and true generalized visual reasoning. Our benchmark are released at https://github.com/RUCAIBox/VIPER.
Bridging Internal Consistency and External Alignment: A Causal and Dynamic Interpretability Framework for LLM Generation
PDF ↗Large Language Models (LLMs) are widely used in high-stakes applications, making their interpretability increasingly important. Existing interpretability methods are typically categorized into internal and external perspectives, which are often studied in isolation and tend to overlook two key aspects: causality and temporal dynamics. Explanations are often limited to surface correlations or static dependencies, failing to capture how influences evolve during autoregressive generation. To address these limitations, we propose a causal and dynamic interpretability framework for LLM generation. We first characterize the backdoor-adjusted causal effects of both the generated prefix and the prompt on the current token using the Structural Causal Model. Next, we introduce two metrics to quantify contextual causal influence and question–answer causal influence. Overall, our work provides a unified causal view of internal consistency and external alignment in LLM generation dynamics.
While Multimodal Large Language Models (MLLMs) are advancing rapidly, accurately evaluating their capabilities remains challenging. Current paradigms primarily rely on holistic scoring and static leaderboards, which fail to disentangle fine-grained competencies. Specifically, they suffer from “Outcome Bias” by validating only final answers and ignoring intermediate reasoning. To address these limitations, we introduce ATOM (AnaTomy Of MLLM), a novel MLLM-as-a-judge framework designed to shift the focus from ranking to fine-grained diagnosis. ATOM decomposes complex reasoning into atomic criteria anchored in visual elements, enforcing verification against explicit visual facts. Validated on a newly constructed benchmark with rigorous human rankings, ATOM achieves state-of-the-art accuracy, surpassing the strongest baseline by up to 7.92%. Moving beyond ranking, ATOM bridges the gap between assessment and alignment: by pinpointing atomic-level failures, it establishes a closed-loop mechanism for targeted self-correction. This approach enables models to identify and rectify errors autonomously, successfully resolving up to 39.95% of previously failed queries without human intervention.
Test-time scaling has enabled Large Language Models (LLMs) to tackle complex reasoning, yet the limitations of current Chain-of-Thought (CoT) evaluation obscure whether performance gains stem from genuine reasoning or mere verbosity. To address this, (1) we propose a novel neuro-symbolic framework for the non-intrusive, comprehensive process-centric evaluation of reasoning grounded in First-Order Logic. (2) Through this lens, we identify four distinct behavioral prototypes and diagnose the failure modes. (3) We examine the impact of inference mode, training strategy, and model scale. Our analysis reveals that extended token generation is not a prerequisite for deep reasoning. Furthermore, we reveal critical constraints: mixing long and short CoT data in training risks premature saturation and collapse, while distillation into smaller models captures behavioral length but fails to replicate logical efficacy due to intrinsic capacity limits.
In multimodal information retrieval (MMIR), candidates relevant to an input query need to be retrieved from a database, where the query and database items span different modalities. As real-world databases evolve, repeatedly annotating and indexing data and re-optimizing domain-specific models across modalities is impractical. We present MULTI-SCORE, a fine-tuning-free, two-stage MMIR approach that couples efficient candidate filtering with fine-grained multimodal re-ranking. Stage-1 adopts Matryoshka representations to efficiently filter out low-relevance candidates without expensive similarity computations on full-scale representations for the entire database. Stage-2 re-ranks the filtered candidates by computing their fine-grained multimodal contextual representations with two scoring functions for semantic alignment using chain-of-thought prompting and question-answering. Experiments demonstrate state-of-the-art zero-shot retrieval on 12 MMIR tasks across 32 datasets while outperforming supervised methods on 23 datasets.
Think Before Writing: Feature-Level Multi-Objective Optimization for Generative Citation Visibility
PDF ↗Generative answer engines expose content through selective citation rather than ranked retrieval, fundamentally altering how visibility is determined. This shift calls for new optimization methods beyond traditional search engine optimization. Existing generative engine optimization (GEO) approaches primarily rely on token-level text rewriting, offering limited interpretability and weak control over the trade-off between citation visibility and content quality. We propose FeatGEO, a feature-level, multi-objective optimization framework that abstracts webpages into interpretable structural, content, and linguistic properties. Instead of directly editing text, FeatGEO optimizes over this feature space and uses a language model to realize feature configurations into natural language, decoupling high-level optimization from surface-level generation. Experiments on GEO-Bench across three generative engines demonstrate that FeatGEO consistently improves citation visibility while maintaining or improving content quality, substantially outperforming token-level baselines. Further analyses show that citation behavior is more strongly influenced by document-level content properties than by isolated lexical edits, and that the learned feature configurations generalize across language models of different scales.
Large Language Models (LLMs) can perform sentiment analysis via natural language instructions, yet their predictions are highly sensitive to prompt phrasing. Prior work has shown that sentiment is encoded linearly in LLM representations, but the model’s ability to utilize this information remains surprisingly fragile to prompt variations. We leverage Sparse Autoencoders (SAEs) and circuit-level analysis to uncover causal mechanisms underlying sentiment prediction. We identify a sentiment analysis circuit and find that prompt sensitivity may stem from task activation failure. The model encodes the sentiment feature consistently, but different prompts trigger varying degrees of circuit activation. Based on this insight, we propose a simple inference-time intervention method that amplifies circuit features to compensate for insufficient activation. Experiments across diverse datasets, templates, and languages show consistent improvements, offering an interpretable and training-free alternative to manual prompt engineering.
PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modeling
PDF ↗Reward models (RMs) are central to reinforcement learning from human feedback (RLHF), providing the critical supervision signals that align large language models (LLMs) with human preferences.Generative reward models (GRMs) provide greater interpretability than traditional scalar RMs, but they come with a critical trade-off: pairwise methods are hindered by a training-inference mismatch, while pointwise methods require expensive absolute annotations.To bridge this gap, we propose the Preference-aware Task-adaptive Reward Model (PaTaRM).Unlike prior approaches, PaTaRM enables robust pointwise training using readily available pairwise data via a novel Preference-Aware Reward (PAR) mechanism, eliminating the need for explicit rating labels. Furthermore, it incorporates a task-adaptive rubric system that dynamically generates instance-specific criteria for precise evaluation.Extensive experiments demonstrate that PaTaRM achieves an average relative improvement of 8.7% over the corresponding base models on RewardBench and RMBench across the Qwen3-8B and Qwen3-14B backbones.Crucially, when used as a reward model for downstream RLHF, it yields an average relative improvement of 13.6% over the corresponding base policies on IFEval and InfoBench, validating its effectiveness for policy alignment.Our code, data, and checkpoints are available at https://huggingface.co/AIJian/PaTaRM
Large language models (LLMs) have achieved strong performance on natural language to SQL (NL2SQL) benchmarks, yet their reported accuracy may be inflated by contamination from benchmark queries or structurally similar patterns seen during training. We introduce SPENCE (Syntactic Probing and Evaluation of NL2SQL Contamination Effects), a controlled syntactic probing framework for detecting and quantifying such contamination. SPENCE systematically generates syntactic variants of test queries for four widely used NL2SQL datasets—Spider, SParC, CoSQL, and the newer BIRD benchmark. We use SPENCE to evaluate multiple high-capacity LLMs under execution-based scoring. For each model, we measure changes in execution accuracy (Delta Accuracy) across increasing levels of syntactic divergence and quantify rank sensitivity using Kendall’s tau with bootstrap confidence intervals. By aligning these robustness trends with benchmark release dates, we observe a clear temporal gradient: older benchmarks such as Spider exhibit the strongest negative values and thus the highest likelihood of training leakage, whereas the more recent BIRD dataset shows minimal sensitivity and appears largely uncontaminated. Together, these findings highlight the importance of temporally contextualized, syntactic-probing evaluation for trustworthy NL2SQL benchmarking.
The Paradox of Outcome Optimization: A Causal Information-Theoretic Bound on Reasoning Shortcuts in LLMs
PDF ↗Large Language Models (LLMs) aligned via outcome-based Reinforcement Learning (RL) frequently exhibit a critical failure mode: they achieve high performance on in-distribution benchmarks while demonstrating brittle reasoning capabilities on out-of-distribution (OOD) tasks. We term this phenomenon Reward-Induced Manifold Collapse. We establish a theoretical framework bridging Structural Causal Models (SCM) and the Information Bottleneck (IB) principle to explain this paradox. We define reasoning as a high-complexity causal process and shortcut learning as the exploitation of low-complexity spurious correlations. Under the implicit inductive bias of Stochastic Gradient Descent (SGD), models optimized for outcome rewards are biased toward shortcut solutions whenever the training distribution allows for a “Markovian Screening” of the true causal mechanism. We derive a new generalization bound based on Semantic Coverage Measure (\eta) rather than sample size, showing why data scaling on homogeneous distributions may fail to correct reasoning flaws. We also show that Process Reward Models (PRMs) function as Topological Filters, enforcing step-wise mutual information constraints that render the low-complexity shortcut manifold inadmissible. These results provide a mathematical grounding for the role of process supervision beyond simple credit assignment.
Logical reasoning with large language models (LLMs) has made significant progress in recent years. However, existing methods still suffer from insufficient rule semantic grounding and weak rule application mechanisms, making it difficult to achieve precise understanding and effective utilization of rules in complex multi-step reasoning. To address this, we propose Leibniz, a theory-of-mind driven neuro-symbolic reasoning framework. Specifically, we construct a bidirectional reasoning model based on multi-agent collaboration, which characterizes the reasoning process from two complementary perspectives, namely the Evolution Agent and the Reduction Agent. The former transforms belief-unstable propositions into stable ones that are beneficial for proving the target conclusion. The latter performs reverse reduction from the target to resolve belief conflicts and distill new inferential insights. Both share a belief state space and achieve efficient collaborative reasoning through continual belief updating. We integrate natural language and symbolic representations throughout the reasoning process. The experimental results demonstrate that Leibniz surpasses existing state-of-the-art models in reasoning accuracy, and further analyses reveal its substantial advantages in reliability and flexibility.
Existing QA benchmarks typically assume distinct documents with minimal overlap, yet real-world retrieval-augmented generation (RAG) systems operate on corpora such as financial reports, legal codes, and patents, where information is highly redundant and documents exhibit strong inter-document similarity. This mismatch undermines evaluation validity: retrievers can be unfairly undervalued even when they retrieve documents that provide sufficient evidence, because redundancy across documents is not accounted for in evaluation. On the other hand, retrievers that perform well on standard benchmarks often generalize poorly to real-world corpora with highly similar and redundant documents. We present RARE (Redundancy-Aware Retrieval Evaluation), a framework for constructing realistic benchmarks by (i) decomposing documents into atomic facts to enable precise redundancy tracking and (ii) enhancing LLM-based data generation with CRRF. RAG benchmark data usually requires multiple quality criteria, but LLMs often yield trivial outputs. CRRF scores criteria separately and fuses decisions by rank, improving the reliability of generated data. Applying RARE to Finance, Legal, and Patent corpora, we introduce RedQA, where a strong retriever baseline drops from 66.4% PerfRecall@10 on 4-hop General-Wiki to 5.0–27.9% PerfRecall@10 at 4-hop depth, revealing robustness gaps that current benchmarks fail to capture. RARE enables practitioners to build domain-specific RAG evaluations that faithfully reflect real-world deployment conditions.
Although large language models (LLMs) have shown exceptional capabilities across a wide range of tasks, reliable evaluation remains a critical challenge due to data contamination, opaque operation, and subjective preferences. To address these issues, we propose League of LLMs (LOL), a novel benchmark-free evaluation paradigm that organizes multiple LLMs into a self-governed league for multi-round mutual evaluation. LOL integrates four core criteria (dynamic, transparent, objective, and professional) to mitigate key limitations of existing paradigms. Experiments on eight mainstream LLMs in mathematics and programming demonstrate that LOL can effectively distinguish LLM capabilities while maintaining high internal ranking stability (Top-k consistency = 70.7\%). Beyond ranking, LOL reveals empirical findings that are difficult for traditional paradigms to capture. For instance, “memorization-based answering” behaviors are observed in some models, and higher in-family scores are found in the OpenAI model family (\Delta = 9, p < 0.05). Finally, we make our framework and code publicly available as a valuable complement to the current LLM evaluation ecosystem.
Recent progress in large language models (LLMs) has focused on test-time scaling to improve reasoning via increased inference computation, but often at the cost of efficiency. We revisit test-time behavior and uncover a simple yet underexplored phenomenon: reasoning uncertainty is highly localized—only a small subset of high-entropy tokens dominantly affects output correctness. Motivated by this, we propose Minimal Test-Time Intervention (MTI), a training-free framework that enhances reasoning accuracy and stability with minimal overhead. MTI includes: (i) Selective CFG intervention, applying classifier-free guidance only at uncertain positions; and (ii) Lightweight negative-prompt guidance, reusing the main model’s KV cache to approximate unconditional decoding efficiently. MTI yields consistent gains across general, coding, and STEM tasks—e.g., +9.28% average improvement on six benchmarks for DeepSeek-R1-7B and +11.25% on AIME2024 using Ling-mini-2.0—while remaining highly efficient.
Backdoor Collapse: Eliminating Unknown Threats Via Known Backdoor Aggregation In Language Models
PDF ↗Backdoor attacks are a significant threat to large language models (LLMs), often embedded via public checkpoints, yet existing defenses rely on impractical assumptions about trigger settings. To address this challenge, we propose Locphylax, a defense framework that requires no prior knowledge of trigger settings. Locphylax is based on the key observation that when deliberately injecting known backdoors into an already-compromised model, both existing unknown and newly injected backdoors aggregate in the representation space. Locphylax leverages this through a two-stage process: first, aggregating backdoor representations by injecting known triggers, and then, performing recovery fine-tuning to restore benign outputs. Extensive experiments across multiple LLM architectures demonstrate that: (I) Locphylax reduces the average Attack Success Rate to 4.41% across multiple benchmarks, outperforming existing baselines by 28.1%–69.3%. (II) Clean accuracy and utility are preserved within 0.5% of the original model, ensuring negligible impact on legitimate tasks. (III) The defense generalizes across different types of backdoors, confirming its robustness in practical deployment scenarios. Our code is available at https://anonymous.4open.science/r/Locphylax.