Computation and Language
Ask Now, Use Later: Benchmarking the Proactivity Gap in Long-Lived LLM Agents
Bin Wu, Guanyun Zou, Bingbing Wang et al. (+2 more)
A long-lived LLM
agent, such as OpenClaw, earns its value by acting on a user's preferences and constraints across sessions, not just the current request. Yet today's
agents keep what a user volunteers but rarely ask for what stays unspoken, leaving a proactivity gap in long-lived LLM
agents: an
agent cannot act on a preference it never obtained. As users delegate more of their affairs to
agents, the impact of this gap grows. We isolate one concrete, controllable slice of this gap as Ask-to-Remember (ATR): the
agent decides whether to ask now for a reusable user preference that the current task does not need but a later session with the same user will. ATR is hard even to evaluate: the right question is underdetermined and its payoff deferred to tasks that may never arise. ATRBench, to the best of our knowledge the first ATR
benchmark, makes it measurable by fixing each user's preferences as hidden ground truth, so success demands asking, not recall. Across eight frontier LLM
agents, defaults fall at least 62 points below an oracle handed the relevant preference, and prompting closes little of it. Diagnostics identify acquisition as the bottleneck. ATRBench surfaces this proactivity gap in current
agents and offers a diagnostic testbed for closing it.
Read the full abstract ▾
All authors Bin Wu, Guanyun Zou, Bingbing Wang, Huan Zhao, Chuan Shi
Computation and Language
ConRAG: Consensus-Driven Evidence Flow for Multi-Hop Question Answering
Yikai Zhu, Kunfeng Chen, Qihuang Zhong et al. (+2 more)
Retrieval-augmented generation (RAG) has emerged as a promising paradigm for enhancing large
language models (LLMs) on multi-hop question answering (QA), which requires
reasoning over evidence from multiple documents. However, current multi-hop RAG approaches still struggle with evidence selection and information transfer, i.e., retrieving irrelevant passages or failing to pass intermediate information to subsequent
reasoning steps, thereby hindering coherent multi-hop
reasoning. To this end, we propose ConRAG, a consensus-driven evidence flow framework that effectively boosts LLM performance on complex multi-hop QA. In short, the core of ConRAG is to select relevant evidence via consensus-driven multi-view ranking, and leverage a slot-based mechanism to transfer intermediate information between dependent
reasoning steps. Extensive experiments across three multi-hop QA
benchmarks show that ConRAG surpasses the baselines by a clear margin. Encouragingly, on Gemma-4-31B-it, ConRAG outperforms the strongest baseline by 5.8 points in average score, while reducing total end-to-end runtime and
token consumption by 75.9% and 67.3%, respectively.
Read the full abstract ▾
All authors Yikai Zhu, Kunfeng Chen, Qihuang Zhong, Juhua Liu, Bo Du
Computation and Language
SMILE-Next: Teaching Large Language Models to Detect, Classify, and Reason about Laughter
Lee Jung-Mok, Kim Sung-Bin, Joohyun Chang et al. (+2 more)
Laughter is a complex social signal that conveys communicative intent beyond amusement. While prior work has focused on isolated laughter analysis tasks, a comprehensive understanding of laughter in real-world scenarios remains underexplored. Therefore, we introduce SMILE-Next, a
dataset for real-world laughter understanding with
multimodal textual representations and question-answer annotations across three tasks: laughter detection, laughter type classification, and laughter
reasoning. Building upon SMILE-Next, we aim to develop a laughter-specialized large
language model capable of nuanced understanding of laughter in real-world contexts. To this end, we propose two key components: laughter-specific Self-Instruct and the Mixture-of-Laugh-Experts (MoLE) framework. Laughter-specific Self-Instruct enhances generalization across tasks and domains by automatically synthesizing diverse laughter-centric instructions. MoLE introduces a task-adaptive expert routing mechanism that dynamically selects specialized experts tailored to each laughter-related task, improving task-specific performance and efficiency. Experimental results show that the combination of our proposed components substantially outperforms
multimodal LLM baselines, advancing robust real-world laughter understanding. Project page is at: https://mok0102.github.io/smile-next/.
Read the full abstract ▾
All authors Lee Jung-Mok, Kim Sung-Bin, Joohyun Chang, Lee Hyun, Tae-Hyun Oh
Computation and Language
ATLAS: All-round Testing of Long-context Abilities across Scales
Deli Huang, Cunguang Wang, Hongyin Tang et al. (+15 more)
Long-context
language models now advertise context windows up to millions of
tokens, yet evaluations typically report a single length or a narrow task family, masking two failure modes: performance can collapse as length grows, and strong
retrieval need not transfer to downstream use. We present ATLAS, a benchmarking framework that redefines long-context evaluation as length-dependent capability profiling. ATLAS contributes three methodological principles:(i) a layered taxonomy separating foundational operations from application workloads so failures can be attributed, (ii) length-aware AUC scoring that integrates score-length curves over a fixed 8K-1M grid, replacing single-point metrics with full degradation profiles, and (iii) ATLAScore, a harmonic-mean aggregate over taxonomy categories that penalizes imbalanced profiles, with end-to-end uncertainty propagation from subset scores through the nonlinear final aggregate. We instantiate the framework across eight capability dimensions with nine auditable components and 6,438 instances, and evaluate 26 models. Gemini-3.1-Pro-Preview leads at 128K, Claude-Opus-4.6 leads at 1M. Rankings reshuffle substantially between ATLASscore@8K-128K and ATLASscore@8K-1M: 7 models move by at least two ranks, and the two taxonomy layers share only 61% of cross-model variance, with individual rank gaps up to 12 positions. These results support reporting long-context quality by capability and length, not by a single headline score.
Read the full abstract ▾
All authors Deli Huang, Cunguang Wang, Hongyin Tang, Zhe Tang, Linsen Guo, Dongyu Ru, Ruoshi Yuan, Ziyue Zhu, Xiaoyu Li, Ziwen Wang, Chen Zhang, Anchun Gui, Wen Zan, Jiaqi Zhang, Xuezhi Cao, Jingang Wang, Xunliang Cai, Yixin Cao
Computation and Language · Cross-listed from cs.CR
SilentRetrieval: Hijacking Retrieval-Augmented Generation via Semantically-Preserving Adversarial Data Poisoning
Jiachen Qian
Retrieval-Augmented Generation (RAG) mitigates LLM
hallucinations but introduces a critical vulnerability: corpus integrity. We present SilentRetrieval, a two-stage data poisoning attack that hijacks RAG systems through adversarially crafted yet fluent documents. Stage 1 uses Coordinated Beam Search, a multi-
token joint optimization method with a fluency-similarity objective, to keep a poisoned host document retrievable while constraining perplexity. Stage 2 uses Context-Adaptive Trigger Generation, a lightweight trigger-fusion step driven by a frozen LLM, to integrate manipulation triggers into document content. Under a one-poisoned-document-per-query evaluation with synthetic target answers, SilentRetrieval achieves 84.6%/81.3% HR@10 and 57.5%/54.8% ASR-LLM on Natural Questions and MS MARCO, while maintaining near-benign perplexity. Cross-model evaluation across four target LLMs shows nontrivial effectiveness under a fixed trigger generator, and transfer tests against unseen retrievers, including ColBERT and commercial
embedding models, yield 64.7% average HR@10 under the same injected-corpus protocol. In a sampled Wikipedia-scale evaluation, SilentRetrieval retains 74.2% HR@10 at a 0.016% poisoning ratio. Combined
retrieval-side and generation-side defenses reduce attack success substantially but incur a latency trade-off. Human evaluation shows substantially lower flag rates than disfluent baselines, while remaining numerically more suspicious than benign content at the current sample size.
Read the full abstract ▾
Computation and Language
StoryLens: Preference-Aligned Story Rewriting via Context-Aware Narrative Enrichment
Hanwen Cui, Yuting Mei, Yuhang Fu et al. (+2 more)
Story rewriting aims to adapt existing narratives to diverse reader preferences while preserving plot consistency and narrative coherence. Unlike conventional work on style transfer, we argue that effective story rewriting demands context-aware narrative enrichment beyond surface-level stylistic adaptation. Our pilot human study shows that style adaptation alone provides only marginal gains in reader satisfaction (2.3%), while context-enhanced rewriting substantially improves user preference
alignment (24.5%). Motivated by this, we introduce STORYLENSBENCH, a large-scale
benchmark for preference-aligned story rewriting, comprising structured story books, multi-dimensional reader preference profiles, and ranked context-aware rewritten stories. Building on this
benchmark, we propose STORYLENSEVAL, a reward model for estimating reader satisfaction over rewritten stories, and STORYLENSWRITER, a two-stage rewriting model combining supervised fine-tuning with GRPO-based reinforcement learning. We further establish a comprehensive evaluation framework covering fidelity, coherence, and reader satisfaction. Experimental results demonstrate that STORYLENSWRITER consistently outperforms strong generation and personalization baselines, highlighting the importance of context-aware narrative enrichment for personalized story rewriting.
Read the full abstract ▾
All authors Hanwen Cui, Yuting Mei, Yuhang Fu, Dingyi Yang, Qin Jin
Computation and Language
PromptEmbedder: Efficient and Transferable Text Embedding via Dual-LLM Soft Prompting
Yu-Che Tsai, Kuan-Yu Chen, Yuan-Hao Chen et al. (+4 more)
Large
Language Models (LLMs) have demonstrated remarkable efficacy in text
embedding, yet current adaptation methods like LoRA face significant bottlenecks in computational efficiency and cross-architecture transferability. Whenever a new backbone emerges, existing approaches require costly retraining from scratch. To address this, we propose PromptEmbedder, a novel dual-LLM framework that decouples
embedding knowledge from specific backbone weights. PromptEmbedder utilizes a Prompting LLM to generate instruction-aware soft
prompts for a frozen
Embedding LLM via a differentiable generation process with continuous relaxation, ensuring full gradient flow during contrastive training. By localizing task-specific knowledge within the Prompting LLM, adapting to new architectures requires only retraining a lightweight linear
alignment matrix. Evaluations on the MTEB
benchmark show that PromptEmbedder achieves comparable performance with LoRA finetuning while reducing GPU
memory by 40% and accelerating training by 3.7x. Our approach establishes a scalable, architecture-agnostic paradigm for efficient LLM-based representation learning.
Read the full abstract ▾
All authors Yu-Che Tsai, Kuan-Yu Chen, Yuan-Hao Chen, Yu-Han Chang, Ching-Yu Tsai, Yu-Hsiang Chuang, Shou-De Lin
Computation and Language
ConvMemory: A Lightweight Learned Memory Reranker, a Negative Attribution Result, and a Research-Preview Conflict Editor
Taiheng Pan
We describe ConvMemory, a small 3.6M-parameter learned reranker for conversational long-term
memory retrieval, trained with cross-encoder teacher supervision over fused dense and lexical features. On the LongMemEval
memory family, ConvMemory operates above the BGE-large cross-encoder in Recall@10 at 12-47x lower latency, remains within 0.025 Recall@10 of mxbai-rerank-large-v1 on Clean500 while running 28x cheaper; under Stress1000 distractors the Recall@10 gap widens to 0.081 but ConvMemory still operates at 117x lower latency; these LongMemEval numbers are single-run or single-seed and are reported as indicative cost-frontier evidence, not
benchmark-grade. We then publish a rigorous negative attribution result on a previously claimed mechanism: a five-seed retrained ablation with paired bootstrap shows that ConvMemory's learned temporal window is statistically significant on aggregate but not temporally specific, with the largest effects on hard non-temporal controls and no significant effect on multi-hop temporal queries. The honest description of the mechanism is cheap cross-encoder
distillation in a fused dense+lexical feature space, not temporal-structure exploitation. We additionally release CCGE-LA, a low-amplitude conflict-aware candidate-set editor over ConvMemory, as a research preview with modest but consistent gains on supersession and stale/rescue slices on LoCoMo. All results are
retrieval-stage; ConvMemory does not match mxbai-rerank-large-v1 in absolute LoCoMo MRR, and the report is single-author and not yet independently audited.
Read the full abstract ▾
Computation and Language
Challenges in Explaining Pretrained Clinical Text Classifiers
Kristian Miok, Matej Klemen, Blaz Škrlj et al. (+1 more)
Explaining the predictions of neural models in clinical NLP remains a significant challenge, especially for complex tasks involving long, unstructured medical texts. While post-hoc methods like LIME and SHAP are widely used, they often fall short when applied to clinical narratives. In this paper, we identify core limitations of
token-level and perturbation-based explanation techniques through targeted demonstra- tions on a hospital length-of-stay prediction task. Our findings reveal issues such as overemphasis on non-informative
tokens, instability in at- tributions, and high-confidence predictions for incoherent input variants. These results underscore the need for explanation strategies that are clin- ically meaningful, semantically grounded, and robust to linguistic noise.
Read the full abstract ▾
All authors Kristian Miok, Matej Klemen, Blaz Škrlj, Marko Robnik Šikonja
Computation and Language
Prompting Is All You Need: Multi-view Prompting Large Language Models for Aspect-Based Sentiment Analysis
Nils Constantin Hellwig, Niklas Donhauser, Jakob Fehle et al. (+2 more)
Recent work explored the capabilities of Large
Language Models (LLMs) in Aspect-Based Sentiment Analysis (ABSA) through few-shot prompting, requiring substantially fewer annotated examples while achieving notable improvements over zero-shot baselines. However, a performance gap remained compared to models fine-tuned on hundreds of examples, and the computational costs of LLM inference present practical barriers to deployment. We introduce LLM-based Multi-View Prompting (LLM-MvP), which adapts the multi-view principle of considering multiple element orderings to LLM prompting. By combining schema-constrained decoding with a context-free grammar and prefix batching, LLM-MvP achieves performance competitive or superior to fine-tuned approaches while substantially reducing computational overhead. Extensive experiments across five
benchmark datasets demonstrate that LLM-MvP closes the gap between few-shot prompting and fine-tuned models, offering a practical and efficient solution for ABSA.
Read the full abstract ▾
All authors Nils Constantin Hellwig, Niklas Donhauser, Jakob Fehle, Udo Kruschwitz, Christian Wolff
Computation and Language
Knowledge Dependency Estimation for Reliable Question Answering
Chaodong Tong, Qi Zhang, Nannan Sun et al. (+2 more)
Reliable question answering requires identifying not only whether an answer is correct, but also which available knowledge the prediction depends on. In realistic LLM-based QA, this knowledge may come from context,
retrieval, decomposition, or intermediate
reasoning, forming a noisy and redundant candidate space rather than a clean gold evidence set. We study knowledge dependency estimation: estimating the sensitivity of a fixed black-box QA model to different candidate knowledge units. The challenge is to obtain fine-grained dependency scores without exhaustive test-time perturbation while modeling redundancy, substitutability, and complementarity. We propose
Knot, a structured rank-aware knowledge dependency
estimator.
Knot learns from subset-level counterfactual supervision, models subset sensitivity through coverage over latent dependency factors, and derives rank-aware unit scores to identify influential candidates. Across multiple-choice and generative QA
benchmarks,
Knot outperforms all compared baselines in subset-sensitivity prediction and produces more faithful unit rankings than deployable baselines without extra QA-model calls; when used for practical risk screening, its dependency scores help flag error-prone QA predictions early.
Read the full abstract ▾
All authors Chaodong Tong, Qi Zhang, Nannan Sun, Lei Jiang, Yanbing Liu
Computation and Language · Cross-listed from cs.AI
MemCog: From Memory-as-Tool to Memory-as-Cognition in Conversational Agents
Zihan Li, Xingyu Fan, Feifei Li et al. (+1 more)
Existing
agent memory systems universally follow what we term a
Memory-as-Tool paradigm where a single query triggers one-shot
retrieval of flat passage lists, suffering from passive invocation,
reasoning-
retrieval decoupling, and structural mismatch between retrieved fragments and the
agent's navigational needs. We propose MemCog, a
Memory-as-Cognition system that makes
memory access an integral part of the
reasoning process. MemCog organizes user knowledge as Navigable
Memory Store with associative link graphs, exposes Cross-Dimensional
Navigation Interface for multi-step
reasoning-driven traversal, and employs Proactive
Reasoning Protocol that drives
agents to spontaneously initiate
memory exploration from conversational context. We additionally construct ProactiveMemBench, the first
benchmark for evaluating proactive
memory triggering. Experiments show that MemCog achieves state-of-the-art on passive QA
benchmarks (92.98 on LoCoMo, 95.8 on LongMemEval) while substantially outperforming baselines on ProactiveMemBench, demonstrating the advantage of
Memory-as-Cognition.
Read the full abstract ▾
All authors Zihan Li, Xingyu Fan, Feifei Li, Wenhui Que
Computation and Language
Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts
Liu O. Martin, Lucas Bandarkar and Nanyun Peng
Modern large
language models (LLMs) achieve state-of-the-art machine translation performance, but they do so as broad generalists largely trained for many tasks and capabilities unrelated to translation. Thus, they are heavily overparameterized for this task, resulting in excessive
memory and compute requirements. In this paper, we present a method for aggressively
pruning experts from modern mixture-of-experts LLMs while incurring negligible degradation in translation quality. Our approach exploits expert specialization and the separability of
multilingual capabilities in LLMs to identify experts irrelevant to translation. And because of the modular nature of MoEs, these can be easily pruned without any training. Without retraining, we are able to prune half of all experts with negligible degradation and 70% with only minor losses. With a very short SFT, we prune 75% of experts while recovering baseline performance, and in some settings remove nearly 90% while maintaining reasonable translation quality. Overall, our results show that translation requires only a fraction of the LLM, enabling substantial compression of the MoE blocks that contain over 90% of parameters.
Read the full abstract ▾
Computation and Language
Personality, Role, and Expressive Style in Large Language Models: An Interactionist Analysis
Moe Nagao, Koichiro Terao, Mikio Nakano et al. (+1 more)
Prompt-based personality control is a key technique for designing large
language model (LLM) dialogue
agents that behave consistently across social contexts. However, specifying Big Five personality traits (BFTs) in a
prompt does not ensure that the intended traits are expressed in generated utterances. This paper investigates this mismatch from an interactionist perspective, viewing personality expression as a context-dependent outcome shaped by the interplay between trait specification and situational factors. We analyze how perceived BFT expression in LLM-generated dialogue is influenced by three
prompt factors: personality traits, dialogue roles, and expressive styles. Using a factorial design that combines six personality conditions, three roles, and three expressive-style conditions, we generate 1,080 LLM-
agent dialogues in each of English and Japanese. We then evaluate the target
agent's utterances using an LLM-as-a-judge framework to estimate expressed Big Five traits. The results show that expressed personality is shaped not only by explicit trait specification, but also by dialogue role and expressive style. These effects are trait-specific: dialogue role strongly influences Openness, expressive style substantially shapes Conscientiousness and Agreeableness, and explicit trait specification dominates Neuroticism. Even without explicit personality-trait specification, social and expressive conditions induce distinct personality-like impressions. Cross-linguistic comparisons show broadly similar patterns between English and Japanese dialogues, with noticeable differences only under specific combinations of personality, role, and expressive style. These findings suggest that personality control in LLM
agents should be understood not as a direct consequence of trait prompting, but as a context-dependent process involving personality specification, social role, and expressive style.
Read the full abstract ▾
All authors Moe Nagao, Koichiro Terao, Mikio Nakano, Naoto Iwahashi
Computation and Language · Cross-listed from cs.AI
MIRA: A Bilingual Benchmark for Medical Information Response Audit
Mengyu Xu, Qiaoxin Yang, Qianqian Wang et al. (+3 more)
Existing safety evaluations for large
language models overlook whether responses preserve comparable medical information across different user phrasings of the same question. To address this, we introduce the Medical Information Response Audit (MIRA), a bilingual, controlled
benchmark that assesses whether LLMs provide comparable medical information across user-side
language, register, and health literacy signals. MIRA contains 4,320
prompts built from 60 medically reviewed, low-risk health questions. Across five mainstream LLMs, models answered all medical questions, but responses to low health-literacy signals consistently omitted more key information, provided fewer concrete next steps, and offered less support for independent judgment. We term this pattern Differential Information Dilution (DID). A comparison with 300 real-world health queries provides preliminary evidence of rank-order validity. A knowledge-guided mitigation
prompt reduces information dilution for most models, with the largest reductions in underinformative simplification observed for Claude (~8%) and Qwen (~6%). Code and data are available at https://github.com/Rainxu09/MIRA.
Read the full abstract ▾
All authors Mengyu Xu, Qiaoxin Yang, Qianqian Wang, Xiwei Dai, Weiyi Wu, Chongyang Gao
Computation and Language · Cross-listed from cs.CV
VCap: Hypergeometric Rewards for Weak-to-Strong Visual Captioning
Xingyu Lu, Jinpeng Wang, Yi-Fan Zhang et al. (+13 more)
Visual captioning requires models to capture visual content faithfully while minimizing both omission and
hallucination. As the dominant paradigm for captioning, MLLMs have achieved strong performance through scaling and high-quality data. Recently, RL has emerged as a key route to driving MLLMs toward higher precision and broader coverage, however, existing reward designs for captioning fail to provide fine-grained and reliable signals for factual verification, limiting their effectiveness. To address this, we propose VCap, a Witness-Adjudicator reward that pairs the reference caption (a witness) with the visual signal (an adjudicator). By explicitly verifying factual consistency between the reference and policy-generated captions grounded in the visual signal, VCap delivers a reward signal with hypergeometric-distribution-level precision for caption quality verification. This design enables effective learning even from imperfect references, facilitating weak-to-strong generalization in RL training. In our experiments, an 8B model trained with VCap outperforms open- and closed-source SOTA models on multiple image and
video captioning
benchmarks. Human evaluation further confirms its strong
alignment with factual correctness. Additionally, VCap improves MLLM perceptual capability, generalizes across tasks, and surpasses best-of-N
distillation, challenging prior assumptions about RLVR.
Read the full abstract ▾
All authors Xingyu Lu, Jinpeng Wang, Yi-Fan Zhang, Yankai Yang, Yancheng Long, Yiyang Fan, Xuanyu Zheng, Haonan Fan, Kaiyu Jiang, Tianke Zhang, Changyi Liu, Bin Wen, Fan Yang, Tingting Gao, Han Li, Chun Yuan
Computation and Language
Beyond pass@k: Redundancy-Aware RLVR for Multi-Sample Code Generation
Le Bronnec Florian, Alexandre Verine, Rio Yokota et al. (+1 more)
LLMs for code generation are commonly evaluated in repeated-sampling settings using Pass@k, where multiple candidate programs are executed against unit tests under a finite sampling budget. While recent verifier-based reinforcement learning (RLVR) methods improve executable correctness, how these objectives affect redundancy among sampled programs remains poorly understood. In this work, we study implementation-level redundancy in code generation using JPlag, a plagiarism-detection system for code. Across models and
benchmarks, we show that correctness-only RLVR often concentrates generations around repeated implementations, whereas Pass@k-aware objectives maintain lower redundancy and improve larger-budget performance. Motivated by these observations, we augment RLVR with direct anti-redundancy rewards based on JPlag similarity. Across 3 models and 3
benchmarks, discouraging near-duplicate generations reliably improves finite-budget executable performance, often matching or outperforming specialized Pass@k-aware objectives.
Read the full abstract ▾
All authors Le Bronnec Florian, Alexandre Verine, Rio Yokota, Benjamin Negrevergne
Computation and Language
The Missing Piece in Pre-trained Model Evaluation: Reward-Guided Decoding Unlocks Task-Oriented Behavior Without Parameter Updates
Shaobo Wang, Guo Chen, Ziyue Wang et al. (+5 more)
With the rapid progress of large
language models (LLMs), reliably evaluating the capabilities of pre-trained LLMs has become increasingly important. The challenge is that base pre-trained models are optimized for next-
token prediction and often fail to follow instructions or produce well-formed answers under standard prompting and direct decoding. As a result,
benchmark performance can conflate model capability with decoding-induced failures to produce task-oriented outputs, while exposing such behavior often relies on costly post-training. Recent decodingonly approaches attempt to reshape output distributions, but such methods can be inefficient and brittle across open-ended tasks. To address these limitations, we propose Energy-Based Decoding (EBD), a training-free, reward-guided framework for activating task-oriented behaviors from frozen pre-trained LLMs across both open-ended and objective tasks. EBD augments decoding with an external lightweight reward model, steering generations toward high-utility responses while anchoring them to the pre-trained model prior through a reward-tilted target distribution. We show that EBD shifts base-model outputs toward more instructionfollowing behavior, increasing behavioral similarity to post-trained counterparts and enabling a fairer inference-time evaluation of accessible pre-trained-model behavior. Empirically, EBD outperforms baselines across five models and six
benchmarks, improving Qwen3-8B-Base on AlpacaEval2.0 from 8.8 to 44.5, reducing Mistral-7B Math500 latency by 18.9x relative to prior decoding work, and remaining robust to reward-model size.
Read the full abstract ▾
All authors Shaobo Wang, Guo Chen, Ziyue Wang, Zhengyang Tang, Qingyang Liu, Xingzhang Ren, Dayiheng Liu, Linfeng Zhang
Computation and Language
Beyond Imitation: Reflective On-Policy Self-Distillation for LLM Reasoning
Ziqi Zhao, Xinyu Ma, Liu Yang et al. (+6 more)
On-policy self-
distillation (OPSD) improves the
reasoning capabilities of large
language models (LLMs) by providing dense
token-level supervision for on-policy rollouts. However, existing OPSD methods often yield limited gains on complex
reasoning tasks and suffer from severe training instability. We identify two key causes: conditioning the self-teacher on a complete verified solution encourages imitation of complete reference trajectories rather than extraction of transferable
reasoning insights, while indiscriminate full-response
distillation imposes superfluous supervision on already-valid
reasoning prefixes. Together, these issues suppress
reasoning diversity and contribute to late-stage mode collapse. We propose Reflective On-policy Self-
Distillation (ROSD), which distills transferable
reasoning insights rather than complete reference trajectories. For each erroneous rollout, a self-reflector contrasts it with a correct rollout from the same group to derive a corrective idea and identify the sentence containing the first
reasoning error. The corrective idea provides the self-teacher with targeted guidance, while the diagnosed error boundary allows ROSD to mask out the
distillation loss over the valid prefix and apply
token-level
distillation only from the first erroneous sentence onward. Experiments across multiple
reasoning benchmarks and model backbones show that ROSD consistently outperforms standard OPSD and reinforcement learning baselines, better preserves
reasoning diversity, stabilizes training, and mitigates late-stage mode collapse. Code is available at https://github.com/ZiqiZhao1/ROSD.
Read the full abstract ▾
All authors Ziqi Zhao, Xinyu Ma, Liu Yang, Yujie Feng, Daiting Shi, Jingzhou He, Xin Xin, Zhaochun Ren, Xiao-Ming Wu
Computation and Language
KSAFE-MM: A Multimodal Safety Benchmark via Localized Contextualization for Korean Cultural Risks
Yongwoo Kim, Sojung An, Yunjin Park et al. (+8 more)
Multimodal Large
Language Models (MLLMs) exacerbate safety risks by introducing vulnerabilities across multiple modalities, such as
language and
vision. Current MLLM safety evaluation tools, however, suffer from major limitations: 1) English-centric
dataset construction, and 2) a focus on generic risks that are not tied to local cultural contexts. This paper introduces KSAFE-MM, a
benchmark for Korean
multimodal safety evaluation that covers both general safety risks and culture-specific vulnerabilities. KSAFE-MM consists of two complementary parts: KSAFE-MM-G evaluates globally shared risks in Korean contexts through linguistic contextualization, which transforms generic safety queries into contextually grounded
multimodal samples. In contrast, KSAFE-MM-C targets safety vulnerabilities that are culture-dependent, using localized visual queries drawn from real-world. It pairs these visual queries with jailbreak-style textual queries to cover
multimodal safety risks involving cultural visual cues and malicious textual intent. We evaluate 12 state-of-the-art MLLMs on KSAFE-MM and reveal culturally grounded vulnerabilities that translation-based evaluation fails to capture. Notably, jailbreaking strategies substantially amplify attack success rates, with ProgramExecution yielding up to 74.2% ASR compared to 13.4% for standard queries. Furthermore, we identify a systematic trade-off between safety and over-refusal, where models achieving low ASR tend to exhibit excessive refusal behavior on benign queries. These findings highlight the urgent need for culturally grounded safety evaluation beyond English-centric
benchmarks.
Read the full abstract ▾
All authors Yongwoo Kim, Sojung An, Yunjin Park, Jungwon Yoon, Dujin Lee, HyunBeom Cho, Jaewon Lee, Wonhyuk Lee, Youngchol Kim, JeongYeop Kim, Donghyun Kim
Computation and Language
MemGuard: Preventing Memory Contamination in Long-Term Memory-Augmented Large Language Models
Hyeonjeong Ha, Jeonghwan Kim, Cheng Qian et al. (+7 more)
Memory-augmented large
language models extend
reasoning beyond a fixed context window by maintaining long-term
memory across interactions. However, existing
memory systems often collapse stable user facts, episodic events, and behavioral rules into a shared space, allowing functionally distinct
memories to be retrieved and used as interchangeable evidence. We identify this failure mode as heterogeneous
memory contamination, where context-specific events become overgeneralized claims, or semantically relevant but functionally incompatible
memories mislead generation. To this end, we introduce MemGuard, a type-aware
memory framework that preserves functional
memory boundaries during
memory construction and
retrieval. It assigns each
memory an explicit functional role at write time, maintains relations across type-isolated
memories, and selectively composes evidence only from necessary
memory types, reducing contamination from irrelevant or functionally incompatible evidence. Across
hallucination and long-horizon conversation
benchmarks, MemGuard improves
memory reliability by up to 28.27% while retrieving up to 5.8x fewer
memory tokens than prior methods. These results suggest that reliable long-term
reasoning depends on principled organization and selective use of heterogeneous
memory.
Read the full abstract ▾
All authors Hyeonjeong Ha, Jeonghwan Kim, Cheng Qian, Jiayu Liu, William M. Campbell, Yue Wu, Yuji Zhang, Kathleen McKeown, Dilek Hakkani-Tur, Heng Ji
Computation and Language
Integrated and Cross-Architecture Interpretation of LLM Reasoning
Leonardo Matthew Yauw, Wei-Bin Kou and Yujiu Yang
Understanding how LLMs reason is hindered by a practical asymmetry: while their generated outputs are observable, the underlying
reasoning patterns remain opaque. Relying on single probes, such as Mutual Information Peak (MIP) or Deep-Thinking Ratio (DTR), risks underestimating the genuine inferential structure. To response this deficiency, we present an Integrated, cross-Architecture
Reasoning (IAR) framework, designed to provide a unified approach to LLM
reasoning interpretability. Specifically, we first propose to use bandwidth-calibrated MIP coupled with Tukey IQR peak-detection to isolate
reasoning-crucial
tokens at the output layer. Second, we performed an overlap analysis between MIP-picked
tokens and DTR-deep
tokens to trace the cross-layer trajectories of those
tokens. This also discloses whether
reasoning-crucial
tokens are computation-intensive as well, further facilitating to understand how
reasoning patterns evolve across model layers. Finally, we apply a Jaccard stability metric over multi-domain problems to verify if the MIP-identified
tokens are
reasoning quality-guaranteed. Extensive experiments on six models across four domains (mathematics, code, logic, and common sense) demonstrate IAR's generalizable interpretation capabilities across architectures. Our code is available at https://github.com/LeonardoMatthew/IAR.
Read the full abstract ▾
Computation and Language
Beyond Chunk-Local Extraction: Cross-Chunk Graph Augmentation for GraphRAG
Jiaming Zhang, Yibo Zhao, Jing Yu et al. (+2 more)
GraphRAG extends
retrieval-augmented generation by organizing corpora as explicit knowledge graphs, enabling graph-based
retrieval for complex question answering. However, existing frameworks extract entities and relations within individual chunks, leaving cross-chunk relations – those whose evidence spans multiple passages – systematically absent from the index. Exhaustive LLM-based recovery of such relations is impractical due to the combinatorial explosion of chunk combinations. We present CrossAug, a GNN-guided CROSS-Chunk Graph AUGmentation method that enriches GraphRAG indices with cross-chunk relational structure as an offline step before query-time
retrieval. CrossAug derives training supervision through self-supervised graph corruption, uses a
topology-aware GNN to score subgraphs for missingness, and applies evidence-grounded LLM completion only to selected high-scoring regions. Experiments on three LLM-based GraphRAG frameworks across four multi-hop and long-document QA
benchmarks demonstrate that CrossAug consistently improves performance, confirming the benefit of cross-chunk graph augmentation for
retrieval-based question answering. Our code is available at https://github.com/DonFinliani/CrossAug.
Read the full abstract ▾
All authors Jiaming Zhang, Yibo Zhao, Jing Yu, Jianxiang Yu, Xiang Li
Computation and Language
ResearchMath-14K: Scaling Research-Level Mathematics via Agents
Guijin Son, Seungyeop Yi, Minju Gwak et al. (+3 more)
The frontier of mathematics is defined by problems whose solutions are not yet known. However, whether
language models can meaningfully engage with such problems without human intervention remains unclear. A major obstacle is the lack of large-scale research-level math
datasets. To this end, we introduce ResearchMath-14k, a set of $14{,}056$ problems curated from academic sources via a multi-
agent pipeline. ResearchMath-14k spans 11 mathematical domains and ranks above existing math
datasets on knowledge, novelty, and procedural difficulty. To our knowledge, it is the largest research-level mathematical problem set available for training. We additionally generate $220$K teacher trajectories through targeted prompting, followed by behavioral filtering. Notably, however, generating correct trajectories is nontrivial at this level, and two LLM judges label only $3.7\%$ and $4.3\%$ of sampled ResearchMath training trajectories as correct. Nevertheless, across three model families, full-parameter training on ResearchMath improves performance on graduate- and research-level mathematics
benchmarks by $2.1$ points over the starting checkpoints. In comparison, training on existing
datasets such as DASD and Nemotron-SFT-Math-v4 changes performance by $0.0$ and $-0.5$ points, respectively. Notably, mixing DASD with ResearchMath yields higher scores than
token-matched DASD alone on
benchmarks covering olympiad short-form ($+2.0$), graduate- and research-level short-form ($+0.8$), graduate- and research-level symbolic ($+2.6$), and proof evaluation ($+5.7$). Further analysis suggests that research-level mathematical content and greater
reasoning diversity may help explain why ResearchMath provides complementary supervision to contemporary
datasets. We make ResearchMath-14k publicly available for future works on research-level mathematical
reasoning.
Read the full abstract ▾
All authors Guijin Son, Seungyeop Yi, Minju Gwak, Hyunwoo Ko, Wongi Jang, Youngjae Yu
Computation and Language
Where Does Toxicity Live? Mechanistic Localization and Targeted Suppression in Language Models
Himanshu Beniwal and Mayank Singh
Large
language models frequently generate toxic, hateful, or harmful content, yet existing mitigation methods rely on costly retraining or output-level filtering with no mechanistic insight into where toxicity originates internally. We introduce Meow2X and TRNE, two complementary retraining-free frameworks that localize toxicity to specific layers and
neurons by analyzing activation differentials between toxic and neutral
prompts, then suppress them via inference-time scaling or minimal rank-one weight edits – without any gradient descent. Evaluations across five LMs, two
benchmarks, and 90 configurations using dual safety evaluators demonstrate consistent toxicity reduction while preserving
language modeling quality. Our analysis reveals that toxicity is disproportionately encoded in early MLP layers, varies across architectures, and is systematically underestimated by single-evaluator setups – underscoring the need for multi-evaluator safety assessment. By bridging mechanistic interpretability with practical detoxification, our framework offers a principled path toward safer, more transparent
language models.
Read the full abstract ▾
Computation and Language
Rethinking Visual Neglect: Steering via Context-Preference for MLLM Hallucination Mitigation
Jingwen Wu, Xijun Zhang and Ge Song
Object
hallucination remains a primary obstacle to the reliable deployment of
Multimodal Large
Language Models (MLLMs). Current inference-time mitigation methods mainly assume
hallucinations stem from visual neglect, steering models to enhance visual reliance. In contrast, our systematic interventions on multiple MLLMs show that pushing toward more visual reliance may exacerbate
hallucinations on some models, while less may mitigate
hallucinations. This result suggests that attributing
hallucinations solely to visual insufficiency is underdetermined. We argue that the image, as a context, simultaneously competes with the model's parametric knowledge and the textual context. For this, we propose a training-free framework, Context-Preference Activation Steering (CAS). It extracts two semantically distinct Context Preference Vectors (CPVs) via two small sets of designed conflict samples and applies them via single-pass signed residual injection at mid-early MLP layers during inference to control information reliance. Experiments show that CAS substantially mitigates object
hallucinations without increasing decoding latency and preserves native text-generation quality.
Read the full abstract ▾
Computation and Language
Auditing Stance Asymmetry in Generative Explanations
Jiarui Han
Bias evaluation for
language models has made substantial progress on bounded comparisons, such as overt derogation, stereotype association, or label-sensitive differences under controlled substitutions. Open-ended explanations raise a different problem: they guide interpretation by assigning responsibility, legitimacy, context, and grievance. A model can avoid hostile
language while making one side structurally understandable and another personally at fault, overreacting, or less worth taking seriously. We call this stance-bearing asymmetry in generative explanations. We propose
Symmetry Decomposition Evaluation (SDE), which tests paired situations with concrete group labels, structural-role rewrites, and explicit support or counter-evidence. In a controlled 32-family prototype suite, this decomposition shows that surface differences are not all alike: some weaken under structural or evidence control, while others remain as stable differences in how the model assigns blame, context, or legitimacy. Targeted case review and judge comparison suggest a broader difficulty for evaluating open-ended framing asymmetries: judge readings shift across operationalizations, and scalar scores can flatten distinctions that readers use to interpret explanatory stance. SDE therefore reframes generative bias evaluation as an audit of explanatory stance – what stance each side receives, how it changes under decomposition, and where automatic scoring becomes unstable.
Read the full abstract ▾
Computation and Language
An Evolutionary Approach for Designing Stable and Highly Expressible Low-Immunogenicity Therapeutic mRNA Sequences
Dhawa Sang Dong, Mausam Gurung and Suraj Kandel
Messenger RNA (mRNA) sequences as therapeutics require optimized design to ensure efficient translation, structural stability, and minimal immunogenicity. This study presents a two-stage in-silico framework that integrates deep learning and evolutionary computation for rational mRNA optimization instead of existing state-of-the-art models. In the first stage, a pretrained CodonTransformer (BERT-like Large
Language Model) generates biologically coherent mRNA sequences encoding the target antigen. In the second stage, a genetic
algorithm (GA) evolves these candidate sequences through codon-aware crossover and synonymous mutation guided by human codon usage preferences. Fitness functions for evaluation combined translation-related metrics (CAI, tAI, codon-pair bias), mRNA structural stability (local and global MFE via RNAfold, GC content), and reduced immunogenicity (CpG/UpA motif frequency). Over successive generations (38th, 40th, and 42nd), the GA improved (achieved CAI values of 0.73 to 0.74 and tAI values of 0.63 to 0.64) CAI and tAI by over 6% and codon-pair bias is high and consistent (0.97 ) and improved ribosomal accessibility at the 5' end, with an unpaired_30 fraction reaching 0.87; Global Minimum Free Energy (MFE) converged to a balanced range of -346 to -356 kcal/mol, achieving approximately 84% base-paired structural stability, and reduced immune-stimulatory motifs - lowering the average immune penalty to 27.3 in the final generation. Linear Design produces hyper-stable transcripts (MFE < - 2000 kcal/mol) that risk translation inefficiency due to extreme
rigidity, and BiLSTM-CRF focuses solely on high CAI (0.96 to 0.98) without structural constraints, our framework achieves an optimal translation-stability
equilibrium, highlighting the proposed BERT-GA framework as an effective, data-driven approach for the design and optimization of in-silico mRNA sequences.
Read the full abstract ▾
Computation and Language
KVoiceBench, KOpenAudioBench, and KMMAU: Agent-Driven Korean Speech Benchmarks for Evaluating SpeechLMs
Haechan Kim, Seungjun Chung, Inkyu Park et al. (+2 more)
Speech language models (SpeechLMs) have achieved substantial progress by extending large
language models (LLMs) to the
speech modality. However, SpeechLM evaluation remains heavily centered on English, limiting reliable assessment of
multilingual speech capabilities. Straightforward
benchmark transfer through ASR, translation, normalization, and TTS can corrupt
language-specific instructions, answer constraints, and spoken forms; for audio understanding, transferring source-
language audio also fails to preserve target-
language speaker attributes, accents, and paralinguistic properties. To address these limitations, we propose two human-
agent benchmark-construction frameworks: one transfers source-
language SpokenQA
benchmarks into target-
language SpokenQA
benchmarks, and the other converts target-
language ASR corpora into audio understanding
benchmarks using transcriptions and speaker metadata. Using these frameworks, we construct and publicly release three Korean
speech benchmarks: KVoiceBench and KOpenAudioBench for Korean SpokenQA, and KMMAU for Korean audio understanding, comprising 12,345 samples in total. We evaluate eight recent SpeechLMs and find that English-Korean performance gaps vary substantially across models and task families, and that SpokenQA and audio understanding rankings diverge, revealing complementary weaknesses invisible to English-only evaluation.
Read the full abstract ▾
All authors Haechan Kim, Seungjun Chung, Inkyu Park, Jihoo Lee, Jonghyun Lee
Computation and Language
Periodic RoPE for Infinite Context LLMs
Simin Huo
The ability to process ultra-long contexts is crucial for large
language models (LLMs) to perform long-horizon tasks. While recent efforts have extended context windows to 1M and beyond, model performance degrades when sequence length exceeds the pre-trained range of positional encodings (e.g., RoPE), i.e., position exhaustion. This fundamental limitation must be overcome to achieve a truly infinite context. To address it, we propose Periodic RoPE (P-RoPE), a positional encoding mechanism designed to circumvent this exhaustion. It operates in conjunction with sliding window
attention (SWA) to capture local dependencies and relative positions within each window. This local layer is then complemented by a global
attention layer with No Positional Encoding (NoPE), enabling unbounded interaction across the entire sequence without positional constraints. By stacking these two types of layers, the model avoids the need for positional extrapolation to generalize longer and theoretically supports an infinite context window. Empirical results show that our model, MiniWin, outperforms MiniMInd with standard GPT architectures in long-context efficiency and stability. Our work provides a possible pathway toward LLMs with genuine infinite-context understanding. The code is available at \href{https://github.com/Cominder/miniwin}{https://github.com/Cominder/miniwin}.
Read the full abstract ▾