跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+1072新增论文
+845其中 AI 相关
+94已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-09 20:39 · 全站最近入库：2026-11-09 13:22各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 21:32 — 2026-11-09 21:32（北京时间） · 点击数字查看对应简报；AI 相关包含在新增论文内。

最近完成检查：2026-11-09 20:39最近发现新增版本：2026-11-09 13:22库内论文最新更新时间：2026-11-07 02:59

每分钟刷新已保存的数据。空检查不代表采集异常；新收录可能包含历史论文。AI 导读按新增版本计数。

18,172 已收录论文15,506 AI 相关4,527 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

时间范围跟随排序：版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常，查看状态 →

找到 5,189 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 20:38（北京时间）
版本更新 · v4arXiv:2509.23166

Test-Time Policy Adaptation for Enhanced Multi-Turn Interactions with LLMs

Chenxing Wei, Hong Wang, Ying He, Fei Yu, Yao Shu

原文摘要

Large Language Models (LLMs) employ multi-turn interaction as a fundamental paradigm for completing complex tasks. However, their performance often degrades in extended interactions, as they are typically trained on static, single-turn data, which hinders their ability to adapt to real-time user feedback. To address this limitation, we first propose a new paradigm: Test-Time Policy Adaptation for Multi-Turn Interactions (T2PAM), which utilizes user feedback from the ongoing interaction as a reward signal to estimate a latent optimal policy aligned with user preferences, then updates a small subset of parameters to steer the model toward this policy, ultimately enabling efficient in-conversation self-correction. We then introduce Optimum-Referenced One-Step Adaptation (ROSA), a lightweight algorithm that operationalizes T2PAM. ROSA guides the model parameters toward a theoretical optimal policy in a single, efficient update step, avoiding costly iterative gradient-based optimization and minimizing computational overhead. We provide a rigorous theoretical analysis guaranteeing that the policy of ROSA converges to the preference of user as the number of interactions increases. Extensive experiments on challenging benchmark demonstrate that ROSA achieves significant improvements in both task effectiveness and efficiency.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2512.06869

Role-aware Heuristic Episodic Attention for Conversational LLMs

Wanyang Hong, Zhaoning Zhang, Yi Chen, Libo Zhang, Baihui Liu, Linbo Qiao, Zhiliang Tian, Dongsheng Li

原文摘要

Large language models often lose track of persistent instructions and relevant information as multi-turn conversations grow. We study this cumulative contextual decay through three related failure modes: attention pollution, dilution, and drift. We propose REA(Role-aware Heuristic Episodic Attention), a context-management framework that assigns different persistence and representation policies to instructions and episodic interactions. Instructional Memory retains identified global constraints in a dedicated prefix. Episodic Memory preserves user inputs and compresses model replies, while heuristic retrieval selects raw text, compressed representations, or omission for each historical turn. On Long-MT-Bench+, REA improves the judge score from 6.32 to 7.36 on a 10-point scale, a 16.5% relative gain over the Vanilla baseline, and reduces average latency by 2.91x. Additional evaluations show aggregate gains on three backbones spanning 1.7B-7B parameters and on Chinese and English role-playing tasks. These results support role-aware context management as a practical approach to maintaining conversational continuity and instruction adherence.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.04971

Why Geometric Continuity Emerges in Deep Neural Networks: Residual Connections and Rotational Symmetry Breaking

Kyungwon Jeong, Won-Gi Paeng, Honggyo Suh

原文摘要

Weight matrices in deep networks exhibit geometric continuity---principal singular vectors of adjacent layers point in similar directions. While this property has been widely observed, its origin remains unexplained. Through experiments on toy MLPs and small transformers, we identify two mechanisms: residual connections create cross-layer gradient coherence that aligns weight updates across layers, and symmetry-breaking nonlinearities constrain all layers to a shared coordinate frame, preventing the rotation drift that would otherwise destabilize weight structure. Crucially, a nonlinear but rotation-equivariant activation reaches less than half of ReLU's continuity, isolating symmetry breaking---not nonlinearity alone---as the active ingredient. Activation and normalization play distinct roles: activation concentrates continuity in the leading singular direction, while normalization distributes it across multiple directions. In transformers, continuity is projection-specific: Q, K, Gate, and Up (which read from the residual stream) develop input-space ($\mathbf{v}_1$) continuity; O and Down (which write to it) develop output-space ($\mathbf{u}_1$) continuity; V alone, lacking an adjacent nonlinearity, develops only low continuity.

cs.LG · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.31303

The Value of Memory in Language Agents

Zehao Lin, Erenkai Zhu, Helan Hu, Xixuan Hao, Chunyu Li, Kai Chen

原文摘要

Memory enables language agents to carry useful distinctions from past experience into future decisions. Its value depends on which distinctions survive, whether a bounded agent can use them, and what can be recovered or revised as demands evolve. We develop a decision-theoretic framework that follows this value from raw history through persistent state, operational use, and costly recovery and revision. Original and retained value are expressed in units of avoided task loss; risk-cost frontiers describe attainable recovery and preservation across uses. Unknown demand yields value profiles, while interactive feedback changes the distribution of future work. A cost-aware simulation condition and five minimal counterexamples expose the roles of information timing, retrieval access, component interaction, feedback and demand uncertainty. In particular, a pointer with zero mutual information about the answer can still reduce the cost of exact recovery. Diagnostics on LoCoMo, a two-project update task and longitudinal cases exhibit failures of preservation, use and measurement. At fixed retrieved evidence, a saved ledger changes subsequent answers; at fixed agent outputs, an evaluator revision changes scores. The framework connects these distinctions to operational risk bounds and testable principles for allocating memory, preserving recovery options and revising evidence dependencies.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.32151

The LLENS Platform for On-Premises Large Language Model Agents in a Japanese University Hospital: Development and Evaluation Study

Ken Enda, Jien Saito, Masami Mukai, Feng Han, Emir Turkes, Tatsumi Kusakabe, Reona Umeki, Kazuhiro Sumiyoshi 等 15 位作者

原文摘要

Large language model agents that use tools could take on part of the electronic health record and desk work that occupies physicians' time, but in many hospitals the medical information systems run inside closed networks, so the model and the agent harness must also be inside the network. We designed, implemented, and deployed LLENS, a clinical support AI agent platform that runs inside the closed network of Hokkaido University Hospital on a single host with 8 NVIDIA H200 GPUs, the open-weight model Kimi K2.6 (~1 trillion parameters), and a web application rebuilt from Open WebUI. The harness has 10 components, including page image reading, full-text attachment input, a path-addressed document tree without a vector index, a network-isolated sandbox, artifact return, self-checking of generated output, subagents, and saved procedures. Kimi K2.6 was chosen from 3 open-weight candidates and scored 480/500 on the 2026 Japanese Medical Licensing Examination; it generated 119 tokens/s for a single user and served 16 concurrent requests at 908 tokens/s in total. On 180 questions over hospital documents (3 seeds each), mean accuracy was 10.7% with no documents, 81.3% with fixed retrieval (top 10 chunks of one lexical search), and 97.2% with interactive access, in which the model reads with document tools and the sandbox (15.9 percentage points above fixed retrieval, 95% CI 10.9-21.3); the difference lay in filtering/counting questions (16.7% vs 87.8%). In an automated test of 13 work scenarios (4 runs each), the expected components were triggered in all runs for 12, and 5 scenarios were run with synthetic data on the deployment host. The platform has run on the hospital network since September 14, 2026.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2609.01198

FinLifeBench: Exhaustive Life-Event History and Financial-State Reconstruction from Longitudinal Banking Dialogue

Hangyeul Lee, Juyoung Oh, Jaeyong Ko, Sunmin Kim, Jaeik Park, Hyunkyu Kim, Jungmin Son, Pilsung Kang

原文摘要

Repeated banking interactions require assistants to maintain complete, current, and traceable customer records as life changes emerge incidentally in routine requests. Existing benchmarks emphasize question answering, bounded episodes, or targeted recall rather than exhaustive longitudinal reconstruction. We introduce FinLifeBench, which evaluates two tasks over the same cumulative dialogue: reconstructing every life-event instance with its first-establishing session and reconstructing a complete 34-path financial state at consecutive checkpoints. The benchmark contains 6,000 eight-turn Korean banking sessions from 20 independent synthetic trajectories, with deterministic, exhaustive gold for 24 event types and 34 state paths and consensus quality assurance. Across eleven LLMs under a full-context condition, event-anchor recall falls from 0.591 at 15 sessions to 0.445 at 300. Errors are driven primarily by omitted events rather than poor anchor localization, while financial-state reconstruction frequently treats superseded or potentially outdated information as current; the best GCA@15 reaches 0.470. Performance on the two reconstruction tasks is only weakly associated. These results show that models can localize evidence for recovered events while still failing to maintain complete and temporally valid longitudinal records.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34809

Position Matters More Than Channel: Talking a Language Model Out of a Correct Answer

Xinyi Ni

原文摘要

Language models now answer questions alongside other claims: a note in the system prompt, a message from another agent, their own earlier answer. We ask what it takes to talk a model out of an answer it gets right on its own, and separate three factors that earlier studies vary together: where the advice sits in the prompt, who is said to have written it, and what the model has already written. On Qwen3.5-27B and four-option MMLU-Pro items, with the rest of the advice held fixed in each comparison, we find three things. First, position matters far more than the API channel, in opposite directions for a bare claim and an argument: a bare "the answer is X" does most harm just before the model answers, while the same claim with a false rationale does most harm before the question, where it raises error on items the model otherwise almost never misses from 1% to 67%; moving the claim between the system message and the user message at the same position changes error by a few points. Second, attributing the note to an authority rather than to another participant, and phrasing it assertively, each raise error by 10-27 points, while the two authorities we tested differ by under two. Third, replaying a correct worked answer before the advice protects it whether the text is shown as the model's own or as someone else's, and the same replay keeps wrong answers from being corrected by a bare correct note. The position result replicates under an independently written prompt. Code, items and all result tables are released at https://github.com/coolkillercat/safety.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.00326

Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification

Charles Weng, Dingwen Li, Alexander Martin

原文摘要

Safety pipelines often block content by thresholding a model's first-token unsafe score. We show that a single prompt's score is not a reliable probability and compare two repairs: mean ensemble and calibration with development labels. Across seven vision-language models and two benchmarks, an input's unsafe probability varies across 15 task-preserving rewordings with a standard deviation of 0.16-0.30 on average; the inputs that vary most are more often misclassified. The 15-prompt mean has lower negative log-likelihood (NLL) than the best single prompt, even one chosen with evaluation labels, on 13 of 14 model-benchmark pairs. Development labels change the picture: jointly selecting and calibrating a single prompt outperforms the similarly calibrated mean for all seven models on held-out data from the development benchmark. Under cross-benchmark evaluation without refitting, however, the advantage reverses, with the calibrated mean outperforming the single prompt on six of seven models. We recommend evaluating first-token scores as a prompt family, using the mean ensemble as the label-free baseline, calibrating one prompt when development labels exist and deployment data match, and calibrating the mean when they may not.

cs.CL · cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34791

Text Is the Bottleneck: A Small-Scale Replication of LLM Game Playing in Riichi Mahjong

Mingge Xu

原文摘要

We replicate the riichi-mahjong track of "Can Large Language Models Master Complex Card Games?" (arXiv:2509.01328) under a drastically smaller budget: models of at most 2B parameters, one 16 GB consumer GPU, about five GPU-days. (1) With matched data and labels, text is the bottleneck: on the same 100k-decision training set, a 2.6M-parameter convolutional board encoder agrees with the strong AI Mortal on 80.6% of test decisions, above every text-prompt LLM we train (best 77.8%), and places better at the table (2.85 over 1,000 games vs. ~3.0 for the LLMs; paired difference at n=100 not significant). Scaled to 1.71M decisions, a 4.9M-parameter Transformer reaches 94.7% agreement, beats the 0.8B LLMs on identical seeds (-0.45 ranks, sign test p=0.005), and is statistically equivalent to parity against three Mortal champions (TOST within 0.10 ranks, p=6e-4); parity against a teacher-opponent is roughly the ceiling of pure imitation. (2) Training toward the teacher raises teacher agreement more than a 2.5x parameter increase does (+2.5 vs. +1.5 points); on human-record agreement both effects are +1.0, and neither is visible at the table. (3) Few-point agreement differences do not resolve into playing strength within a family: +2.5 agreement points coincide with a paired placement difference of +0.02 +/- 0.12 ranks (p=1.0), and +2.0 points at 2B with -0.06 (p=0.37) - both below the ~0.25-rank resolution of 100 paired games. We also document a paper-code mismatch in the original winner filter, reconstruct its two private engine patches, and show that a divide-and-conquer pipeline makes template-generated Japanese rationales 93% sentence-exact (43% without facts). Per Tenhou's terms we release no trained weights - only code, configurations, and evaluation protocols.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34790

BehaviorMem: Building and Evaluating User Memory from Real-World E-Commerce Behavior Streams

Sizhe Wang, Xinghao Fang, Rui Qiu, Zhenjia Ma, Dagui Chen, Kun Kuang, Hongcheng Guo

原文摘要

User memory helps personalized systems build on their understanding of users over time. In e-commerce, users convey their preferences and needs through both explicit language and implicit interactions with products. We study memory built from logs of these interactions, examining how it can \emph{remember} concrete evidence and established interests, and use that evidence to \emph{infer} underlying needs and preferences that inform judgments beyond well-observed product domains. Existing memory methods primarily target language-based interactions, and evaluations offer limited real behavioral data and limited separation of these capabilities. This leaves open how to jointly support and separately evaluate remembering and inference from behavioral logs. We formulate the problem of continual behavioral user memory around these joint requirements and develop a dedicated evaluation framework and memory method to address them. \textbf{BehaviorMemBench} is a benchmark built from real e-commerce logs covering 1{,}194 users and 17{,}462 behavioral batches over more than three years, with automatic future-behavior labels and a task-independent memory interface to separately evaluate remembering and inference under a common protocol. \textbf{BehaviorMem} is a memory design that organizes explicit behavioral facts and inferred user characteristics across long-term and short-term scopes in a two-axis state, with evidence-grounded formation, continual maintenance, and query-independent readout. Experiments across two builders and two task readers show that BehaviorMem leads the evaluated agentic memories on remember and infer, and reveal different method rankings across the two capabilities and competitive statistical baselines. Validation on a public dataset further supports its usefulness for downstream tasks based on behavioral histories.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34785

Natural Jammers: When Non-Robust Features Become Antagonistic

Jiaming Zhang, Fuyao Zhang, Pengjun Xie, Wei Yang Bryan Lim

原文摘要

Deep neural networks are known to rely on non-robust features that align poorly with human perception. Prior work has largely framed these features as opportunistic shortcuts that help standard models on clean data. We show instead that non-robust cues can be antagonistic: they can act as Natural Jammers that actively mislead standard models on clean, unmodified images. Using disagreement between a standard model and robust models as an analysis lens, we identify a Jammed Set on which robust models outperform standard models. We provide causal evidence for Spectral-spatial Locking: jamming arises only when high-frequency components are precisely aligned with specific spatial structures. Disrupting this lock through spectral filtering or small geometric shifts can restore correct predictions. Crucially, we uncover a Substitution Paradox: while individual jammers are fragile to geometric shifts, the aggregate prevalence of jamming remains approximately stable and is redistributed rather than eliminated under such transformations. Our results reframe natural non-robustness as a failure field over the empirical model-data geometry: one that redistributes rather than disappears under small transformations.

cs.CV · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v5arXiv:2510.15552

Think Parallax: Solving Multi-Hop Problems via Multi-View Knowledge-Graph-Based Retrieval-Augmented Generation

Jinliang Liu, Jiale Bai, Shaoning Zeng

原文摘要

Large language models (LLMs) still struggle with multi-hop reasoning over knowledge-graphs (KGs), and we identify a previously overlooked structural reason for this difficulty: Transformer attention heads naturally specialize in distinct semantic relations across reasoning stages, forming a hop-aligned relay pattern. This key finding suggests that multi-hop reasoning is inherently multi-view, yet existing KG-based retrieval-augmented generation (KG-RAG) systems collapse all reasoning hops into a single representation, flat embedding space, suppressing this implicit structure and causing noisy or drifted path exploration. We introduce ParallaxRAG, a symmetric multi-view framework that decouples queries and KGs into aligned, head-specific semantic spaces. By enforcing relational diversity across multiple heads while constraining weakly related paths, ParallaxRAG constructs more accurate, cleaner subgraphs and guides LLMs through grounded, hop-wise reasoning. On WebQSP and CWQ, it achieves state-of-the-art retrieval and QA performance, substantially reduces hallucination, and generalizes strongly to the biomedical BioASQ benchmark.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34768

DynEdit: Dynamic Entropy-Guided Sequential Editing for Large Language Models

Jinhu Fu, Yan Bai, Yihang Lou, Li Sun, Jiahong Wu, Xiangxiang Chu, Sen Su

原文摘要

Knowledge editing aims to update outdated or incorrect knowledge in large language models without full retraining. Existing parameter-editing methods remain limited for long-sequence knowledge: single-point editors provide insufficient coverage for extended targets, while fixed chunk-based editors distribute editing effort uniformly across rigid predefined windows, failing to prioritize difficult, information-dense tokens. In this work, we show that token entropy is closely associated with both editing difficulty and semantic importance: high-entropy regions are harder to edit and often contain key factual content. Motivated by this observation, we propose DynEdit, a dynamic framework for long-sequence knowledge editing. DynEdit uses token-level entropy to guide non-uniform sliding, increasing the coverage of difficult high-entropy regions, and further applies entropy-guided token-level optimization to strengthen updates on information-dense tokens within each editing window. Experiments on three benchmarks and two backbone models show that DynEdit consistently outperforms competitive baselines across diverse long-form editing settings, achieving gains of up to +14.0 BLEU points on challenging paraphrase queries and near-saturated semantic scores on several diverse knowledge domains.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34759

From Belief Inference to Task Action: A Multi-Level Evaluation of Theory of Mind on Verifiable Tasks

Zhenshuai Yin, Yongxin Guo, Mingda Zhang, Zhongze Luo, Haoyue Liu, Lianmin Zhang, Xiaoying Tang

原文摘要

Effective collaboration requires language models to understand what others know and believe, and to act accordingly. Models that assume others share their information may omit necessary communication and undermine coordination. Many theory-of-mind (ToM) benchmarks assess mental-state inference through story questions, but correct answers do not establish that models can use these judgments in subsequent tasks. We introduce a multi-level evaluation built on existing, verifiable tasks. We extend household collaboration and calendar scheduling with asymmetric information while retaining their original constraints and validators. Independent interfaces assess world understanding, information access, beliefs, decisions, and execution. Controls supply personal evidence or correct beliefs to distinguish state recovery from state use. Across six model configurations, these abilities diverge. On matched scheduling cases, Claude recovers all agent beliefs correctly but produces subjectively optimal plans in only 67.9% of cases. In household collaboration, supplying beliefs raises Qwen-27B success from 55.0% to 75.0%, yet Qwen-4B answers every fixed supplied-belief source question correctly while completing only 20.0% of the corresponding continuation condition. These results show why ToM evaluation must connect perspective inference to independently verified decisions and task consequences.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.17558

Firefly: Illuminating Large-Scale Verified Tool-Call Data Generation from Real APIs

Yuxuan Lu, Ziyi Wang, Yingzhou Lu, Yisi Sang, Jiri Gesi, Xianfeng Tang, Yimeng Zhang, Zhenwei Dai 等 15 位作者

原文摘要

Training tool-calling agents requires large-scale trajectory data with verifiable labels, yet existing approaches either synthesize environments that diverge from real API behavior or generate tasks without ground-truth outcomes for verification. We present FireFly, a pipeline for generating verified tool-call data from real-world MCP servers. Our key insight is to invert the standard synthesis pipeline: rather than generating tasks and hoping they are solvable, we first let a strong LLM explore real APIs along graph-guided DAG structures, then synthesize tasks backward from observed outcomes, guaranteeing label correctness by construction. To handle the scale of real-world tool spaces (${\sim}$1,000 tools), we build a pairwise tool graph and sample sub-DAGs to focus exploration on semantically coherent workflows. To address environment drift in live APIs, we construct a retrieval-augmented simulator that caches all exploration results and replays them during training and evaluation, enabling fully offline and reproducible RL. Applying this pipeline yields 5,144 verified tasks spanning 240 servers and 993 tools. A 4B-parameter model trained with GRPO on FireFly matches Claude Sonnet 4.6 on our held-out test set and shows improvements on multiple tool-calling benchmarks including Tau2-Bench, MCPMark, and MCP-Atlas.

cs.SE · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34756

Mechanistic Insights into LoRA: Layer Sparsity for Adaptive Fine-Tuning via Path Patching

Fei Zuo, Yizhou Huang, Kezhi Wang

原文摘要

Low-Rank Adaptation (LoRA) has emerged as the dominant paradigm for parameter-efficient fine-tuning of large language models. Standard all-layer LoRA applies adaptation throughout the transformer, despite heterogeneous contributions from its components. In this work, we conduct a systematic mechanistic interpretability analysis to understand how LoRA affects model internals. We observe that LoRA's adaptation effect is highly non-uniform, with significant component-wise and layer-wise sparsity. Most critically, we find that important layers are highly task-dependent, with limited overlap across tasks. Motivated by these insights, we propose Adaptive-LoRA, a calibration-driven method that automatically identifies task-specific layers via frozen-base zero-ablation, enabling efficient layer selection without exhaustive search. Extensive experiments on Qwen2.5-1.5B across 11 diverse reasoning tasks demonstrate that Adaptive-LoRA outperforms full-layer LoRA while using only 30% of layers, achieving a $3.5\times$ reduction in adapter parameters. This work bridges mechanistic interpretability and practical efficiency, providing both fundamental insights and a principled approach for adaptive fine-tuning.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34746

DecisionTraceBench: Tracking Decision-Critical Information in Multi-Party Spoken Conversations

Wenchao Du, Fangxin Shang, Long Wu, Runze Cui, Qing Yang, Yehui Yang

原文摘要

In high-stakes domains such as finance, critical information evolves through multi-party interaction. Recovering these changes and their sources is fundamental to reliable speech understanding. Existing speech benchmarks focus on transcription, local understanding, and turn-level state prediction, leaving joint decision evolution across speakers insufficiently evaluated. We introduce DecisionTraceBench, a speech state-tracking benchmark for Decision-Critical Information State Tracking (DCIST). It jointly evaluates decision event sequences, speaker attribution, scope, and temporal evidence, using trajectory-level F1 to measure complete event-chain recovery. Our dataset comprises Chinese financial dialogues involving two to four speakers, with 100.39 hours of synthetic audio and 8.61 hours of human-performed recordings. On a fixed cohort of human-performed recordings, five cascaded systems and two direct-audio models achieve DCIST scores of 5.73% to 21.35%. Ablations with speaker-labeled reference scripts and no temporal localization requirements yield a maximum score of only 38.72%. Audio perturbation experiments further show that overall transcription error rates do not adequately reflect complete trajectory recovery. These findings reveal limitations in jointly understanding information provenance and state evolution, providing a benchmark for evaluating decision-process understanding and decision traceability in high-stakes speech applications.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2512.21711

Do Latent Tokens Think? A Causal and Adversarial Analysis of Chain-of-Continuous-Thought

Yuyi Zhang, Boyu Tang, Tianjie Ju, Sufeng Duan, Gongshen Liu

原文摘要

Latent tokens are gaining attention for enhancing reasoning in large language models (LLMs), yet their internal mechanisms remain unclear. This paper examines the problem from a reliability perspective, uncovering fundamental weaknesses: latent tokens function as uninterpretable placeholders rather than encoding faithful reasoning. While resistant to perturbation, they promote shortcut usage over genuine reasoning. We focus on Chain-of-Continuous-Thought (COCONUT), which claims better efficiency and stability than explicit Chain-of-Thought (CoT) while maintaining performance. We investigate this through two complementary approaches. First, steering experiments perturb specific token subsets, namely COCONUT and explicit CoT. Unlike CoT tokens, COCONUT tokens show minimal sensitivity to steering and lack reasoning-critical information. Second, shortcut experiments evaluate models under biased and out-of-distribution settings. Results on MMLU and HotpotQA demonstrate that COCONUT consistently exploits dataset artifacts, inflating benchmark performance without true reasoning. These findings reposition COCONUT as a pseudo-reasoning mechanism: it generates plausible traces that conceal shortcut dependence rather than faithfully representing reasoning processes.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34737

LADDER: Diagnosing Domain-Level Self-Assessment Under Noisy Feedback

Zhenshuai Yin, Jianshu Liang, Xuyue Xiong, Yongxin Guo, Mingda Zhang, Zhongze Luo, Haoyue Liu, Lianmin Zhang 等 9 位作者

原文摘要

Knowing when they are reliable is essential for language models to support human-AI collaboration. Yet interaction feedback is not always complete or correct, and may mislead models' understanding of their own capabilities. We introduce LADDER to study how language models assess themselves under unreliable feedback. Our findings show that models can accurately report supplied performance information while still struggling to form trustworthy self-assessments from interaction experience. Apparently accurate assessments can obscure capability differences and do not necessarily predict future task performance. These findings suggest that trustworthy self-assessment should be evaluated against observed task performance and its predictive value. Our study provides empirical evidence of the limits of model self-knowledge and a basis for evaluating capability awareness in reliable human-AI collaboration.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34735

iPrefBench: A User-Centric Benchmark for Personalized Mobile Use Agents

Yuting Ning, Ruijia Cheng, Sushobhan Ghosh, Peitong Duan, HungShi Lin, Abhishek Sundararajan, Shreyash Pandey, Huan Sun 等 10 位作者

原文摘要

An agent that truly knows and acts on behalf of its user has been a long-standing promise of AI assistants. Computer-use agents bring this promise within reach by capably executing fully specified tasks across apps. However, real-world user requests are often underspecified. To act correctly for the particular user, an agent must infer some missing information from the user's activity history and ask for other information that only the user can provide, without making unsupported assumptions or burdening the user with unnecessary questions. We present iPrefBench, a benchmark for evaluating the infer and ask capabilities of personal mobile-use agents in an executable iOS environment with an interactive simulated user. Grounded in a coherent persona with 3,051 past activity events, iPrefBench comprises 130 underspecified user tasks across 26 apps. Each omitted user-dependent requirement is, by construction, either history-inferable or ask-only. iPrefBench evaluates personalized mobile agents on both whether agents complete tasks correctly for the user (i.e., personalized task completion) and whether they seek user input when necessary (i.e., user-agent interaction). Evaluation of 7 frontier agents reveals a substantial gap between execution and personalization. Agents achieve high action completion scores (up to 91.6%) but low full task success (up to 47.7%) because personalization lags behind. They especially struggle with ask-only information, failing to request enough necessary information. Our findings highlight a challenge for personalized agents in knowing what to infer and when to ask. We hope that iPrefBench can serve as a valuable testbed for building agents that know their user and work with them.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2608.00583

A False Average: Pooled CoT-Monitor Accuracy Conceals a Reasoning-Dependent Fragility

Shikhar Shiromani, Leo Richter, Noah Siegel

原文摘要

Chain-of-thought (CoT) monitors are reported by their aggregate accuracy on a pool of reward hacks. We show that this number is a false average. On Terminal Wrench, about 77% of hacks are given away by the actions alone, and the monitor's pooled accuracy is dominated by them; on the remaining 23%, which an action-only monitor misses, the same monitor is fragile. We expose the fragility with a controlled attack: we rewrite only the agent's reasoning to read as good-faith engineering, leaving every command and output byte-identical. One gradient-free rewrite drops a held-out monitor's catch rate on that subset from about 95% to between 4 and 11%, while the pooled rate falls only about 25 points, near what the subset's share of the pool times its collapse predicts. The rewrite stays truthful about what happened and lies only about intent, so the trace-only monitors we test recover little, even when warned about the attack; of our defenses, only information from outside the trace helps much. The attack transfers across monitor and agent families, and the dataset's own hack labels show the same gradient with no monitor in the definition. Live agents told to hack on 27 tasks, with no monitor feedback, produce hollow hacks, but a monitor calibrated on each model's own honest solves catches nearly all of them. Applying the same rewrite to these live hacks keeps all 111 action-visible ones caught and lowers catch on the 16 whose actions score below 0.5 only from 16 to 13: the effect lands where the false average predicts, but the collapse stays a controlled result. A linear probe on a surrogate monitor's activations out-reads its verdict on the obfuscated hacks (AUC up to 0.98), but a behavioral control shows this edge does not, by itself, mean the monitor knows more than it says. Monitor accuracy should be reported on the reasoning-dependent subset, where the defense is load-bearing.

cs.CR · cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.03825

Dynamic Short Convolutions Improve Transformers

Oliver Sieberling, Bharat Runwal, Rameswar Panda, Yoon Kim

原文摘要

Transformers have become the dominant architecture for large language models, largely due to the scalability and flexibility of attention, feed-forward layers, residual connections, and normalization. This paper introduces dynamic short convolutions as an additional neural network primitive for improving Transformers. Unlike static short convolutions, dynamic convolutions use input-dependent filters, which preserves the locality bias of convolution while increasing expressivity. Motivating experiments show that applying dynamic short convolutions to key, query, and value representations improves performance on challenging associative recall tasks compared with static convolutional variants. Across language-modeling experiments ranging from 150M to 2B parameters, dynamic convolutions consistently outperform standard Transformers and Transformers augmented with static short convolutions. Fitting scaling laws indicates a 1.33$\times$ compute advantage over compute-matched Transformers when dynamic convolutions are applied to the key, query, and value vectors, and a 1.60$\times$ advantage when adding dynamic convolutions after every linear layer. Dynamic convolutions also offer improvements on linear RNNs (Mamba-2/Gated DeltaNet) and mixture-of-experts architectures. We make these gains practical with custom Triton kernels that enable efficient training with a manageable end-to-end slowdown. These results suggest that dynamic short convolutions are a scalable, hardware-efficient, and expressive primitive for advancing Transformer-based language models.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.02423

Investigating and Alleviating Harm Amplification in LLM Interactions

Ruohao Guo, Wei Xu, Alan Ritter

原文摘要

Large language models (LLMs) can serve as helpful assistants, yet they can equally function as harm amplifiers that enable malicious users to achieve harmful outcomes beyond their capabilities through extended interactions. This risk manifests along two axes, i.e., democratizing domain expertise that allows novices to produce specialized harmful content, and scaling harmful operations at volumes that manual effort cannot match. Existing works, however, often overlook how LLMs compound harm across multi-turn conversations. We introduce HarmAmp, a new benchmark for multi-turn harm amplification scenarios spanning twelve risk categories. Each scenario is grounded in real-world threats and satisfies rigorous criteria, i.e., substantive amplification, operational specificity, and multi-turn necessity. We further propose TrajSafe, a proactive monitor that anticipates harmful trajectories and intervenes through actions such as probing users' genuine intents and steering the models towards safer completion. Our extensive experiments demonstrate that TrajSafe significantly reduces the harmfulness incurred in multi-turn interactions while preserving a low over-refusal rate and the target model's general capabilities. Our work offers a promising paradigm to alleviate the nuanced safety risks in LLM interactions.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34684

Knowing When the Evidence Is Enough: Requirement-Directed Search with Horn-Closure Certificates

Xinyu Zhang, Weiwei Sun

原文摘要

After each Web search, a search-augmented model must decide whether its evidence suffices: answer, search again, or abstain. Prevailing systems leave this to holistic model judgement, which can accept a context naming every entity yet lacking one required fact. We instead make sufficiency structural: a per-query requirement hypergraph states which facts jointly yield another fact. This raises two challenges: deriving requirements from the question alone, and linking each requirement to passages about the right entity. Certify therefore decomposes the question, searches per requirement, and accepts links only after entity checks. Horn closure over the links yields a certificate, the minimal sufficient passages and their derivation, or an abstention naming open requirements. Given correct retrieval, requirements and links, the decision is sound, and complete up to the entity check. Pooled over three corpora with unanswerable twins, Certify decides significantly more accurately than search agents, graph retrieval and sufficiency judges in default configurations. On the same evidence, it matches a proprietary model's sufficiency label without an additional model call. Over 21M Wikipedia passages, entity checks add 4 to 13 points of decision accuracy where gold evidence exists. Finally, answers from its certificate are 80% correct, against 57% from all retrieved passages.

cs.IR · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34673

Agent Memory as Recommendation: Recommender-System Design for Long-Term Memory of LLM Agents

Xinyu Zhang, Weiwei Sun

原文摘要

Large language model (LLM) agents that serve users across many sessions must choose, for every request, the few stored memories that enter a limited context window. This choice largely decides whether the agent answers correctly, and it runs on every request at Web scale. This selection faces three challenges: a large and growing store, a per-request cost budget, and multi-session questions that need all of their evidence. Many memory systems build notes or graphs with hundreds to thousands of LLM calls per history, and each is designed on its own, so it is unclear which component yields the accuracy. These challenges mirror the properties of recommendation, so we formalize agent memory as budgeted slate recommendation, with the request as the user, stored memories as items and the context window as a slate. The formalization places memory systems in one design space of recommender cascades and factorizes accuracy into the completeness that each stage controls. Guided by it, we propose AgentMemRec, which adapts candidate generation, learning to rank and slate construction from recommender systems and selects memories without any LLM call. On multi-session questions of LongMemEval, AgentMemRec outperforms memory and context-selection methods, including A-Mem, HippoRAG 2, DynamicRAG and T-Mem, by 10 to 31 points of accuracy. It matches an LLM reranker that makes 64 calls per query at 3.2 times lower serving cost, and on LoCoMo it outperforms Mem0 and Zep without any LLM call at write time. The factorization attributes the gains to slates that more often hold all of the evidence, so recommender-system design offers an effective and efficient foundation for the long-term memory of LLM agents.

cs.IR · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34659

AdaTutor: Adaptive Tutoring through Pedagogical Strategy Selection

Rajeev Bhatt Ambati, Shashank Srivastava, Snigdha Chaturvedi

原文摘要

Effective tutoring requires determining not only what information to provide, but also which form of assistance to deliver. A tutor may address the same student error using various pedagogical strategies, adapting its approach as the interaction progresses. We introduce Comparative Pedagogical Supervision (CPS), a training framework to learn to select the optimal strategy and feedback for student learning. We evaluate CPS on math and science reasoning datasets. After five tutoring turns, CPS outperforms the strongest fixed-strategy baselines. The learned tutor transfers across datasets, and its strategy selection can also inform a separate API-based tutor.

cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34653

Heard Enough to Act: Early Tool Calling in Spoken Dialogue Systems

Yizhou Peng, Chen Chen, Donghang Wu, Hexin Liu, Eng Siong Chng

原文摘要

Latency-sensitive spoken agents must know not only which tool to call, but also when they have heard enough to act. Waiting for the turn to complete delays execution even when the request is already actionable. However, neither existing spoken agents nor spoken dialogue corpora define the earliest point at which a request becomes actionable. We introduce a synthetic spoken-dialogue pipeline that provides earliest-actionable-point supervision, marking when the observed speech grounds a tool and its required arguments. Through controlled task generation, speech synthesis and alignment, and gated annotation, we construct 26,694 dialogues totaling 786 hours across 48 domains, with disjoint training and test argument values. We additionally create an audio rendering of $τ^2$-bench for external evaluation. We train an omni-modal language model on this corpus, enabling causal, chunk-level decisions and partial-transcript generation for downstream call completion. The model achieves tool-state macro-F1 scores of 0.48 and 0.47 on held-out dialogues and the external benchmark, respectively. For correct calls, the estimated mean lead over a turn-end cascade, using the same formatter, is 1.8-2.0 seconds in-domain and 1.1 seconds on $τ^2$-bench. These results demonstrate a data-driven path toward spoken agents that act while listening. We will open-source the pipeline and the training scripts.

cs.CL · eess.AS

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34639

Cross-Loop Query Handoff: How Later Loops Redirect Two-Hop Retrieval in the Ouro Looped Language Models

Ziyu Mu, Zehua Wang

原文摘要

How do later loops in a pretrained looped language model repair two-hop retrieval? We study Ouro-1.4B and Ouro-2.6B by transplanting attention queries between loops at the same layer, head and final prompt position. On synthetic in-context questions, injecting the layer-indexed collection of loop-3 queries throughout loop 1 rescues 90% and 97% of questions answered incorrectly at loop 1 and correctly at loop 3, versus 24% and 21% for shuffled donors. The rescue also holds under greedy generation in Ouro-2.6B; in Ouro-1.4B it is limited to four-candidate ranking. Crossed donor-recipient swaps favor retrieval through the donor's bridge entity over copying its answer, although a substantial positional component remains in Ouro-2.6B. Late keys and values rescue far fewer targets. Freezing queries at their loop-1 values also damages two-hop accuracy more than a matched question that supplies the bridge. On natural questions, rescue is smaller and replicates only in Ouro-2.6B, and crossed swaps do not establish bridge redirection. These results identify a transferable collection of late-loop queries that supports bridge-conditioned retrieval on synthetic scenes; they do not locate where the bridge is computed or establish a single-vector handoff.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2608.12571

Is this Citation on Point?

Apurv Verma

原文摘要

In 2023, a New York judge sanctioned two attorneys in \emph{Mata v. Avianca} for filing a brief containing hallucinated citations generated by ChatGPT. These failures can be largely caught by database lookups. The harder problem is detecting citations that point to real cases but do not support the propositions for which they are offered, a failure mode that existing evaluations of LLMs for legal use cases largely overlook. In this paper, we study proposition-level citation support verification through controlled perturbations of real legal citations obtained from two legal corpora. We either replace the cited case or change only the pinpoint page within the same case. Our evaluation of fourteen model configurations on the resulting examples shows that models catch $93$--$100\%$ of wrong-case corruptions, but they only catch $37$--$61\%$ of wrong-pinpoint corruptions on court opinions and $52$--$83\%$ on legal briefs. When models fail to catch wrong-pinpoint corruptions, they accept the citation based on topical overlap rather than page-level support. Larger-scale models and extended reasoning narrow the gap but do not close it: GPT-5.4 with high reasoning effort still misses $40\%$ of pinpoint mismatches on court opinions and $18\%$ on briefs. Prompting the model to verify support at the cited page improves recall, but it also raises the false positive rate. These results show that citation misrepresentations remain difficult to detect automatically. Recognizing the right legal topic and verifying support for the cited proposition are distinct capabilities, and current models conflate them.

cs.DL · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.34626

Entity Matching using Decision Models

Aaron Steiner, Ralph Peeters, Christian Bizer

原文摘要

Entity matching is the task of deciding whether two records describe the same real-world entity. Current entity matching methods rely on pre-trained language models (PLMs) which require task specific fine-tuning or generative large language models (LLMs) which do not require task-specific fine-tuning but are slow and expensive at matching large record sets. Decision models are a new class of LLM-based models that return a probability for each predefined answer option rather than generate text. This paper investigates decision models as an alternative to generative LLMs and fine-tuned PLMs for entity matching. We evaluate six decision models in a zero-shot setting across eight benchmarks and compare their performance with that of the generative LLM GPT-6 Luna and the PLM-based matcher Ditto. Our experiments show that, without task-specific training data, the four strongest decision models achieve mean F1 scores close to that of GPT-6 Luna, with the best model slightly surpassing it. This model also outperforms Ditto on six of the eight benchmarks. All decision models are faster than GPT-6 Luna, and most are less expensive. For three of the four strongest decision models, the predicted match probabilities are well calibrated without requiring any task-specific recalibration. A model cascade that routes the least confident record pairs to an ensemble of generative LLMs improves the mean F1 scores of the four strongest decision models by 1.9-4.0 percentage points. Using the best decision model and its best-performing prompt for each benchmark, the cascade surpasses the highest published F1 scores from literature on four of the eight benchmarks.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗