QA
-
How Far Can Coding Agents Go in Visual Generation?
2026-10-13 18:18:41Hengguang Zhou, Yuanhao Ban, Jiaqi Feng, Xiaohuan Pei, Justin Cui, Cho-Jui HsiehAbstract
Modern image generation largely relies on specialized generative models that leave the process of image construction implicit in learned parameters. We explore an alternative: general-purpose coding agents that construct images through executable programs. Our study of zero-shot drawing on AgentDrawBench, a 200-prompt benchmark, reveals a distinctive capability profile. Agents excel at preserving structure and satisfying precise visual constraints, achieving competitive results in domains where compositional accuracy matters more than photorealistic appearance, while remaining limited in perceptual richness. Their visual expressiveness varies with the drawing interface, and a qualitative study on reconstruction shows how procedural skills extracted from human demonstrations support different artistic media. We then examine code as an external visual language of thought: a structured, executable representation that agents can construct and inspect. We test it in visual question answering task and find reconstructed visual code retains substantial information for question answering and, in some domains, supports higher accuracy than the original image. These abilities raise a further question: do agents draw on a shared internal representation of scene content beyond a particular drawing program? We explore an internal visual language-of-thought hypothesis, in which a shared, compositionally structured representation supports visual operations across drawing languages. Cross-harness drawings and images assembled from independent single-pixel queries exhibit related layouts and tones. Agents also maintain self-generated room arrangements across viewpoints after visible drawings are removed. These results demonstrate that the coding agent would maintain a consistent and coherent world inside its imagine even before the harness gets involved and the generation starts.
Abstract (translated)
URL
https://arxiv.org/abs/2610.19025
PDF
-
GeoReader: Learning Latent Geometry for Spatial Reasoning in Vision-Language Models
2026-10-13 16:45:37Rong Li, Shijie Li, Jiangyong Huang, Lingdong Kong, Jiaming Zhou, Dongyue Lu, Zeying Gong, Baoxiong Jia, Junwei Liang, Siyuan HuangAbstract
Vision-language models (VLMs) struggle with metric and multi-view relationships that remain implicit in RGB observations. Geometry foundation models recover this structure from RGB, but existing methods either fuse the estimated features with visual tokens and learn to use them from question answering alone, or align VLM latents with geometry features. Capturing geometry, however, does not ensure its use: zeroing the aligned latents of a released feature-alignment model leaves its accuracy unchanged. We introduce GeoReader, a framework that learns geometric representations for the VLM that reads them. First, we train lightweight Geometric Readers to produce continuous input tokens, supervising them with geometry descriptions decoded by a frozen VLM, so that the Readers express geometry in a form the target model can already read. We then freeze the Readers to preserve this interface while fine-tuning the VLM on spatial QA, with geometry injection conditioned on the question. Geometry descriptions serve only as training targets: at inference, the VLM answers directly from the continuous geometric tokens without generating a geometry chain of thought. With geometry predicted from RGB images, our approach reaches 94.3 on MindCube-Tiny with Qwen3-VL-4B and, with a multi-source QA mixture, an average of 72.5 across six additional spatial benchmarks with Qwen3.5-9B. These results suggest that a useful geometric representation is one the receiving model can use, not merely one from which geometry can be reconstructed.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18694
PDF
-
EgoOmniWorld: Unifying Egocentric Understanding via Ego-Action-World Interactions
2026-10-13 14:53:16Songruo Chen, Yuqian Fu, Yu Li, Mingkang Dong, Sifan Zhou, Bin Ren, Xu Zheng, Tianwen Qian, Mohamed Elhoseiny, Yanwei FuAbstract
Egocentric video understanding spans diverse capabilities, modalities, and evaluation formats, yet these are often studied in isolation despite their tight coupling in real-world interaction. We propose the Ego-Action-World interaction loop as a unifying principle: the wearer perceives the world, forms goals, acts, changes the world, and uses feedback to guide subsequent behavior. Based on this view, we introduce EgoOmniWorld, a benchmark organized around four complementary interaction structures: Node captures ego, action, and world states; Edge captures directed relations among them; Chain captures linked dependencies; and Loop captures feedback-closed interactions. We develop an interaction-first, evidence-grounded data pipeline that extracts structured interactions from egocentric videos, grounds them in visual and auditory evidence, and generates questions according to their interaction structure. Required modalities and single- or multi-turn formats follow naturally from the supporting evidence and dependencies. EgoOmniWorld contains 24,166 QA groups, including 20,284 training and 3,882 test samples. Experiments on representative open- and closed-source multimodal large language models reveal substantial and uneven limitations across interaction structures. Models trained on EgoOmniWorld improve in-domain interaction understanding and generalize to existing egocentric benchmarks. EgoOmniWorld provides a unified foundation for evaluating and advancing egocentric video understanding. Data, code, and models will be released.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18499
PDF
-
Beyond Answer Sharing: TrueMAD for Targeted Peer Engagement in Multi-Agent Debate
2026-10-13 12:08:38Dezheng Bao, Sihang Du, Renhong Huang, Zhizhang Yuan, Yifei Sun, Yang YangAbstract
Multi-agent debate (MAD) aims to improve large language model reasoning by exposing agents to alternative solutions and enabling iterative revision. However, standard broadcast-style MAD makes peer answers visible without requiring agents to engage with a particular peer claim. As a result, reliable reasoning may be suppressed by conflicting or majority responses, limiting achievable performance, while erroneous reasoning may persist and degrade final predictions without actionable feedback. We propose TrueMAD, a backbone-agnostic test-time plug-in that introduces one-to-one, free-form peer advice between answer broadcasting and self-revision. Each source agent responds to a target's current reasoning through support, rebuttal, refinement, or further questions, while the target retains control over how to revise. The same mechanism can augment standard MAD and its improved variants without parameter updates. Across AIME, GPQA-Diamond, and TruthfulQA with Qwen3-30B-A3B, TrueMAD improves agent-level and strict-majority accuracy over MAD by 3.62 and 3.52 percentage points on average, while TrueDMAD improves DMAD by 4.72 and 4.06 points, respectively. Analyses at the utterance, agent-transition, and group levels further show more peer-specific deliberation, more correction-oriented revisions, and more favorable consensus dynamics.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18240
PDF
-
WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution
2026-10-13 09:52:02Hang He, Linting Wang, Guanming Liu, Bo Mao, Jinxiu Liu, Yiran Suo, Maosen Zhou, Zhiheng Xi, Guanting Dong, Zhou Shao, Rui Zheng, Jie Zhou, Chengcheng Wan, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing HuangAbstract
Scaling interactive environments is important for training long-horizon agents through reinforcement learning. These environments often require further testing and refinement after initial generation, and correcting defects in a task may also require updating its work context and verifier. However, existing construction pipelines offer limited support for such iterative refinement while keeping these components aligned. We introduce WorkRover, a framework for training long-horizon cowork agents through recursive environment evolution. Its harness-driven synthesis enables construction agents to jointly build and validate semantically aligned work contexts, executable tasks, and evidence-grounded verifiers, adapting their actions to workspace state and execution feedback rather than following a fixed construction workflow. To support environment evolution, we introduce Factorial Compatibility Replay (FCR), which uses rollout feedback to evaluate combinations of component revisions through dependency-closed replay and selects the smallest validated repair among tested candidates while preserving task requirements and evaluation constraints. To provide finer-grained training signals, we propose Verifier-to-Turn Credit Assignment (VTCA), which augments trajectory-level GRPO advantages with signals from deterministic workspace-progress checks and assigns a bounded share of positive-progress credit to prerequisite turns using recorded tool-output and artifact dependencies. Using WorkRover, we post-train Qwen3.5-35B-A3B to obtain WorkRover-35B-A3B and evaluate it on several cowork-agent benchmarks. Its reported scores exceed those of Qwen3.5-35B-A3B by 26.27 percentage points on JobBench, 29.32 percentage points on OfficeQA Pro, and 413.25 Elo points on GDPval v2. These results suggest that WorkRover offers a promising foundation for scalable training of long-horizon cowork agents.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18067
PDF
-
Seeing Through the Forgery Is Not Enough: Regressive Sycophancy under Fabricated Evidence in Medical Vision-Language Models
2026-10-13 08:55:06Sadif Ahmed, Soumya Swagata Biswas, Rifat Shahriyar, Chowdhury Mohammad Rakin HaiderAbstract
A medical vision-language model may analyze a given CT scan and correctly detect both kidneys in view. Continuing this conversation, if a user provides a prescription with an opposite direction, medical VLMs often sycophantically choose to realign their diagnosis. In this work, we study regressive sycophancy under fabricated evidence, where a model abandons a correct answer when pushback comes with a fake document. We test four open VLMs on 400 VQA-RAD yes/no questions in 19,200 multi-turn conversations, with and without a fabricated prescription. Without a disclosure that documents are synthetic, the models abandon 80.1% of 6,488 correct answers. Notably, under one script each, the same prescription raises the flip rate of Qwen2.5-VL-7B from 45.9% to 89.6% and lowers that of Qwen3-VL-8B from 77.7% to 50.9%. A model rejected the prescription while still preserving the correct answer in 915 of 3,232 conversations with the prescription (28.3%), and 40% of these flipped in later turns. System-prompt sentences that disclose the documents as synthetic lower one model's flip rate from 77.8% to 47.3%. Thus, we conclude that awareness about a forgery does not prevent a model from adopting its answer.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17989
PDF
-
Virtual Head Attention
2026-10-13 08:39:20Xibo Ding, Guoxia Wang, Jinle Zeng, Jiabin Yang, Sirui Tao, Chao Yang, Dianhai YuAbstract
The design of attention mechanisms in long-context large language models faces an ``impossible triangle'' among KV cache size, training and prefill FLOPs, and model quality. GQA achieves strong quality but incurs high KV cache overhead; alternative approaches such as MLA and MFA can compress the cache, but MLA suffers from quality loss while MFA substantially increases training and prefill compute. We propose Virtual Head Attention (VHA), which simultaneously improves all three vertices through two lightweight linear operations: Q Premix applies a near-identity transformation to queries within each KV group to recover virtual head diversity from halved physical query heads, and Linear Postmix fuses inter-head features via a low-rank $I + AB^\top$ residual structure. Since VHA introduces no nonlinear operations, Premix can be folded into the query projection and Postmix into the output projection at inference time, reducing the model to a standard GQA-2 architecture that directly reuses all existing GQA inference optimizations. With only two KV heads, the group size in large models readily reaches 64, fully exploiting Tensor Core parallelism during decoding. Across four scales with 4K-context 50B-token pretraining, VHA consistently surpasses the standard GQA baseline at every scale ($+$0.05\% to $+$0.81\%), while reducing KV cache by $4\times$ vs.\ GQA-8 at dense scales and $2\times$ vs.\ GQA-4 at the MoE scale, and lowering training and prefill FLOPs by 7.9\% (1.7B, $S{=}4096$). On an industrial 30B-A3B MoE model trained on 500B tokens at 8K context, VHA outperforms both MLA and GQA-4 by $+$0.70\,/\,$+$2.60\,\% on a 12-benchmark base-model evaluation suite; in a separate matched 10B-A1B MoE comparison after long-context mid-training, VHA attains an average RULER score of 84.4 over 8K--128K versus 73.0 for GQA, while using two rather than four KV heads in the full-attention layers.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17959
PDF
-
Visual Dependence Does Not Imply Answerability: A Systematic Analysis of Answerability Signals in Frozen InstructBLIP
2026-10-13 07:56:00Minjun Kim, Hyunmin Song, Changwon LimAbstract
Visual question answering models often produce plausible responses even when the available visual evidence is insufficient to support a valid answer. Although prior work has introduced benchmarks and abstention mechanisms for unanswerable visual questions, it remains unclear which signals derived from a frozen vision-language model reliably indicate answerability on unseen image-question pairs. We systematically evaluate answerability signals on a stratified 10K subset of UNK-VQA using frozen InstructBLIP, covering predictive uncertainty, cross-modal alignment, visual grounding, visual dependence, internal representations, and learned probes. Answerability discrimination remains weak, with test AUROC largely confined to 0.50-0.54. CLIP alignment yields the highest point estimate but does not significantly outperform generation confidence under paired bootstrap analysis. Targeted removal of question-relevant visual evidence causes substantially larger changes in answer confidence than same-size random masking, yet intervention sensitivity provides essentially no answerability discrimination. Increasing probe capacity, combining multiple signals, and adding reason-aware supervision do not provide robust improvements, while post-freeze cross-validation yields similarly weak discrimination. Selective-answering analysis shows limited usable coverage under low-risk operating criteria. Failure-mode analysis shows that the relative utility of individual signals varies across perturbation mechanisms. Taken together, these results show that, on our UNK-VQA subset with frozen InstructBLIP, visual dependence, uncertainty, semantic alignment, and static representations are not robust standalone proxies for answerability. Our findings emphasize the need to distinguish a model's reliance on visual evidence from whether that evidence is sufficient to support an answer.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17902
PDF
-
GraphOPD: Graph-Augmented On-Policy Distillation for LLM Agents
2026-10-13 07:31:01Bohan Lin, Liyi Chen, Zhuoning Guo, Muyang Li, Qimeng Wang, Yan Gao, Yao Hu, Yudong ZhangAbstract
On-policy distillation post-trains large language model agents by supplying dense, step-level guidance from a teacher policy when the reinforcement-learning reward is sparse and arrives only once per trajectory. Existing instantiations allocate this guidance by the size of the teacher-student divergence at each step, on the single-turn intuition that a large disagreement marks a mistake worth correcting. Once decisions chain over many turns, that rule misfires, since an early drift enters every later context both policies condition on, leaving the teacher consistent with the drifted trajectory instead of flagging its cause, while interchangeable steps register large but outcome-irrelevant divergences. We demonstrate this on an agentic benchmark, where distilling the highest-divergence steps brings no consistent benefit over random selection. To this end, we introduce GraphOPD, the first method to bring graph-based structural augmentation into on-policy distillation for agent capabilities. It reads which steps enabled which later ones from the environment's own record of state changes, immune to the drift that corrupts the teacher-student gap, organizes them into a dependency graph, scores each step by a random-walk stationary distribution over it, and fuses that structural credit with the divergence signal into a trajectory-relative mask concentrating supervision on each rollout's highest-aptitude steps. Across three model scales and eleven baselines on ALFWorld, WebShop, and SearchQA, GraphOPD shows competitive performance throughout, improving over the strongest baseline by up to +5.8 pp. An executed-replay audit further shows that this structural credit score tracks true causal impact far above chance, that both fused signals are independently necessary, and that the same signal transfers to out-of-domain tool-integrated reasoning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.08959
PDF
-
ECMR: Evidence-Aware Deliberation Control for Multi-Agent LLM Reasoning
2026-10-13 07:30:30Yuan Liu, Haoyue Peng, Fanliang Hu, Lei Huang, Wenli CaiAbstract
Multi-agent Large Language Model (LLM) systems typically commit to a deliberation protocol before inference. A fixed protocol cannot allocate computation differently to resolved and unresolved instances, and it may mistake repeated support for independent confirmation. We study this failure as semantic false consensus: a confident group answer supported by narrow, duplicated, or correlated rationales. We introduce ECMR, an evidence-aware controller over frozen role-conditioned LLMs. At each step, ECMR summarizes answer concentration, support diversity and independence, contradictions, role behavior, verification history, and remaining budget, then selects whether to propose, challenge, verify, arbitrate, diversify, correct a role, or stop. We formulate this decision as finite-horizon budgeted control and derive sufficient conditions under which state conditioning, risk-triggered verification, and role correction improve cost-adjusted utility. Across a declared held-out evaluation of 9,000 records from MedQA, CaseHOLD, and FinQA, ECMR-Full increases pooled accuracy from 62.1% to 84.9%, reduces false-consensus rate from 40.7% to 17.0%, and raises evidence-support rate from 58.0% to 88.7%. State shuffling loses 11.7 accuracy points with only 0.7 fewer rounds, while harder traces shift from early stopping toward arbitration. ECMR-Full uses the largest token budget, yielding the strongest absolute reliability rather than uniformly superior token efficiency.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17869
PDF
-
Navalia: Advancing Cowork Agents through Verifiable Task Synthesis and Long-Horizon Post-Training
2026-10-13 07:19:22Hengqi Liu, Ziyuan Liu, Yang Qin, Wanting Zhou, Zichuan Wang, Jiachen Liang, Xu Chu, Shaowei Chen, Zhaokai Luo, Mu ChuanAbstract
Large language model (LLM) agents are evolving from conversational assistants into Cowork Agents capable of autonomous planning, tool use, and complex task execution. Financial analysis requires these agents to integrate cross-document evidence and perform multistep numerical reasoning, but expert task authoring is costly, while direct LLM synthesis struggles to produce challenging tasks with reliable answers. We introduce Navalia, a framework for Cowork Agents that combines verifiable task synthesis with long-horizon post-training. Using corporate financial reports as its data source, Navalia semantically selects supporting evidence from complete fact sets and constructs executable computation chains to generate analytical tasks with traceable evidence and actual data dependencies between steps. We further use independent LLM review and programmatic checks to verify evidence consistency and answer correctness. The verified tasks are then used for trajectory collection, cold-start supervised fine-tuning (SFT), and reinforcement learning (RL) to improve long-horizon task execution. Navalia, initialized from Qwen3.6-35B-A3B, achieves consistent gains over the base model across all five benchmarks: OfficeQA Pro, FinSearchComp, Finance Agent Benchmark, APEX-Agents-AA, and Workspace-Bench-Lite. These results demonstrate that Navalia improves evidence integration, tool use, and long-horizon task execution capabilities and generalizes to broader Cowork settings, including general office tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26887
PDF
-
From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models
2026-10-13 04:35:53Hyeonggeun Han, Taehyun Cho, Sangwoo HongAbstract
Large vision-language models often exhibit object hallucination, generating claims that are not supported by the input image. To mitigate this issue, we propose Evidence-Guided Visual Claim Admission (EGVCA), a training-free framework that admits a semantic claim only when it receives sufficient image-dependent support. EGVCA measures this support by comparing the claim likelihood under the observed image and a visual-null input. For captioning, it controls semantic expansion without altering token generation. For binary visual question answering, it combines atomic visual support with the image-induced change in the model's native Yes--No margin. Experiments on CHAIR, POPE, and MME across three LVLM backbones show that EGVCA substantially reduces object hallucination while preserving caption quality. It achieves the highest POPE F1 in eight of nine backbone--sampling settings and improves the native MME total score on all three backbones. These results demonstrate the effectiveness of claim-level admission as a simple alternative to token-level distribution correction.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17706
PDF
-
MedEvo: Medical Reasoning Evolution through Knowledge Navigation
2026-10-13 04:11:26Xiaoliang Wang, Zhuo Chen, Chuanyang Zheng, Quanbin Wang, Jiangwei Lao, Yingao Liu, Jiajia Liu, Jinming Cao, Honglin Lin, Zhihong Lu, Hao Jin, Jian Wang, Lijun WuAbstract
Medical multimodal reasoning models are typically trained on teacher-generated rationales, discarding the valid reasoning in students' own failed trajectories, while general-domain correction methods offer little guidance for errors in image reading, anatomical context, or differential diagnosis. We propose MedEvo, which uses medical knowledge, namely a tag taxonomy and error-type diagnosis, to navigate both trajectory repair and data selection. For each failed rollout, a gold-blind auditor locates the first consequential error and provides a minimal, answer-free hint; the same student retains the valid prefix and regenerates the erroneous suffix. After each round, MedEvo identifies capability gaps across medical tags on a held-out feedback set and prioritizes related questions for new rollouts and repairs. With supervised fine-tuning (SFT) alone, MedEvo raises Qwen3.5-9B mean accuracy across seven benchmarks from 65.57% to 69.12% (47.40% to 54.50% on the multimodal track of MedXpertQA (MedXpertQA-MM)), surpassing teacher-distillation SFT (66.93%). It also raises Qwen3.5-35B and Qwen3.5-397B mean accuracy from 67.91% to 73.04% and from 71.83% to 74.95%, respectively. Our analyses show that failed trajectories contain reasoning worth preserving, that repair success does not imply learning value, and that independent feedback is key to sustained improvement.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17690
PDF
-
DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter
2026-10-13 03:49:32Xuyan Ye, Xiyan Fu, Shengda Fan, Jiawei Cui, Yankai Lin, Wei LuAbstract
Reinforcement learning with verifiable rewards (RLVR) for multi-turn agents relies on sparse terminal rewards that say little about individual steps. Within-group step comparisons depend on reliable state matching and sufficiently diverse sampled outcomes, while direct step scoring may require costly external evaluators, human annotations, or task-specific heuristics. We introduce DeltaTTG to learn a reusable progress signal through a shared progress meter, using low-cost supervision from diverse policies' recorded trajectories without task-specific progress rules. Its supervision target, Token-to-Go (TTG), the number of tokens remaining to be generated until task completion, serves as an observable proxy for remaining completion effort. During actor training, changes in the frozen meter's estimates redistribute outcome-derived advantages across steps without altering the reward, preserving each trajectory's advantage sign and token-average scale. On ALFWorld, PaperSearchQA, and BFCLv3, DeltaTTG improves the average score over GRPO by 5.7 and 7.2 points with Qwen3-4B and Llama-3.1-8B-Instruct, respectively, outperforms all compared RL baselines on average, and keeps actor-training GPU-hours close to GRPO. Further experiments show improved sample efficiency and out-of-domain performance, and support the meter's ability to distinguish step quality and guide test-time action selection across actor sizes. The code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17666
PDF
-
Dissociating Answerability from Extraction in LLM Question Answering over Radiology Reports
Abstract
Answering questions from radiology reports involves two decisions: whether the report contains an answer and which span gives it. A false answer presents an unsupported claim as fact; an over-abstention hides what the report says. Yet RadQA results are reported as a single F1 that merges both. We propose an evaluation method that separates them, reporting false-answer and over-abstention rates beside F1 and splitting F1 changes into extraction and answerability terms. We apply it to 13 LLMs on the RadQA test set, zero-shot and with three cumulative demonstration sets, the last adding an unanswerable example. Models with similar F1 differed more than twofold in false-answer rate. The unanswerable example changed when models answered, not how well they extracted: 12 of 13 models abstained more often, and three-quarters of the answers they withdrew had F1 $\geq$ 0.5. Within models, false answers showed no clear association with extraction quality, but acting on the decision itself helped: keeping an answer only when enough other models also answered lowered the zero-shot false-answer rate by 4 points without a detectable F1 change. Because the two decisions can change independently, we suggest radiology QA evaluations report them separately, not through F1 alone.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17576
PDF
-
Budgeted KV-Cache Repair for Evolving Contexts
2026-10-12 19:44:57Han Yu, Wenhui Zhu, Jincheng Cao, Rui Cai, Yuanda Xu, Zhengze Zhou, Xiwen Chen, Hao Wang, Xuanzhao Dong, Han Shi, Menglin Zhou, Shayan Mohajer Hamidi, Lin Su, Mina Razghandi, Sayan Ghosh, Qing Lan, Rohit Jain, Alborz GeramifardAbstract
Language-model agents revise their working context as information becomes obsolete, but deleting text does not erase its influence on cached computation. Later key-value (KV) states remain conditioned on the deleted content, while recomputing the entire retained suffix can be costly. We introduce budgeted KV-cache repair: recovering task performance by recomputing a user-chosen fraction of the retained suffix. Comparing repair allocations under identical relevance scores and token budgets, we find that refreshing coherent regions can substantially outperform refreshing isolated tokens. This finding motivates Semantic-Unit Refresh (SUR), which allocates the budget to query-relevant text regions using inexpensive embedding scores, without training or an auxiliary transformer forward. To execute these repairs efficiently, a layer-packed executor processes selected states together at each layer while preserving full causal cache access. Across five models, SUR achieves strong recovery in controlled deletion tasks while recomputing only 2-5% of the suffix. On four natural multi-hop QA datasets, it achieves near-full-recompute exact match at a 5% budget with lower post-edit latency. The executor accelerates fragmented refresh by up to 16.1x over interval-wise execution.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17338
PDF
-
When Agents Commit Too Soon: Diagnosing Premature Commitment in LLM Agents
Abstract
Tool-using LLM agents make early choices about what to search for, which evidence to trust, and what to do next. A wrong early choice can steer every later tool call, so an agent may produce a coherent but incorrect result. Existing evaluation does not reveal this process: final-answer scoring comes only after completion, and agreement across runs can be high even when every run is wrong. We ask whether hidden states reveal that repeated runs have already settled into stable behavior. Our contributions are threefold. First, we introduce representational commitment, which measures hidden-state similarity across repeated runs at the same step. On Llama-3.1-70B running ReAct on HotpotQA, we show that higher step-4 similarity is associated with lower whole-trajectory variability (r = -0.35); we find the same pattern in action sequences and replicate it on two other models and StrategyQA. This provides an internal measure for studying when agent behavior stabilizes. Second, we establish the measure's scope. Shared evidence explains much of the observed association, the HotpotQA result disappears on hard questions, and correct and confidently wrong questions overlap. We therefore show that the measure detects stability, not correctness, and cannot by itself establish that commitment was premature. Third, we evaluate monitoring and intervention. A monitor built from three partial runs predicts whole-trajectory variability (mean AUROC 0.81). A commitment prompt reduces variability by 28% relative to a token-matched filler, but not significantly relative to the standard agent after correction, and it does not improve accuracy. Together, these contributions provide a measurement and evaluation framework for studying how agent trajectories stabilize and for testing when additional evidence, sampling, or review may be useful.
Abstract (translated)
URL
https://arxiv.org/abs/2606.22936
PDF
-
PitchIn: Natural-Language Vocal Coaching from Unstructured Community Feedback
2026-10-12 16:12:36Hayeon Bang, Hounsu Kim, Joonyong Park, Jung-Jae Yu, Juhan NamAbstract
Effective vocal coaching connects performance issues with corrective advice, but recordings paired with expert diagnoses and advice are scarce. Online singing communities provide a potential source of supervision, but observations and advice are scattered across unstructured comments. We introduce PitchIn, audio-language models for natural-language vocal coaching, and the PeerCoach Dataset: 14,445 instruction examples covering Focused QA, Overall Feedback, and Recognition QA. We compare target combinations on Qwen2.5-Omni and train PitchIn on Focused QA and Overall Feedback using Qwen2.5-Omni and MOSS-Music. On VocalCoachBench, the proportion of requested diagnosis-correction pairs matching an expert issue with valid advice rises from 21.3% to 61.0% for Qwen2.5-Omni and from 31.4% to 51.2% for MOSS-Music. Direct training on source comments yields only 19.8% for Qwen2.5-Omni. For Qwen, Focused QA alone reaches 65.6%, but adding Overall Feedback improves alignment with expert reviews and human preference. Both models improve expert-review alignment over pretrained baselines and are significantly preferred for usefulness (92.0% and 67.3%, respectively). Demo examples are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17036
PDF
-
MIND: Lightweight and Effective Memory Injection Defense for LLM Agents via Intent-Aware Information Bottleneck
2026-10-12 16:09:44Dongyi Liu, Haixing He, Xiaobao Wu, Jia LiAbstract
Memory-augmented LLM-based agents are vulnerable to memory injection attacks: Agents may retrieve poisoned memory from attackers, which diverts their behavior from initial user intent and finally causes task failure. However, existing defense mechanisms either incur high computational cost or suffer from information redundancy in multi-turn contexts. To address these challenges, we propose Memory Intent-Aware Neural Denoising(MIND), a lightweight defense framework for memory injection attack. Our preliminary analysis reveals that benign and poisoned trajectories exhibit distinguishable relationships between the initial user intent and subsequent behavior. Building on this observation, MIND employs an intent-aware Information Bottleneck(IB) to extract compact intent--behavior representations from the initial intent and turn-level behavior. The IB preserves intent-relevant cross-turn attack signals while filtering task-irrelevant and repetitive information, and a lightweight detector identifies malicious memories from the resulting representations. As such, MIND mitigates information redundancy in multi-turn contexts while avoiding the overhead of repeated LLM auditing. Extensive experiments show that MIND reduces attack success rates while preserving task accuracy and inference efficiency. Notably, on ReAct-StrategyQA, MIND reduces mean ASR-r and ASR-a by 58.1% and 53.5%, respectively, while matching the undefended agent in average accuracy and latency.
Abstract (translated)
URL
https://arxiv.org/abs/2607.28103
PDF
-
The Value of Being Wrong: Self-Mined Visual In-Context Learning from Errors
2026-10-12 15:38:27Shu Zhao, Tan YuAbstract
Visual in-context learning (ICL) enables vision-language models (VLMs) to adapt at test time via demonstration examples, but its effectiveness depends critically on the examples selected. Recent counterfactual selection methods improve ICL by constructing demonstrations that expose how individual attribute changes affect the answer. However, they require multi-stage pipelines involving attribute extraction, caption engineering, and composed image matching. We propose a simpler alternative grounded in a key observation: a model's own prediction errors are natural counterfactual demonstrations that directly encode its failure modes. Our method, SMILE (Self-Mined Visual In-Context Learning from Errors), builds a model-specific pool of hard negatives offline by recording the target VLM's own mistakes. At test time, the prediction serves as a direction signal that surfaces, in the semantic embedding space, the past mistakes most likely to recur for the current query. This prediction-conditioned identification provides a unified mechanism for both closed-set classification and open-ended visual question answering without task-specific design. Across four benchmarks and four VLMs, SMILE delivers an average task-score gain of +3.09 points over the SOTA baseline at 5x lower latency. Project page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.16977
PDF
