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

+0新增论文
+0其中 AI 相关
+1已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-08 23:55 · 全站最近入库：2026-11-08 06:02各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 00:42 — 2026-11-09 00:42（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：2026-11-08 23:55最近发现新增版本：2026-11-08 06:02库内论文最新更新时间：2026-11-06 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。

17,100 已收录论文14,661 AI 相关4,057 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 5,059 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-08 23:54（北京时间）
新论文 · v1arXiv:2611.06065

ProcessPrefBench: Evaluating Cross-Task Transfer of Individual Process Preferences from Naturalistic Human-AI Interaction

Yiwen Yang, Guanming Liu

原文摘要

In natural human-AI dialogue, users' corrections, requests, and self-descriptions carry reusable signals about how they want work to be done. These signals reveal individual process preferences, user-specific standards for evidence, depth, analytical focus, procedure, and delivery. We ask whether a preference revealed in one dialogue can shape an assistant's later answer when the task does not restate it. ProcessPrefBench uses three months of natural dialogues from two knowledge workers to build 66 source-target cases separately judged for source evidence and target applicability. Sources are preselected, so the benchmark tests application rather than retrieval. We test three models with target-only, full-history, and distilled-preference prompts. Target-specific 0-2 rubrics score preference realization, and a task gate flags severe failures. Compared with target-only prompting, full-history and imperative distilled-preference prompting raise preference realization by 0.204 and 0.244 points on the 0-2 scale, respectively, with no reliable advantage from distillation. Fabricated-evidence failures rise from 5.8% under target-only prompting to 10.1% under full-history and 10.4% under imperative prompting. ProcessPrefBench highlights a tension in learning from dialogue, where drawing on prior interactions can improve preference realization yet also invite fabricated evidence. We hope it helps build assistants that honor a user's process preferences without sacrificing factual reliability.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06062

Adapting Qwen3 to Tajik from 0.6B to 8B: Lower Text Loss Did Not Mean Better Understanding

Tohir Saidzoda, Madina Abdusalomzoda, Nasimjon Rajabov, Akhrorkhon Ikromov, Komron Giyosiev, Abubakr Abdullozoda, Amin Jalilov, Jonibek Kishvarov 等 10 位作者

原文摘要

Small LoRA adapters trained on Tajik news and books lowered Qwen3's Tajik text loss at every size we tried, from 0.6B to 8B, by 0.60 to 1.45 nats per token, but none of them improved Tajik reading comprehension where it can be measured. On Belebele the 1.7B and 8B adapters left accuracy unchanged and the 4B adapter lowered it by 3.7 points; at 0.6B the score rose by 4.2 points but sits near chance and coincides with a change in which answer letter the model prefers, so we do not read it as understanding. At 8B, continued pretraining on about 375 million tokens of raw Tajik text was followed by a Belebele gain of 9.9 points. After a further supervised fine-tuning stage, our main model, e3, is 11.4 points above Qwen3-8B on Belebele; its English MMLU changed by -1.2 points, and a loss larger than 3.3 points is excluded. The same heavy recipe applied to Qwen3-0.6B-Base brought its text loss close to that of the 8B model after books-only pretraining, while its multiple-choice answers stayed near chance and its English MMLU fell by 7.6 points. We report 15 trained models and their 5 starting models from June to October 2025, scored in 2026 on one harness with paired intervals, and we report what went wrong: lost training records, a thinking-mode dataset with almost no reasoning in it, later branches whose training data was never documented, and a 4-bit release that contains no transformer layers.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2607.05721

SpanUQ: Span-Level Uncertainty Quantification for Large Language Model Generation

Yimeng Zhang, Yingying Zhuang, Ziyi Wang, Yuxuan Lu, Pei Chen, Aman Gupta, Zhe Su, Ming Tan 等 15 位作者

原文摘要

Uncertainty estimation is essential for the trustworthy deployment of LLMs: it underpins self-refinement and serves as a decision signal for downstream systems, from verifiers to collaborating agents. However, existing approaches operate at suboptimal granularities: token-level scores lack semantic coherence, while sequence-level scores fail to localize errors. We formalize Span-Level Uncertainty Estimation (SLUE), a new task that targets the natural granularity for uncertainty: semantically coherent text spans, each conveying a single assessable unit of meaning. To address this task, we introduce SPANUQ, a lightweight probe that distills the uncertainty knowledge from expensive multi-sample inference into a single forward pass over LLM hidden states. SPANUQ employs a DETR-style span decoder to simultaneously detect spans and estimate their uncertainty via a Mixture of Beta distribution, trained with a principled combination of Beta NLL regression and contrastive ranking objectives. We construct SPANUQ-BENCH, the first span-level uncertainty benchmark comprising 20K prompts, 293K annotated spans, and continuous soft labels derived from multi-sample claim verification. Experiments on five LLM backbones show that SPANUQ consistently achieves the best span-level uncertainty quality , outperforming the strongest probe baseline and all sampling-based methods while being 10~20x faster. Its DETR-based span detector attains 0.910 F1, surpassing the best heuristic by 39.4%, enabling precise error localization that sequence-level methods cannot provide. The same architecture ports to five LLMs spanning two model families, with one probe trained per backbone, and we additionally observe that sequence-level uncertainty is partially decomposable, suggesting that span-level estimation subsumes sequence-level as a special case. The project page is available damon-demon.github.io/SpanUQ.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06048

SVGFaith: A Two-Layer Meta-Evaluation of Faithfulness Metrics for Text-to-SVG Generation

Guotao Liang, Haitao Zhou, Biao Liu, Huanxuan Liao, Lingfeng Qiao, Qian-Wen Zhang, Qian Yu, Di Yin 等 9 位作者

原文摘要

Text-to-SVG systems are commonly evaluated by applying raster-image metrics to rasterised outputs, yet whether these scores capture fine-grained prompt faithfulness remains unclear. We introduce SVGFaith, a two-layer meta-evaluation framework in which the more faithful candidate of each pair is determined by construction or by human judgment. SVGFaith-Diagnostic contains 1,842 controlled contrastive pairs across eight capability axes, split evenly between two directions: two renderings compared against one prompt, and two descriptions compared against one rendering. SVGFaith-Arena contains 1,327 human-annotated comparisons of real generator outputs. On SVGFaith-Diagnostic, CLIP and three other image--text embedding metrics reach at least 0.789 pairwise accuracy on object identity but at most 0.672 on spatial relations and attribute binding, where chance is 0.5. We further introduce SVGFaith-Score, an interpretable pointwise metric. A domain-tuned decomposer turns each prompt into a dependency graph of atomic visual requirements, a visual question answering model scores whether each requirement is satisfied, and a requirement's score is discounted when its prerequisites score low. In addition to the overall score, the per-requirement scores show which requirements are judged unsatisfied and which inherit discounts from their prerequisites. SVGFaith-Score significantly outperforms all nine baselines, including a Qwen3.5-122B-A10B judge, on Arena and in both Diagnostic directions. Code and data: https://github.com/Yukinonooo/SVGFaith.

cs.CV · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06043

Confucius4-R2T2: A Stable-Prefix Streaming ASR Model for Real Real-Time Transcription

Daniel Huang, Qi He, Shaoxiong Duan, Yingjie Li, Qiao Cheng, Yitao Duan

原文摘要

Streaming speech recognition is increasingly consumed by machines, such as translators and voice agents, that act on each token as it appears. A token that is later revised costs wasted work or a wrong action. In this work, we present Confucius4-R2T2 (short for Real Real-Time Transcription), a stable-prefix streaming ASR model that never modifies its earlier output while maintaining high accuracy and low latency. R2T2 is obtained by post-training Qwen3-ASR-1.7B on 1,500 hours of Chinese and English audio while freezing its pre-trained audio encoder and projector. The adaptation is based on the concept of Longest Stable Prefix (LSP), a new notion that we introduce in this work. We prove that LSP is monotone under the prefix order, consistent with the full-audio output, and maximal among labels that equal a whole earlier output. Learning from LSP labels converts a full-context model into a stable-prefix streaming one without any architectural change. Extensive evaluations show that R2T2 brings streaming recognition close to the base model's full-context quality while delivering stable text with low emission lag. Its gains extend beyond the post-training languages, demonstrating a practical path from full-context ASR to stable streaming. We open-source the model weights and code at https://huggingface.co/netease-youdao/Confucius4-R2T2 and https://github.com/netease-youdao/Confucius4-R2T2.

cs.CL · cs.AI · cs.SD

查看论文
arXiv ↗PDF ↗
版本更新 · v6arXiv:2603.15159

To See is Not to Master: Teaching LLMs to Use Private Libraries for Code Generation

Yitong Zhang, Chengze Li, Ruize Chen, Guowei Yang, Xiaoran Jia, Yijie Ren, Jia Li

原文摘要

Large Language Models (LLMs) have shown strong potential for code generation, yet they remain limited in private-library-oriented code generation, where the goal is to generate code using APIs from private libraries. Existing approaches mainly rely on retrieving private-library API documentation and injecting relevant knowledge into the context at inference time. However, our study shows that this is insufficient: even given accurate required knowledge, LLMs still struggle to invoke private-library APIs effectively. To address this limitation, we propose PriCoder, an approach that teaches LLMs to invoke private-library APIs through automatically synthesized data. Specifically, PriCoder models private-library data synthesis as the construction of a graph, and alternates between two graph operators: (1) Progressive Graph Evolution, which improves data diversity by progressively synthesizing more diverse training samples from basic ones, and (2) Multidimensional Graph Pruning, which improves data quality through a rigorous filtering pipeline. To support rigorous evaluation, we construct two new benchmarks based on recently released libraries that are unfamiliar to the tested models. Experiments on three mainstream LLMs show that PriCoder substantially improves private-library-oriented code generation, yielding gains of over 20% in pass@1 in many settings, while causing negligible impact on general code generation capability.

cs.SE · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06025

When Every Choice Is Correct: A Causal Study of Numerical Preference in Language Models

Hongnian Huang, Hongliang Liu

原文摘要

When a language model chooses a number between 1 and 30, every integer in the range is valid, yet the model can favor particular answers. We use numerical choice to study how preference is processed across layers and tokens. Perturbation probing and gradient attribution nominate neurons for activation interventions. In a selected Qwen3-4B case, a pre-answer edit changes 17 to 15 while both runs first emit 1. After the edit ends, selective cache and attention-output replacements trace part of its continuing influence on the second digit. Replacing one attention head's output or the other heads' outputs jointly produces similar immediate readout contributions; both interventions restore 17 but have different final log-odds effects on 7 versus 5. Holding later multilayer perceptron (MLP) outputs at their values from the edited run without either replacement reduces this effect gap from 0.119 to 0.0025 nat. Later computation therefore changes how similar local contributions affect the continuation. Retrospective estimates across eighteen contexts capture much of the probability movement but miss the selected answer switch. Prompt tests also separate influence from answer changes: one held-out prompt retains influence without switching, and only one of twelve fresh wordings qualifies for the reader/MLP test; there, joint MLP clamping again reduces the gap between the two reader effects while the answer remains 17. These results use numerical choice as a controlled case for studying persistent state and conditional downstream processing; they do not establish the origin or a complete circuit of natural numerical preference.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06022

Who Pays for the Knowledge Models Copy? Model Collapse as Rogers' Paradox

Nam H. Le

原文摘要

Generative models trained on the output of earlier models lose knowledge across generations, rare knowledge first. The remedies that work all bring in information from outside the loop, and the theory of model collapse treats that supply as given: a fixed archive, drawn from a fixed distribution. We argue that the supply is the problem, and that cultural evolution has the theory of it. We read model collapse as Rogers' paradox: a population that comes to rely on copying ends up no better off than one that never copied, because copies go stale unless someone keeps learning for themselves. We test the identification in three steps. First, Rogers' model, fitted only to fully synthetic training, predicts to within a few percentage points how much rare knowledge language models, from word trigrams to a pretrained transformer, retain under other mixtures of fresh data. Second, when the world itself changes, a lineage that copies its own past needs more fresh data to keep up the faster the world changes, as Rogers' model says it must and as analyses that assume a fixed target cannot; a lineage that restarts every generation from fixed pretrained weights is largely insulated, because those weights are knowledge that is never copied. Third, we close the loop: when the people who write a model's training data can ask the model instead of finding things out, the supply of fresh data becomes endogenous. At equilibrium, the model is exactly as reliable as it is costly for people to find out for themselves, the population is no better off than without the model, and every increase in the share of model-written text must be paid for by more people learning for themselves. Users who check what they are told and relearn when it is wrong break the paradox. Filters that select plausible text are conformist and accelerate collapse; only checks that carry new information act as critical learning.

cs.LG · cs.CL · q-bio.PE

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06010

Holistic Parallel Supervision for Expressive Speech to Speech Translation

Hao Zhang, Ismail Rasim Ulgen, Bismarck Bamfo Odoom, Rui Liu, Zihan Zhang, Berrak Sisman, Philipp Koehn

原文摘要

Expressiveness is central to speech-to-speech translation (S2ST), yet existing datasets rarely align source and target speech jointly in content, prosody, timing, and speaker identity. Consequently, current systems often rely on indirect conditioning or transfer mechanisms. To obtain this supervision, we introduce a training-data construction strategy that combines professional dubbing, which provides human-performed correspondence in content, prosody, and timing, with multilingual voice conversion to reduce speaker mismatch. We evaluate this supervision using existing expressive S2ST systems and DirectS2ST, an end-to-end method developed in this work that predicts the complete target-speech representation directly from source speech. Experiments on English-German and English-Spanish show that holistic parallel supervision improves expressive realization across all evaluated systems. DirectS2ST exhibits the largest gains in content, prosody, timing, speaker similarity, and human-rated expressiveness. To our knowledge, this work provides the first systematic investigation of holistic parallel supervision for expressive S2ST.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06006

Familiar letters, not related languages, let unseen languages into a shared space of meaning

Yuliang Liu, Xinyu Wang, Xiang Bai

原文摘要

Hundreds of languages have written text but almost no translations, and whether a model learning one from untranslated text builds on a related language or familiar letters is unclear: they usually coincide. Here we show that familiar letters, not a related language, let an unseen language into a shared space of meaning. GlyphMind, a tokenizer-free vision transformer that reads text as pixels and lets any sentence be shown in any script, was pretrained to align translations across 27 writing systems. Adapted on untranslated Lao, a language and script it had never seen, it retrieved English translations seven times as often as the strongest simple baseline. Rewritten in unfamiliar letters, Lao did not enter although its close relative Thai was in pretraining; Santali, which had none, entered in Devanagari, familiar from Hindi, but not in its own script. Transliteration into familiar scripts may thus open multilingual models to languages without translations.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.06001

How Far Can You Steer? Behavior-Capability Frontiers in Contrastive Activation Addition

Deepanshu Raj

原文摘要

Activation steering can modify language-model behavior without parameter updates, but behavioral efficacy alone does not determine whether an intervention is usable. Contrastive Activation Addition (CAA) steers targeted behaviors using activation directions derived from contrasting examples, potentially degrading unrelated model capabilities in the process. We therefore study CAA through a behavior-capability frontier, selecting interventions that maximize behavioral gain subject to an explicit bound on capability degradation. We evaluate this framework across four instruction-tuned models, seven behaviors, and capability-loss budgets of 1%, 3%, and 5% (absolute percentage points). Validation-selected behavioral effects generalize strongly to held-out data ($ρ= 0.945$), while joint behavior-capability survival among nonzero operating points rises from 37.5% at the 1% budget to 72.7% at 3%, with limited additional improvement at 5%. Control experiments indicate that these effects depend on learned behavior directions rather than generic activation perturbations. Representational geometry is significantly but only modestly predictive of causal cross-behavior transfer. Finally, secondary capability, out-of-distribution, and open-ended evaluations show that interventions satisfying protected benchmarks can still degrade untested capabilities or transfer unevenly to free-form generation. These results characterize activation steering as a constrained intervention whose usable range depends on model, behavior, direction, layer, and evaluation budget, rather than a universal control knob.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.21849

Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift

Sungjun Lim, Heedong Kim, Andrew Lee, Kyungwoo Song

原文摘要

Mechanistic interpretability aims to explain a model's behavior by identifying causally responsible internal structures. Dictionary-based explainers such as sparse autoencoders and transcoders are a primary tool, but their faithfulness under out-of-distribution (OOD) shift has received little systematic attention. We show that distribution shift rotates the subspace that the model actively uses, misaligning the explainer's dictionary trained on in-distribution (ID) activations. We formalize this misalignment as the faithfulness gap, a geometric distance between the ID dictionary and the OOD-active subspace, and show that it bounds excess OOD projection loss. To reduce this gap, we propose the Geometry-Adaptive Explainer (GAE), which realigns the dictionary with the OOD-active subspace while preserving the original sparse feature activations. This requires only unlabeled OOD activations and no gradient updates. For the adapted dictionary, we also bound the excess projection loss caused by any remaining faithfulness gap. Empirically, GAE even matches or surpasses all training-based baselines in causal faithfulness across multiple models and OOD settings.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05984

On the Design of Hyper-Connections: Simplification, Stability, and Scalability

Xiangdong Zhang, Xiaohan Qin, Sunan Zou, Tuo Dai, Bizhao Shi, Zhuo Xia, Yebin Yang, Yizhou Chen 等 11 位作者

原文摘要

Hyper-Connections (HC) architectures are increasingly adopted in frontier language models, yet their components remain insufficiently understood and introduce non-negligible training overhead. In this work, we systematically study the HC architecture to understand its components and identify simpler, more efficient designs. We find that learned read/write gating without mixing and unconstrained mixing with fixed gates form two effective interaction groups: either alone achieves performance comparable to mHC. We also identify repeated access to the expanded residual state as a major source of training overhead, even without mixing. These findings lead to SimpleHC, which simplifies the architecture by removing residual mixing and simplifies execution through shared memory access, using the resulting additive structure and shared gate inputs. Layer-specific gates share a group-entry state, enabling batched reads and deferred writes; scalar corrections exactly preserve preceding sublayer output contributions. Our analyses support both the retained residual structure and the design of shared memory access. SimpleHC incurs only 2.4% extra end-to-end training time over a vanilla residual baseline at the largest tested scale. It maintains performance comparable to mHC when scaling from 15B to 30B MoE models, exhibits smoother gradient dynamics during standard training, and remains stable under stress. We hope SimpleHC will encourage a broader pursuit of simplicity in foundation model design, making performance gains more accessible and practical at scale.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2603.03415

Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs

Mingyu Jin, Yutong Yin, Jingcheng Niu, Qingcheng Zeng, Wujiang Xu, Mengnan Du, Wei Cheng, Zhaoran Wang 等 10 位作者

原文摘要

In this work, we investigate how the internal representations of Large Language Models (LLMs) change with inputs of increasing difficulty. Although sparsity changes may not be obvious for individual samples, dataset-level results reveal a consistent statistical trend: as task difficulty increases (e.g., harder questions, longer contexts, or more answer choices), the last hidden states of LLMs become systematically sparser. In short, \textbf{the harder the task, the sparser the representation}. This sparsity--difficulty relation is observable across diverse models and domains, suggesting that harder inputs drive more concentrated activation patterns in the last hidden state. Through a series of controlled analyses and learning-dynamics experiments, we show that representational density is a learned property of data familiarity: models develop rich, distributed representations for mastered patterns, while unfamiliar inputs default to sparser activations. Our finding is also actionable. We illustrate this with Sparsity-Guided Curriculum In-Context Learning (SG-ICL), which uses sparsity to select few-shot demonstrations matched to the query's difficulty, outperforming standard CoT and Auto-CoT baselines on MATH-500. Our study provides new insights into how LLM representations reflect task difficulty and how this signal can be leveraged for inference. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.04878

TraceRouter++: Causal Semantic Routing Realignment via Optimal Transport

Chuancheng Shi, Shangze Li, Wenhua Wu, Hengzhi Chen, Fei Shen, Li Zhuo, Zhiyong Wang, Tat-Seng Chua

原文摘要

Despite their remarkable capabilities, large foundation models (LFMs) remain susceptible to adversarial manipulation. Current safety intervention methods primarily rely on the \textit{locality hypothesis}, rigidly suppressing isolated neurons or features. However, harmful semantics manifest as distributed, cross-layer circuits, rendering localized ablations both brittle against complex jailbreaks and detrimental to general utility. To bridge this gap, we propose TraceRouter++, a routing-level safety intervention framework that explicitly traces safety-relevant causal routing pathways activated under harmful inputs and realigns their unsafe activation states. Our framework operates in a \textit{Discover-Trace-Realign} paradigm: it first pinpoints a sensitive onset layer via attention divergence and disentangles dense activations into interpretable routing features using sparse autoencoders (SAEs). Subsequently, it maps these features to downstream causal circuits via feature influence scores (FIS). Crucially, departing from rigid suppression paradigms, TraceRouter++ introduces a constrained optimal transport (OT)-based soft routing realignment. By formulating safety intervention as a tractable coordinate-wise marginal optimal transport approximation, our method redirects unsafe activation marginals toward safe-reference marginals within causally identified routing subspaces while preserving non-selected coordinates at the intervention point. This mechanism safely redirects malicious semantic flows while leaving orthogonal computational routes intact. Extensive experiments across representative large language models (LLMs), multimodal large language models (MLLMs), and diffusion models (DMs) show that TraceRouter++ outperforms evaluated baselines, improving adversarial robustness while maintaining a favorable safety-utility trade-off. WARNING: This paper contains unsafe model responses.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2505.09338

Llama See, Llama Do: A Mechanistic Perspective on Contextual Entrainment and Distraction in LLMs

Jingcheng Niu, Xingdi Yuan, Tong Wang, Hamidreza Saghir, Amir H. Abdi

原文摘要

We observe a novel phenomenon, contextual entrainment, across a wide range of language models (LMs) and prompt settings, providing a new mechanistic perspective on how LMs become distracted by ``irrelevant'' contextual information in the input prompt. Specifically, LMs assign significantly higher logits (or probabilities) to any tokens that have previously appeared in the context prompt, even for random tokens. This suggests that contextual entrainment is a mechanistic phenomenon, occurring independently of the relevance or semantic relation of the tokens to the question or the rest of the sentence. We find statistically significant evidence that the magnitude of contextual entrainment is influenced by semantic factors. Counterfactual prompts have a greater effect compared to factual ones, suggesting that while contextual entrainment is a mechanistic phenomenon, it is modulated by semantic factors. We hypothesise that there is a circuit of attention heads -- the entrainment heads -- that corresponds to the contextual entrainment phenomenon. Using a novel entrainment head discovery method based on differentiable masking, we identify these heads across various settings. When we ``turn off'' these heads, i.e., set their outputs to zero, the effect of contextual entrainment is significantly attenuated, causing the model to generate output that capitulates to what it would produce if no distracting context were provided. Our discovery of contextual entrainment, along with our investigation into LM distraction via the entrainment heads, marks a key step towards the mechanistic analysis and mitigation of the distraction problem.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05954

Robots in the City: Persona-Based Perceptions of Urban Robots

Keonhee Jang, Kanghoon Lee, Seokwoo Yun, Namwoo Kim, Yoonjin Yoon

原文摘要

Pedestrians increasingly encounter robots in urban public spaces without interacting with them, but how robots change the perceived safety of a street remains unclear. We examine how robot presence and type affect perceived safety across urban contexts. We generate 18 images of Seoul streets that vary in time of day, road class, and robot condition (no robot, a delivery robot, or a humanoid). Two multimodal large language models (MLLMs), Qwen3.8-27B and Gemma4-26B, each conditioned on 500 synthetic personas, choose the safer image in each pair, and 25 human respondents complete the same task. Both models prefer streets without robots and favor the delivery robot over the humanoid. Robots are chosen more often at night and less often by older personas, whereas road class has smaller effects. In the model rationales, the delivery robot appears as a physical obstacle, whereas the humanoid appears as either a threat or a source of protection. Like humans, both models chose robots more often at night, but chose the humanoid over the delivery robot far less often, and Gemma showed a night effect for the humanoid about three times that of humans. These results show that the effect of a robot on perceived safety depends on when it appears, and that persona-conditioned MLLMs capture only part of human judgments.

cs.HC · cs.CL · cs.RO

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.36513

MedConstBench: Benchmarking How Constitutions Steer LLMs in Medical Ethics

Yibo Yang, Zerui Huang, Haomiao Yang, Zixia Wang

原文摘要

As large language models advance rapidly, they are increasingly used to answer medical questions, and many such models are now deployed locally. Once a model is deployed, its behavior is governed mainly by the ethical policy written into its constitution. However, most existing work focuses on factual correctness rather than the trade-off between competing ethical values. Even benchmarks that do evaluate ethics treat it as a fixed property of the model rather than as a variable, so how a constitution reshapes medical-ethics behavior remains largely unexplored. We introduce MedConstBench, a medical-ethics benchmark that treats the deployment constitution as a controlled variable, is organized around Beauchamp's four principles, and is scored by human-calibrated multi-model peer review. It includes MedDilemma, our own dataset of clinical dilemmas, and supports a systematic analysis of how constitutions steer LLMs. Evaluating ten open-source LLMs under four constitutions, we find that the effect of a constitution is dimension-selective and model-dependent, and that no constitution is universally optimal. Motivated by these findings, we propose a do-no-harm adaptive selector that chooses a constitution per model, matching the per-model oracle while guaranteeing a worst-case floor and generalizing to unseen models at low cost.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05927

Modality-Aware Expert Pruning for MoE-Based Multimodal Large Language Models

Donggeun Kim

原文摘要

Mixture-of-Experts (MoE) enables multimodal large language models (MLLMs) to scale capacity efficiently, but expert parameters dominate memory even though only a few experts are activated per token: they account for over 90% of model size in modern MoE-MLLMs (e.g., 54 GB out of about 58 GB in BF16 for InternVL3.5-30B-A3B and Qwen3-VL-30B-A3B). Expert pruning has emerged as an effective compression strategy, but existing methods aggregate importance uniformly across tokens, ignoring that the same expert may serve different roles for vision versus language processing. To address this, we propose MAEP (Modality-Aware Expert Pruning), a training-free framework that decomposes expert importance by modality in a single forward pass. MAEP introduces (1) Delta-H Output ($ΔH$), a redistribution-aware metric measuring the MoE output change when an expert is removed, and (2) low-attention vision tokens as a vision-side pruning signal that is dataset-stable: its expert-importance ranking remains stable across different datasets. Text-side importance serves as a preservation signal while vision-side importance serves as a pruning signal, and experts are ranked globally across layers. Experiments on three MoE-based MLLMs across multimodal and text benchmarks show that MAEP outperforms modality-agnostic baselines on the majority of multimodal benchmarks while remaining competitive on text-only tasks, reducing the BF16 expert weight footprint by about 27 GB (about 47% of total memory) so that the full weight footprint of an approximately 58 GB model fits within a single 40 GB GPU. Code is available at https://github.com/nota-github/MAEP.

cs.CV · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05919

Routes of Reasoning: Reading and Reshaping Test-Time Compute in MoE Models

Shengjie Wang, Mohsen Fayyaz, Nanyun Peng

原文摘要

Reasoning models allocate test-time compute across distinct behaviors such as planning, deriving, and verifying. However, models frequently "overthink," continuing to verify and re-derive long after finding a correct answer. This wastes expensive compute and can even degrade performance. Existing interventions, such as token budgets or steering, primarily regulate reasoning duration rather than targeting specific behaviors. In this work, we demonstrate that Mixture-of-Experts (MoE) routers provide a natural, training-free interface for behavioral control. We first split reasoning traces into sentence-level segments and label each with one of six reasoning behaviors (planning, execution, restatement, fact, reflection, and conclusion), so that every token carries the behavior of its segment. Linear probes on router logits predict this label nearly as well as probes on the residual stream, the model's full hidden state and the standard target for probing, while using about half as many features. These signals generalize: experts associated with a reasoning behavior on one dataset predict the same behavior on the others, across six datasets spanning five domains (math, code, science, knowledge, and commonsense). Building on this, we selectively edit expert router scores during inference to curb overthinking. Across four models and four benchmarks, this intervention reduces the number of generated tokens by a median of 30.4% (and up to 63.3%). Crucially, it encourages the model to stop once the problem is solved, rather than skipping essential reasoning steps. Accuracy never drops by more than 0.5 percentage points, and frequently improves. Ultimately, MoE routing offers both an interpretable lens into reasoning states and a lightweight mechanism for optimizing test-time compute.

cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05911

Autonomous agents for electricity-system planning

Yoon Pyo Lee, Samrendra Roy, Jay Phil Yoo, Kazuma Kobayashi, Souvik Chakraborty, Syed Bahauddin Alam

原文摘要

Planning electricity systems requires repeated decisions that connect regional resources, investment, operation and network constraints. Here we present an autonomous planning framework that carries prepared regional scenarios through proposal, numerical evaluation, feedback interpretation, redesign and final plan selection. Agents propose investment bounds, storage operating conditions and feeder interventions, interpret numerical feedback, and decide how to revise designs and when to stop. The framework automatically executes annual investment-and-dispatch optimization and representative-feeder verification, applies fixed acceptance criteria and retains the lowest-cost accepted candidate. We evaluate forty carbon-constrained trajectories across ten system-year scenarios in five electricity systems and four countries, using two model deployments and single- and multi-agent architectures. All forty retain candidates satisfying the implemented physical and fixed annual operational carbon-budget checks. Twenty-eight of forty matched reference designs exceed the new scenario ceilings. A separate archive of ninety-six trajectories under a common 1,000 Mt annual ceiling provides matched references and broader regional coverage. The results demonstrate autonomous planning under explicit emissions constraints using public regional data and published institutional demand outlooks.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2607.05369

GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks

Kaiyuan Chen, Shuangyu Xie, Letian Fu, Justin Yu, William Pacini, Sandeep Bajamahal, Hudson Kim, Jaimyn Drake 等 24 位作者

原文摘要

For robots to work reliably in commercial and industrial applications, recent advances in agentic coding systems have potential to combine interpretable robot programming with the open-world adaptability of model-free policies. We focus on "Variational Automation" (VA), a class of tasks that have larger variations in object geometry and pose than fixed automation. Model-free policies often struggle to close the reliability gap for VA tasks, which must be executed persistently and reliably in commercial and industrial applications. Motivated by prior work on Task and Motion Planning (TAMP) and the Robot Operating System (ROS), we introduce Graph-as-Policy (GaP), a multi-agent coding harness that generates directed computation graphs with perception, planning, and control nodes from a Modular Open Robot Skill Library (MORSL). GaP generates an internal simulation environment to rehearse task instances with different graphs in parallel to iteratively refine the graph structure and parameters to improve success rates and throughput. Evaluation with 8 new open VA task benchmarks, 4 in simulation and 4 in real-world, suggests that GaP can achieve success rates that significantly outperform baselines. Details, code, and data: https://graph-robots.github.io/gap

cs.RO · cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2602.09870

Steer2Edit: From Activation Steering to Component-Level Editing

Chung-En Sun, Ge Yan, Zimo Wang, Tsui-Wei Weng

原文摘要

Steering methods influence Large Language Model behavior by identifying semantic directions in hidden representations, but are typically realized through inference-time activation interventions that apply a fixed, global modification to the model's internal states. While effective, such interventions often induce unfavorable attribute-utility trade-offs under strong control, as they ignore the fact that many behaviors are governed by a small and heterogeneous subset of model components. We propose Steer2Edit, a theoretically grounded, training-free framework that transforms steering vectors from inference-time control signals into diagnostic signals for component-level rank-1 weight editing. Instead of uniformly injecting a steering direction during generation, Steer2Edit selectively redistributes behavioral influence across individual attention heads and MLP neurons, yielding interpretable edits that preserve the standard forward pass and remain compatible with optimized parallel inference. Across safety alignment, hallucination mitigation, and reasoning efficiency, Steer2Edit consistently achieves more favorable attribute-utility trade-offs: at matched downstream performance, it improves safety by up to 17.2%, increases truthfulness by 9.8%, and reduces reasoning length by 12.2% on average. Overall, Steer2Edit provides a principled bridge between representation steering and weight editing by translating steering signals into interpretable, training-free parameter updates. Our code is available at https://github.com/Trustworthy-ML-Lab/Steer2Edit

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05896

Selective, Risk-Controlled Emotion Forecasting in Conversation: From a Persistence Barrier to Departure Detection with Conformal Guarantees

Bin Wen, Dai-Qiao Zhang, Tien-Ping Tan

原文摘要

Forecasting emotion in conversation is increasingly framed as predicting the next emotional state's value. We show this targets a hard quantity in a state-observed testbed and propose a learnable, deployable reformulation instead. On valence-arousal-dominance trajectories from a dyadic corpus, repeating a speaker's own previous state sets a persistence baseline that no frozen-encoder correction meaningfully improves upon and a state-space filter exceeds; the floor requires the historical state to be observed, since an estimated anchor collapses persistence below a direct forecaster. What survives is the timing of a departure from persistence, learnable on three of four corpora and near chance on the highest-base-rate corpus, a negative result we report. We recast the task as selective departure forecasting: a lead-time-weighted metric with a leakage protocol, and commitment policies whose best choice switches with the departure rate. A dialogue-level conformal risk control bounds the expected loss of committed predictions, meeting the target in all twelve configurations tested, though the guarantee is bought through abstention, with coverage and alarm recall falling sharply on the hardest corpus. One tested language model's confidence fails as a ranking signal where another succeeds, a model-dependent failure not yet attributable to family versus scale. Code and result logs are available from the corresponding author upon request.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05895

Information-Matched Auditing of Conversational Emotion-Shift Forecasting: A Leakage-Safe Evaluation Protocol

Bin Wen, Dai-Qiao Zhang, Tien-Ping Tan

原文摘要

Comparisons in conversational emotion-shift forecasting are easily distorted when a learned model receives rich features but its transition baseline is given a noisy or unavailable current-emotion label, when hyperparameter budgets differ, or when target-turn information enters the representation. We present an information-matched, leakage-safe auditing protocol that closes these gaps with an explicit commitment point, executable causality tests, equal four-configuration validation searches, and seed-dialogue-aware inference. Applying it to six causal model families on four text corpora and two feature-source sensitivity configurations shows that a deployable train-only current-emotion classifier feeding the transition baseline yields a different verdict than an oracle diagnostic where both sides receive gold current emotion. Under the deployable condition, learned models improve over the predicted-label baseline on all four primary corpora; DailyDialog gains are large and survive a 72-comparison Holm correction (best ROC-AUC gain = +0.167, adjusted p=.0144). Under matched gold information, the baseline remains competitive on IEMOCAP and stronger on MELD, while every model retains a large DailyDialog advantage. We additionally report precision-recall AUC, calibration, a strictly utterance-local feature control, same-person self-shift forecasting, and a future-access dose-response test. An eight-item leakage checklist accompanies the protocol. Code, raw seed predictions, and reporting artifacts are available from the corresponding author upon request.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05877

When Does Better Confidence Improve a Knowledge-Graph Audit?

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

原文摘要

Confidence estimates guide human review in LLM-built knowledge graphs, but better error prediction can yield little improvement in an audit. We study this gap through repairability, selection headroom and label validity. Modelling non-adaptive auditing as budgeted covering, we characterise the minimal decision-relevant error statistics and bound the regret of scalar-preserving first-order planning by a structural mass $Δ^{\mathrm{cl}}$. We simulate ideal repair against benchmark-aligned labels for five extraction configurations and two cross-document graphs on Re-DocRED, followed by frozen confirmation on test and replication on 500 hash-selected training documents. On the held-out cross-document graphs, an error-rate $\times$ fan-out queue leaves 3.98-4.74 percentage points more errors than a joint model with the same actions. By comparison, opening shared actions for that joint planner reduces residual errors by 0.89-1.43 points relative to its own fact-only plans, in point estimates. Equivalence between joint and independent-block models remains inconclusive. Generic LLM verification improves three of four model-budget configurations on each held-out set. LLM adjudication identifies possible benchmark omissions.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2605.09252

LLM Agents Already Know When to Call Tools -- Even Without Reasoning

Chung-En Sun, Linbo Liu, Ge Yan, Zimo Wang, Tsui-Wei Weng

原文摘要

Tool-augmented LLM agents tend to call tools indiscriminately, even when the model can answer directly. Each unnecessary call wastes API fees and latency, yet no existing benchmark systematically studies when a tool call is actually needed. We propose When2Tool, a benchmark of 18 environments (15 single-hop, 3 multi-hop) spanning three categories of tool necessity -- computational scale, knowledge boundaries, and execution reliability -- each with controlled difficulty levels that create a clear decision boundary between tool-necessary and tool-unnecessary tasks. We evaluate two families of training-free baselines: Prompt-only (varying the prompt to discourage unnecessary calls) and Reason-then-Act (requiring the model to reason about tool necessity before acting). Both provide limited control: Prompt-only suppresses necessary calls alongside unnecessary ones, and Reason-then-Act still incurs a disproportionate accuracy cost on hard tasks. To understand why these baselines fail, we probe the models' hidden states and find that tool necessity is linearly decodable from the pre-generation representation with AUROC 0.89--0.96 across six models, substantially exceeding the model's own verbalized reasoning. This reveals that models already know when tools are needed, but fail to act on this knowledge during generation. Building on this finding, we propose Probe&Prefill, which uses a lightweight linear probe to read the hidden-state signal and prefills the model's response with a steering sentence. Across all models tested, Probe&Prefill reduces tool calls by 38\% with only 1.7\% accuracy loss, while the best baseline at comparable accuracy only reduces 6% of tool calls, or achieves a similar tool call reduction but incurs a 5$\times$ higher accuracy loss. Our code is available at https://github.com/Trustworthy-ML-Lab/when2tool

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05862

CARTBench: Rubric-Based Evaluation of Shopping Advice for Complex and Ambiguous Requests with Timelines

Yuqin Dai, Zhanwei Zhang, Honghui Yang, Zhouheng Yao, Yueqi Zhang, Jingyu He, Chaoyang Shi, Fanding Wei 等 12 位作者

原文摘要

Large language model (LLM) shopping assistants now recommend products in multi-turn conversations. However, reliable evaluation faces Data scarcity and Difficulty Standardizing Criteria. To address these obstacles, we introduce CARTBench for CART (Complex and Ambiguous Requests with Timelines). For Data scarcity, we curate cases from real commercial shopping needs across 17 retail domains, pairing timestamped dialogue with 1,603 product records. For Difficulty Standardizing Criteria, we develop case-specific rubrics guided by Ideal Response Principles, distilled from excellent replies manually selected from successfully completed orders. These principles also guide Principle-Guided CART Prompting, our structured recommendation strategy. The experimental benchmark contains 1,328 expert-annotated criteria across Core, Excellence, and Safety. To answer the CART question, we evaluate 7 flagship commercial API models, including GPT-6.1-Sol and Claude-Opus-5, and 35B+ parameters open-weight models. In the first full-suite evaluation under the frozen rubrics, 3 of 10 models exceed 90% on Core, while none exceeds 45% on Excellence. Therefore, we identify an understanding-to-execution gap between required behavior and high-quality purchase advice. To examine these limitations, we diagnose shared failures in product evidence and unresolved needs, linked to changing requirements and decisive unknowns. Code and data are available at https://github.com/Da1yuqin/CARTBench.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2604.08567

Multi-User Large Language Model Agents

Shu Yang, Shenzhe Zhu, Hao Zhu, José Ramón Enríquez, Di Wang, Alex Pentland, Michiel A. Bakker, Jiaxin Pei

原文摘要

Large language models (LLMs) and LLM-based agents are increasingly deployed as assistants in planning and decision making, yet most existing systems are implicitly optimized for a single-principal interaction paradigm, in which the model is designed to satisfy the objectives of one dominant user whose instructions are treated as the sole source of authority and utility. However, as they are integrated into team workflows and organizational tools, they are increasingly required to serve multiple users simultaneously, each with distinct roles, preferences, and authority levels, leading to multi-user, multi-principal settings with unavoidable conflicts, information asymmetry, and privacy constraints. In this work, we present the first systematic study of multi-user LLM agents. We begin by formalizing multi-user interaction with LLM agents as a multi-principal decision problem, where a single agent must account for multiple users with potentially conflicting interests and associated challenges. We then introduce a unified multi-user interaction protocol and design three targeted stress-testing scenarios to evaluate current LLMs' capabilities in instruction following, privacy preservation, and coordination. Our results reveal systematic gaps: frontier LLMs frequently fail to maintain stable prioritization under conflicting user objectives, exhibit increasing privacy violations over multi-turn interactions, and suffer from efficiency bottlenecks when coordination requires iterative information gathering.

cs.CL · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2608.13168

Who Is Your Ideal Companion? Evaluating Emotional Companion Capabilities of LLMs Based on Adult Attachment Theory

Junkai Zhou, Shiting Guan, Zhaoyi Zhang

原文摘要

As large language models (LLMs) are increasingly applied for emotional companionship, evaluating their behavior and capabilities in intimate relationships has become a pressing issue. However, existing assessments primarily characterize general personality traits, providing limited insight into model behavior within intimate and emotionally sensitive contexts. Therefore, we introduce adult attachment theory into LLM evaluation and use the Experiences in Close Relationships-Revised (ECR-R) scale to characterize attachment anxiety and avoidance. To evaluate emotional companionship capabilities of LLMs in realistic interaction scenarios, we present an emotional companionship benchmark, ECBench, spanning four scenarios including emotional support, collaborative tasks, conflict resolution, and social guidance, across friendship and romantic relationships. ECBench is utilized to assess model behavior using 11 dialogue-quality metrics and three evaluation methods. We evaluate the attachment tendencies of 32 LLMs and select representative models to investigate how these tendencies manifest in contextualized multi-turn interactions and whether they can be shaped through prompting. Our study provides a theoretical lens from psychology, along with practical tools to understand and select LLMs for emotional companionship.

cs.CL

查看论文
arXiv ↗PDF ↗