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 21:43 · 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 22:09 — 2026-11-09 22:09（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 21:43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 2,348 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 21:40（北京时间）
新论文 · v1arXiv:2611.09785

Are These Programmers Ripe Enough to Eat? Evaluating How LLMs Handle Lexical Knowledge Conflicts in Translation

Dongxu Wu, Sara Court, Micha Elsner

原文摘要

External lexical resources can help large language models (LLMs) translate texts in low-resource languages and specialized domains, but these resources may conflict with models' prior lexical knowledge. To isolate this conflict, we construct a controlled Mandarin-to-English translation task in which familiar source words are assigned new meanings within and across semantic categories. We treat these prompt-supplied meanings as contextual knowledge (CK) and systematically evaluate the effects of word frequency, sentence context, and prompt framing on the translated outputs. Across nine models, CK reliance is greater for less frequent source words and in sentence contexts that support the supplied meanings. Prompt framing can reduce CK reliance, but seven of nine models favor CK even when explicitly instructed to ignore the supplied meanings. Within model families, larger models are more likely than smaller models to utilize CK in their outputs. Three follow-up experiments further show that modal weakening lowers CK reliance, the effect of repetition depends on prompt layout, and the effect of reasoning varies across models. Overall, our controlled study reveals how CK reliance varies across conditions and models, motivating new methods to detect and evaluate lexical conflicts and guide models' lexical choices in LLM-based translation.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.06919

Can LLMs Take Retrieved Information with a Grain of Salt?

Behzad Shayegh, Mohamed Osama Ahmed, Fred Tung, Leo Feng

原文摘要

Large language models have demonstrated impressive retrieval-augmented capabilities. However, a crucial area remains underexplored: their ability to appropriately adapt responses to the certainty of the retrieved information. It is a limitation with real consequences in high-stakes domains like medicine and finance. We evaluate eight LLMs on their context-certainty obedience, measuring how well they adjust responses to match expressed context certainty. Our analysis reveals systematic limitations: LLMs fail to reproduce their prior outputs after observing an uncertain context, respond to expressed certainties in a distorted way, and overrely on complex contexts. To address these, we propose an interaction strategy combining prior reminders (feeding the model's own context-free answer back into the prompt), context-certainty recalibration, and context simplification. This approach reduces obedience errors by 25% on average, without modifying model weights, demonstrating the efficacy of interaction design in enhancing LLM reliability. Our contributions include a principled evaluation metric, empirical insights into LLMs' uncertainty handling, and a portable strategy to improve context-certainty obedience across diverse LLMs.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09770

ROSE: Risk-Aware Orthogonal Subspace Navigation for Lifelong Knowledge Editing in Multimodal Large Language Models

Lingyun Song, Ziyao Chen, Kang Pan, Xiaolin Han, Yifei Zhang, Xiaoqi Wang, Yudai Pan, Xuequn Shang

原文摘要

Multimodal Large Language Models (MLLMs) require continuous factual updates to stay current, yet lifelong knowledge editing remains a daunting challenge. The primary obstacle is the severe Locality Erosion and Catastrophic Forgetting, where sequential edits inevitably interfere with pre-trained capabilities and prior updates. In this paper, we propose \textbf{ROSE}, a \textbf{R}isk-aware \textbf{O}rthogonal \textbf{S}ubspace navigation framework for lifelong knowledge \textbf{E}diting. ROSE addresses forgetting by aiming to restrict parameter updates to the orthogonal complement of previously edited subspaces. Crucially, it introduces a risk-guided mechanism that utilizes a dynamically computed risk mask to safeguard specific factual parameter subspaces essential for model stability. During inference, ROSE employs a multi-granularity knowledge integration strategy, featuring prototype-anchored semantic gating and perturbation-driven subspace fusion to precisely activate edited knowledge for relevant queries while limiting interference on unrelated inputs. Comprehensive experiments on two prominent multimodal benchmarks show that ROSE is competitive with state-of-the-art approaches across five key metrics, while maintaining outstanding stability over $1,000$ sequential updates. Our code is available at https://github.com/sunlitsong/ROSE.

cs.CL · cs.AI · cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2608.11624

Learning to Persuade Exposes How Easily LLMs Abandon Correct Beliefs

Nimet Beyza Bozdag, Emre Can Acikgoz, Gokhan Tur, Dilek Hakkani-Tür

原文摘要

Persuasion is a core dynamic of natural language communication, shaping how large language models (LLMs) update beliefs, resolve disagreements, and reach decisions. As LLMs increasingly debate, advise, and think collaboratively with humans and each other, resistance to harmful persuasion becomes a core requirement for reliable behavior. Yet a single targeted persuasive argument can be enough to collapse model accuracy to near zero, even when the argument is factually false. We formalize this threat as adversarial persuasion and introduce an adversarial reinforcement learning framework that trains persuader agents to change a target model's answer in a single interaction. First, we show that RL-trained persuaders raise persuasion success from approximately 24% to over 93% against the training-time persuadee. Second, we find that these learned strategies generalize across four out-of-distribution benchmarks, maintaining 89-95% persuasion success against the training-time persuadee. They also transfer to unseen models, achieving 83% success on Qwen-14B, 79% on Llama-3.1-8B, and 25% on GPT-4o-mini, although effectiveness can be limited for frontier models, with only 4% success on GPT-5-mini. Third, we demonstrate that a curriculum that bootstraps on more persuadable open-weight models before targeting harder models further increases GPT-4o-mini attack success from 25% to 38%. Moreover, our results reveal that optimized persuaders increasingly rely on credibility-based tactics, including fabricated citations and false authoritative evidence. Together, these findings expose a critical weakness in current LLM agents. Even when they initially reason correctly, they can be steered toward false conclusions by optimized natural language influence. This positions persuasion robustness as a necessary safety criterion for multi-agent and human-AI decision-making systems.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09768

Language as Trait, Politeness as Strategy: Controlled Tests of Linguistic Effects in LLM Social-Deduction Agents

Yuchen Li, Lisa Liu, Runqi Zou, Shuxuan Kuang, Qingquan Zhang, Julian Togelius

原文摘要

Large language model (LLM) agents coordinate through conversation, and the same message can be phrased in simpler or more complex language, bluntly or politely. We test whether such differences change how other agents judge and act, using the LURK Werewolf framework, which separates each utterance's speech-act predicate from its natural-language realization. We rewrite only the realization, check that the game content is preserved, and replay each game day from saved checkpoints against a re-sampled control. We vary linguistic range consistently across a speaker's messages and politeness strategy per utterance. Across 30 games and 4,335 replays, neither manipulation changes the trust other agents place in the speaker, and no effect on votes or night actions survives correction for multiple comparisons; the largest uncorrected effect is that shifting range away from a neutral paraphrase changes about 4 percentage points more of the other players' votes. In natural play, 88% of face-threatening speech is bald on-record regardless of role.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.26935

DunbaaBERT: From Sacrifice to Semantics

Iffat Maab, Waleed Jamil, Raphael Schmitt

原文摘要

Large language models have achieved strong performance across many NLP tasks, yet Urdu remains comparatively underexplored due to limited resources and fragmented evaluation settings. To address this gap, we introduce DunbaaBERT, a family of Urdu RoBERTa-base models trained from scratch with Byte-BPE vocabularies of 32k, 52k, and 96k tokens on a deduplicated 17 GB Urdu corpus. We evaluate DunbaaBERT across intrinsic and downstream Urdu NLP benchmarks covering linguistic acceptability, news classification, offensive language detection, and sentiment analysis, while analyzing vocabulary-size effects on performance, pre-training dynamics, and computational efficiency. Across benchmarks, the DunbaaBERT variants achieve competitive performance against strong multilingual baselines while consistently maintaining favorable efficiency trade-offs. Under our controlled experimental setup, increasing the vocabulary size does not consistently improve downstream effectiveness, with DunbaaBERT$_{32k}$ repeatedly providing the strongest overall efficiency profile. Overall, our results demonstrate that carefully curated Urdu-specific encoder models can remain highly competitive despite comparatively compact model and training scales. All models are released under the MIT license.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2602.02474

MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents

Haozhen Zhang, Quanyu Long, Jianzhu Bao, Tao Feng, Weizhi Zhang, Haodong Yue, Wenya Wang

原文摘要

Most Large Language Model (LLM) agent memory systems rely on a small set of static, hand-designed operations for extracting memory. These fixed procedures hard-code human priors about what to store and how to revise memory, making them rigid under diverse interaction patterns and inefficient on long histories. To this end, we present \textbf{MemSkill}, which reframes these operations as learnable and evolvable memory skills, structured and reusable routines for extracting, consolidating, and pruning information from interaction traces. Inspired by the design philosophy of agent skills, MemSkill employs a \emph{controller} that learns to select a small set of relevant skills, paired with an LLM-based \emph{executor} that produces skill-guided memories. Beyond learning skill selection, MemSkill introduces a \emph{designer} that periodically reviews hard cases where selected skills yield incorrect or incomplete memories, and evolves the skill set by proposing refinements and new skills. Together, MemSkill forms a closed-loop procedure that improves both the skill-selection policy and the skill set itself. Experiments on LoCoMo, LongMemEval, HotpotQA, and ALFWorld demonstrate that MemSkill improves task performance over strong baselines and generalizes well across settings. Further analyses shed light on how skills evolve, offering insights toward more adaptive, self-evolving memory management for LLM agents. Code is available at https://github.com/ViktorAxelsen/MemSkill

cs.CL · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2604.28031

Models Recall What They Violate: Constraint Adherence in Multi-Turn LLM Ideation

Garvin Kruthof

原文摘要

When researchers refine ideas with large language models over several turns, do the models preserve the original brief's constraints? We introduce DriftBench, a benchmark for constraint adherence in multi-turn LLM-assisted scientific ideation. It pairs 38 research briefs from 24 scientific domains, each with hard constraints and banned moves, with four interaction conditions and seven models from five providers (two open-weight), scored on 2,116 runs. Iterative pressure to improve a proposal raises structural complexity in every model and lowers adherence to the brief's hard constraints. A restatement probe shows that the constraints stay legible and reproducible on demand while the proposal departs from them. The knows-but-violates (KBV) rate, the share of pressure runs scored non-compliant on a hard constraint despite a correct restatement, exceeds 50% for most models. A crossed-judge design measures the judge offset. GPT-5.4 scores adherence 0.535 points lower than Opus 4.6 on the same runs. The design finds no judge-by-condition interaction on adherence, so the pressure effect on adherence is judge-stable while KBV levels and the bottom of the model ranking are not. Re-eliciting the brief at checkpoints lowers KBV for some models but not others, and complexity inflation persists. Blind human raters find more constraint violations than the LLM judge, so reported adherence is conservative. We release all briefs, prompts, rubrics, transcripts, and scores.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2603.04996

HiFlow: Hierarchical Feedback-Driven Optimization for Constrained Long-Form Text Generation

Yifan Zhu, Guanting Chen, Bing Wei, Haoran Luo

原文摘要

Large Language Models face a multi-objective optimization problem in constrained long-text generation, balancing global structure, local coherence, and constraint satisfaction. We propose LoopWeaver, a hierarchical workflow with two sequential offline preference-optimization stages. Constraint-aware planning produces structured plans, feasibility filtering removes invalid plans, and a shared deterministic reward constructs preference pairs for plan-level and generation-level DPO. At inference time, LoopWeaver remains feed-forward: it generates and revises a complete plan, filters the revised plan, and then generates segments conditioned on the current sub-plan and running narrative. We evaluate the method on four LongGenBench-style structured scenarios across multiple backbones. Results show improvements in both constraint accuracy and text quality across all four open-source backbones. Our software and data are publicly available at https://github.com/cgtaa-cmyk/LoopWeaver.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2501.08686

Knowledge Graph-based Retrieval-Augmented Generation for Schema Matching

Chuangtao Ma, Sriom Chakrabarti, Arijit Khan, Bálint Molnár

原文摘要

Traditional similarity-based schema matching methods fail to resolve semantic ambiguities and conflicts in complex, domain-specific mappings due to a lack of commonsense and domain-specific knowledge. In parallel, large language models (LLMs) without sufficient background knowledge tend to hallucinate, further complicating schema matching. To overcome these limitations, we introduce KG-RAG4SM, a Knowledge Graph-based Retrieval-Augmented Generation for Schema Matching. This method integrates novel retrieval mechanisms, including LLM-based, vector-based, graph traversal-based, and hybrid techniques with advanced reranking schemes to extract the most relevant subgraphs from large external KGs. Our approach achieves superior results without requiring re-training, as evidenced by improvements of 39.79% in precision and 30.56% in F1 score over LLM-based state-of-the-art (SOTA) methods (e.g., Jellyfish-8B) on the MIMIC dataset, and enhancements of 69.20% in precision and 21.99% in F1 score over pre-trained language model (PLM)-based SOTA methods (e.g., SMAT) on the Synthea dataset. Additionally, KG-RAG4SM scales effectively across KGs from commonsense KGs to domain-specific ones, while its instruction-tuning knowledge exclusion component ensures truthful performance in end-to-end schema matching tasks despite the suboptimal relevance of KG context. Our empirical analysis and case studies from real-world scenarios show that the hallucination problem in LLMs for schema matching is well mitigated by our solution.

cs.DB · cs.CL · cs.IR

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.07062

Learning to Simulate Individuals from Macro Social Signals

Yining Zhao, Bushi Liu, Haofei Yu, Zhengyang Qi, Shanyong Wang, Chuyue Li, Yuxiang Liu, Jiaxuan You

原文摘要

Large language models are increasingly used to simulate how individuals respond to new situations, yet the behavioral reasoning behind these responses is either inherited from pretraining or learned from individual-level annotations, which offer limited behavioral diversity and little supervision of the reasoning itself. We propose to learn behavioral reasoning from prediction markets, whose price trajectories record how populations respond to real-world events at scale. We introduce macro2mind, which trains a language model with GRPO using market signals. A social behavioral decomposition makes behavioral reasoning an explicit step of forecasting: the model infers representative groups of market participants, predicts how each interprets the news and updates its beliefs, reasons about their interactions, and aggregates these responses into a price. A hindsight-regret curriculum with difficulty-aware sampling focuses training on transitions where hindsight-identified groups substantially improve the forecast while prioritizing examples that remain learnable for the current policy. The learned reasoning applies to user simulation without further training. On SWM-Bench, macro2mind achieves state-of-the-art directional accuracy and correlation on Polymarket. Trained on market data, it transfers zero-shot to four user-simulation benchmarks (Humanual, OvertonBench, PRISM, and CAD) and has competitive performance among zero-shot methods. Used as a data generator, macro2mind also raises a downstream simulator's accuracy on unseen users by 15.5 points, outperforming data generated by its backbone by 13.2 points.

cs.LG · cs.AI · cs.CL · cs.SI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.25596

Cognitive Maps in Large Language Models

Albert Qin, Jonathan D. Cohen, Gautam Reddy

原文摘要

It has long been proposed that animals construct cognitive maps to support flexible behavior in novel environments. We tested whether large language models form analogous internal representations using relational reasoning tasks and causal "mosaic" experiments. Models beyond the ten-billion-parameter scale inferred unstated relationships and generalized to synthetic entity names. Across models and tasks, models formed abstract representations of relational structure, additively combining them with representations of entity identity. Relationships were primarily stored as pairwise links and composed retrospectively, rather than prospectively encoded in a global coordinate system. A sparse, distributed set of key-value pairs mediated binding; perturbing them implanted "false memories" that produced predictable behavioral changes. Our findings reveal a conserved organization analogous to a cognitive map supporting flexible and systematic reasoning in language models.

cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09506

On the Effectiveness and Limitations of Open-Weight LLMs for Low-Resource Offensive Language Detection in Danish

Elnur Adl Zarabi, Amit Kumar, Adi Singh, Sierra Nelson, Megan Stubbs-Richardson, Taoran Ji

原文摘要

We study the effectiveness and robustness of open-weight large language models (LLMs) for Danish offensive language detection, an underexplored low-resource task. We evaluate four model families (Llama, Mistral, Gemma, and Qwen) across eleven prompting strategies spanning five design categories, cross-prompt generalization, Danish orthographic variation, and robustness under balanced and naturally imbalanced class distributions. We compare against three BERT-family baselines with distinct pre-training strategies, evaluated via both masked language model prompting and supervised fine-tuning. Even without task-specific fine-tuning, LLMs can surpass all three fine-tuned BERT baselines, and lightweight finetuning yields substantial further gains. No single model or prompt category dominates, underscoring that prompt design interacts with model pre-training in ways that cannot be predicted a priori. Semantically grounded prompts transfer more robustly, while format-constrained patterns consistently fail to generalize. ASCII transliteration of Danish-specific characters (æ, ø, å) generally degrades performance, indicating these characters encode meaningful signal. At the same time, the gap between a balanced test set (best F1 of 0.84) and the corpus's natural class skew (0.62 trained on skewed data, 0.70 trained on balanced data) -- relative drops of 26.2\% and 16.7\% -- shows that class imbalance remains a critical open challenge. Our code and configurations are available at https://github.com/PASHAADC/Danish-LLM-HateSpeech.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09491

An Effective Spoken Language Understanding Approach Leveraging Query-Adaptive Intent Detection

Bing-Long Kuo, Yu-Hsuan Fang, Tien-Hong Lo, Hsueh-Wei Liao, Berlin Chen

原文摘要

Large language models enable fine-tuning-free spoken language understanding (SLU) through in-context learning, but direct query-based retrieval may select semantically similar yet intent irrelevant demonstrations. This paper presents Query-Adaptive Intent Detection via Prototype Extraction (APEX), a two-stage retrieval framework that first selects candidate intents using intent-level prototypes and then retrieves contextual exemplars within the selected intent space. Experiments on the SLURP dataset evaluate joint SLU performance using Exact Match Accuracy (EMA) under full-transcript and perplexity-stratified conditions, with additional retrieval analysis on ASR-generated transcripts. Compared with query-based retrieval, APEX improves EMA by 4.34 and 3.97 percentage points with Qwen3 4B and Gemma-3-4B, respectively, on full transcripts. Under high-perplexity inputs, APEX also yields EMA gains of 7.66 and 2.13 percentage points. Retrieval analysis further indicates that prototype-guided candidate selection achieves higher ground truth intent coverage across multiple input conditions.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.07234

Reformulating KV Cache Eviction Problem for Long-Context LLM Inference

Tho Mai, Joo-Young Kim

原文摘要

Large language models (LLMs) support long-context inference but suffer from substantial memory and runtime overhead due to Key-Value (KV) Cache growth. Existing KV Cache eviction methods primarily rely on local attention weights, neglecting the influence of value representations, output projection, and inter-head interactions. In this work, we reformulate KV Cache eviction from a conventional head-wise, weight-averaging approach into an output-aware, layer-wise matrix multiplication approximation problem. We introduce LaProx, a novel eviction strategy that explicitly models the multiplicative interaction between attention maps and projected value states to accurately quantify token contributions while accounting for inter-head dependencies. Building on this metric, we propose the first unified eviction strategy that assigns globally comparable importance scores to tokens, enabling model-wide selection instead of local, head-wise decisions. Experimental results across 19 datasets on long-context benchmarks LongBench and Needle-In-A-Haystack demonstrate that our approach maintains model performance with only 5% of the KV cache and consistently outperforms prior works across all configurations. Notably, our method achieves up to 2$\times$ accuracy loss reduction under extreme compression scenarios compared to existing state-of-the-art baselines with minimal overhead. Our code is available at https://github.com/huuthomai/LaProx.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09372

Enabling Privacy-Preserving Natural Language Access to Industrial Smart Meter Data

Carlos Martín-Abellán, Mark Kröll

原文摘要

Energy management officers and grid operators increasingly rely on smart meter data for anomaly detection, performance monitoring, and strategic decision-making. However, extracting actionable insights from large-scale, heterogeneous telemetry databases requires advanced SQL and data engineering expertise, creating a critical bottleneck for non-technical personnel. Furthermore, the strict data privacy regulations and security constraints inherent to critical energy infrastructure often prohibit the routing of sensitive consumer data through cloud-based Large Language Models (LLMs). To address these challenges, we present OpenMeter, a schema-agnostic Multi-Agent System (MAS) designed to translate natural language queries into executable code and human-readable insights. This architecture lowers the technical barrier for day-to-day utility operations by empowering non-expert personnel to interact with complex databases through intuitive natural language dialogue, transforming raw telemetry into a highly valuable tool for routine grid workflows and strategic decisions. Multi-scenario ablation studies over this environment demonstrate that full multi-agent orchestration shows its largest observed accuracy gains on resource-constrained local hardware (+6.1 pp EX-R over the strongest single-agent configuration); at frontier scale, individual components retain value but should be enabled selectively, indicating that the cost of full orchestration is justified specifically where privacy demands local execution.

cs.AI · cs.CL · cs.DB · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2604.17139

The Consensus Trap: Collaborative Interleaved Generation to Defend against a Corrupted Majority

Jiayuan Liu, Shiyi Du, Weihua Du, Xinhao Cheng, Mingyu Guo, Vincent Conitzer

原文摘要

Majority voting offers a simple way to combine answers from multiple large language model (LLM) agents. When a majority follows instructions favoring the same incorrect answer, however, agreement reinforces the error. We call this failure the "consensus trap": a minority may identify the error, yet its reasoning cannot change a vote over completed answers. We introduce chunk-interleaved Round-Robin (RR) collaboration, in which agents take turns extending one unfinished reasoning trajectory. A minority of agents with correct reasoning can insert corrections that subsequent speakers read before the answer is complete. RRMaj combines several such trajectories by a final vote. We characterize a majority barrier for outcome-level voting over final answer labels and derive a local sufficient condition for expected progress when the minority's cumulative corrections outweigh opposing corrupted updates. Across model families and verifiable reasoning benchmarks, RRMaj substantially improves accuracy under majority corruption while retaining comparable average performance under minority corruption in the main advisory-instruction setting. RR also strengthens full-response debate. Scoring the shared reasoning between turns shows that, on average, continuations from uncorrupted agents increase support for the correct answer over the incorrect target. Parallel local serving adds modest latency in the measured deployment. These findings show how language shared during reasoning, and the timing of that exchange, can help a minority correct a group's answer.

cs.CL · cs.AI · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2608.30646

BiG-SURE - Bipartite Graph for Semantic Uncertainty and Reliability Estimation of LLMs

Debarpan Bhattacharya, Malay Phadke, Sriram Ganapathy

原文摘要

Reliable uncertainty estimation is a crucial requirement for deploying large language models (LLMs) and vision-language models (VLMs) in safety-critical settings, especially when the model parameters are not accessible (black-box). We propose BiG-SURE, an uncertainty estimator based on cross-temperature semantic agreement. The method samples low-temperature responses as stable semantic anchors and high-temperature responses as probes under meaning-preserving input transformations. It then constructs an anchor-probe Bipartite Graph (BiG) using NLI-based entailment scores and defines confidence through the normalized squared spectral energy of this matrix, with uncertainty given by its complement. This bipartite graph-based Semantic Uncertainty and Reliability Estimation (SURE) score measures whether high-temperature probes remain semantically aligned with the model's stable low-temperature belief or not. We evaluate BiG-SURE on text QA, multilingual QA, and multimodal QA tasks across multiple model families. In these experiments, BiG-SURE improves average abstention AUROC over prior black-box uncertainty estimators, while remaining simple, unsupervised, and applicable to black-box model settings.

cs.CL · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2510.18413

Adamas: Hadamard Sparse Attention for Efficient Long-Context Inference

Siyuan Yan, Guo-Qing Jiang, Yuchen Zhang, Xiaoxing Ma, Ran Zhu, Chun Cao, Jingwei Xu

原文摘要

Large language models (LLMs) now support context windows of hundreds of thousands to millions of tokens, enabling applications such as long-document summarization, large-scale code synthesis, multi-document question answering and persistent multi-turn dialogue. However, such extended contexts exacerbate the quadratic cost of self-attention, leading to severe latency in autoregressive decoding. Existing sparse attention methods reduce these costs by attending to only a subset of key-value (KV) tokens, but fixed or coarse-grained selection strategies can miss important tokens or retain substantial redundancy under constrained token budgets. We introduce \textbf{Adamas}, an accurate and efficient token-level sparse attention mechanism for long-context inference. Adamas applies the Hadamard transform, bucketization and 2-bit compression to produce compact representations, and uses lightweight distance-based estimation for efficient top-$k$ candidate selection. The selected positions are then used to gather the corresponding original KV states for sparse attention. Comprehensive evaluations show that Adamas maintains near-lossless performance within the effective context length of Llama-3.1-8B-Instruct. In efficiency evaluations, Adamas achieves up to $6.5\times$ self-attention speedup and $1.7\times$ end-to-end decoding speedup at a 32K context length. Adamas also maintains performance comparable to full attention on challenging long-form reasoning benchmarks, including AIME 2024 and AIME 2025.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09223

KLEAR: Knowledge-Matched LLM Evaluation in Korean across Application Requirements via Controlled Benchmark Transformation

Seoyeon Kim, Jeonghyun Kang, Hamin Koo, Barom Kang, Heondeuk Lee, Jaehyung Kim

原文摘要

Existing Korean benchmarks for Large Language Models (LLMs) evaluate performance across diverse knowledge domains, but their accuracy scores cannot distinguish whether incorrect responses arise from a lack of required knowledge or a failure to apply that knowledge correctly. Our analysis of existing Korean benchmarks reveals that most questions can be answered through factual recall alone, while those demanding greater application also differ in required knowledge, making it difficult to isolate application ability. To address this, we introduce KLEAR, a framework that constructs knowledge-matched contrasts by varying application demand while preserving the required knowledge. During construction, KLEAR identifies the knowledge underlying each source question as an explicit anchor bundle to guide generation of questions requiring more complex application. Generated candidates then undergo automated and human validation, with iterative revision based on evaluation feedback. Experiments demonstrate that the resulting knowledge-matched sets largely satisfy the intended constraints, enabling controlled evaluation across diverse LLMs. These evaluations reveal that increasing application demand exposes failures that persist even when the required knowledge is provided.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09222

Khmer PFGen-Bench: Building a Judge-Free Benchmark for Open-Ended Generation in a Low-Resource Language

Mengsay Loem, Kentaro Imajo, Ryota Ozaki, Masanori Hirano

原文摘要

Benchmarks for open-ended generation often rely on large language model (LLM) judges, whose reliability in low-resource languages is uncertain. We introduce Khmer PFGen-Bench, to our knowledge the first judge-free benchmark for multi-sentence answers to questions composed directly in Khmer, a low-resource language. Khmer PFGen-Bench adapts the scorer of PFGen-Bench, a benchmark for Japanese that scores responses with keyword rules and n-gram statistics from LLM-generated references. This scoring requires references that are both valid and varied, which LLMs may not produce in Khmer, so we examine how to build such references. We find that post-trained models can give more valid but less varied references than their base variants, and that mixing model families increases variety for the same number of references. Using references from post-trained models in nine families, the benchmark yields model rankings that are stable across reference sets and consistent with a native speaker's ratings. These findings offer guidance for building judge-free benchmarks in other low-resource languages.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09094

AlienBench: A Benchmark of Structural Novelty in Language Model Outputs

Jonas Oppenlaender, Johanna M. Silvennoinen

原文摘要

Large language models (LLMs) are increasingly used for creative tasks such as writing, design, and idea generation. For these uses, it matters whether a model's ideas depart from familiar examples or stay close to them. Novelty is one standard measure of creativity, and current creativity evaluations ask human experts or LLM judges to rate how novel an output is. However, these ratings have no fixed reference for what counts as novel (or, in other terms, departure from the familiar). Measuring this departure requires a task with known familiar defaults. The structured imagination task from cognitive psychology is such a task. In this creativity test, people imagine a creature from another planet. The departure from features of Earth animals, such as bilateral symmetry and legs as mode of locomotion, is used as measure of creativity. We introduce AlienBench, a benchmark that adapts the structured imagination task to language models. Models describe alien creatures, and a panel of three LLM judges extracts ten biological features from the description and labels them as either Earth-typical or a departure. The sum of feature departures is the score of a description, from 0 to 10. Each model describes alien creatures under six prompt conditions that vary the framing of the task and the planetary environment. Each model generates 30 descriptions per condition (180 in total), and its overall score is the mean across these descriptions. We evaluate 10 language models, from frontier to open-weight models. Overall scores range from 3.28 (GPT-6 Luna) to 6.55 (Ministral 3 14B). Interestingly, open-weight models score higher than proprietary models in this task (5.40 vs. 4.58). Under the baseline prompt, the models depart from Earth defaults more often on average than 100 human participants on every feature. We release all prompts, generations, judge outputs, and scores.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09075

Large Language Model Deployment on Resource-Constrained Edge Devices: A Practitioners Survey

Rahul Maliakkal, Yashasvi Makin, Plawan Rath, Rohit Jain, Ankur Sadhoo

原文摘要

Large Language Models (LLMs) have transformed many areas of artificial intelligence (AI), but their substantial resource requirements limit their deployment on mobile and edge devices. Although LLMs have demonstrated impressive capabilities in various applications, the next challenge is deploying these models on resource-limited edge devices. Recent innovations in software optimizations and specialized hardware have made it feasible to deploy LLMs on edge devices. Edge deployment offers benefits including local data processing, cost savings, reduced latency, and personalized responses. This paper reviews the latest techniques that make the deployment of LLMs on edge devices feasible.

cs.LG · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.02540

PixelBench: Can LLMs Understand and Generate Images Pixel by Pixel?

Haoquan Zhang, Feiyang Liu, Juanxi Tian, Wanhua Li, Haodong Duan, Weiyang Liu

原文摘要

Large language models (LLMs) can generate visual assets through text-based formats such as SVG and CAD, but these symbolic representations describe images using high-level primitives rather than explicit raster values. We ask whether LLMs can instead generate pixel images directly through their existing vocabulary, without a learned image tokenizer or decoder. We introduce PixelBench, a benchmark for text-to-pixel (T2P) and image-to-pixel (I2P) generation across multiple resolutions, together with a complementary pixel-understanding evaluation. Models generate an RGB color palette followed by a grid of indices specifying the palette color of each output pixel. In T2P, the strongest models produce recognizable images directly from text. In I2P, the strongest models outperform palette-matched downsampling and specialized pixelization algorithms at low resolutions. Pixel-understanding performance correlates strongly with generation performance. A resolution-scaling study shows that pixel-generation performance often improves initially before declining at high resolutions. Finally, we introduce Pixel Reconstruction Tuning (PRT), which augments visual instruction tuning with caption-palette-grid trajectories. We post-train Molmo2-4B from its pretraining checkpoint by mixing general instruction data with our curated PRT data, improving performance across 16 visual-centric benchmarks, indicating that explicit pixel reconstruction can provide useful auxiliary supervision.

cs.CV · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.38402

LLMs as Judges and Examinees for Medical Item Difficulty Prediction

Xiaomeng Xiong, Marcus Walker, Yanlin Jiang

原文摘要

Predicting medical item difficulty from item text alone remains challenging. Prior work has largely relied on linguistic features or text representations derived from general-purpose and domain-specific language models, but predictive performance has remained modest. This study takes a different approach by using difficulty-related signals that emerge when open-weight large language models (LLMs) interact with test items. Using USMLE and PANCE medical licensure examination items, we examined LLMs in two roles. As judges, the models provided absolute difficulty ratings and pairwise difficulty comparisons. As simulated examinees, they generated response-based signals including correctness, keyed-option probability, response entropy, and the probability margin between the top two options. We then combined the strongest judgment and response signals with baseline item features in downstream prediction models. Pairwise difficulty judgment was the most consistent individual LLM signal across both datasets and outperformed absolute ratings, and adding simulated-examinee signals yielded only small further gains. The best configurations reached a Spearman correlation of $ρ$ = .447 on USMLE, exceeding the best system in the 2024 BEA shared task ($ρ$ = .25), and $ρ$ = .494 on PANCE. These results were comparable to direct absolute judgments from the proprietary frontier model Claude Opus 5 ($ρ$ = .443 and .488, respectively). Overall, judgment and response signals from open-weight LLMs captured difficulty-related information beyond text representations. Because these models can be deployed locally, the approach offers a practical option for predicting the difficulty of unreleased or high-security test items.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09061

Does Reasoning Improve Psychological Depth in Large Language Models?

Ruichen Zheng, Yihe Wang, Fabrice Harel-Canada, Sara Khosravi, Zeynep Senahan Yildiz, Amit Sahai, Nanyun Peng

原文摘要

Benchmark gains from reasoning-enhanced large language models encourage the assumption that these models should broadly perform better on complex generative tasks, but this remains underexplored in open-ended, creative domains shaped by nuances of reader experience. We tested this assumption through the lens of psychological depth in short stories. We compared DeepSeek-R1 vs. DeepSeek-V3, our closest available contrast for reasoning-oriented post-training, and GPT-5 vs. GPT-4o, through 60 blinded pairwise evaluations in which seven human annotators and an ensemble LLM-as-a-Judge rated prompt-matched story outputs. Human evaluations showed no universal reasoning advantage: GPT-5 was slightly preferred over GPT-4o (60.0-62.9% across two reasoning-effort settings), whereas DeepSeek-R1 fell behind DeepSeek-V3 (42.9%). Although this pattern was marked by low inter-annotator agreement (Krippendorff's $α= 0.070$), our annotation-quality checks suggested genuine evaluative subjectivity, plausibly reflecting distinct annotator decision schemes rather than annotation failure. In contrast, the LLM-as-a-Judge showed a strong pro-reasoning bias (98.3% of pairs vs. 55.2% of human votes), consistent across its five configurations and partly tracking sentence length. These findings suggest that, at the frontier, reasoning-related quality differences are small enough to be dominated by reader subjectivity, and that LLM-as-a-Judge evaluators can obscure this subjectivity in nuanced creative-generation tasks.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08995

Same Facts, Different Remedy: Linguistic Profiling and Arbitrariness in LLM and System One Decisions

Colin Priest

原文摘要

Large language models are increasingly used to resolve consumer complaints, and the way complainants write varies with age, education and socioeconomic status (SES). We audit three remedy-decision systems using sociolinguistic rewrites of US consumer financial complaints that are instructed to preserve every fact: two generative LLMs that return a single label (GPT-4o-mini and GPT-5-mini) and a probabilistic "System One" model that returns a distribution over remedies (Jev). Using pre-registered hypotheses tested once on held-out complaints, we separate two phenomena that single-label audits conflate. At the group scale, older, college-educated and higher-SES styles receive slightly more generous remedies; these effects replicate but are small, with every monetary-remedy gap within 0.5 percentage points. At the individual scale, two style versions of the same complaint receive different monetary decisions 2.6-6.3% of the time. In a variance-component analysis of the probabilistic model, persona components shared across the two rewrite models account for about 1.6% of within-complaint variation. The persona gaps are associated mainly with delivery (vocabulary, grammar, politeness) rather than content. Positive controls show that all three systems respond 49-387 times more strongly to remedy-relevant facts than to style. Rewriting every complaint into one standard style reduced GPT-4o-mini's education and SES gaps by 94% and 87% while retaining fact sensitivity, but reversed two of the probabilistic model's gaps and did not make individual decisions more stable. Independent LLM auditors flagged material fact changes in 3.3% and 11.7% of rewrites, and instability persisted among unflagged rewrites. Fairness audits of LLM decision systems should report group disparities and individual instability as separate outcomes.

cs.CY · cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.08975

Beyond Truncation? Entailment-Licensed Elision and the Limits of Semantic Preservation in LLM Decoding

Mahmoud Alyosify, Mirna Imbabi, Eman Elkhamisi

原文摘要

Instruction-tuned large language models (LLMs) answer at greater length than a request needs. We ask whether a training-free controller inside the decoding loop can remove that surplus while preserving content, judged against the simplest possible baseline: truncating the model's own answer to the same length. An entropy-gated logits processor removes 25.5-34.9% of tokens on MT-Bench development questions and 19.9-34.2% on a sealed test split within pre-registered BERTScore non-inferiority margins, yet we prove and confirm by audit that its output is always a prefix of the unmodified answer: the saving is truncation. A survey of 23 training-free mechanism families finds none that selects content better than truncation without oracle labels. We therefore introduce Entailment-Licensed In-Decoding Elision (ELIDE), which, at each sentence boundary, withholds a span from the reader only when a natural-language inference model judges it entailed by the text already shown, while leaving the model's generation unchanged. ELIDE beats matched-length truncation on three model families, but reading 936 of its elisions on 670 public prompts shows that 59% harm the answer, a failure that no automatic measure detects. Scoping the licence to its discourse unit reduces the harmful share to about one elision in ten on two fresh tests.

cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.30844

Fine-Tuned Language Models Exhibit Improved Information Conveyance

Yuwei Cheng, Weiyi Tian, Haifeng Xu

原文摘要

Fine-tuning is often believed to reduce uncertainty and diversity in large language models, but existing analyses overlook output length, a key confounder, and therefore fail to capture how uncertainty is distributed across an entire generation rollout. To address this, we propose Canopy Entropy ($\mathrm{CE}^\star$), a measure that views language generation from a tree perspective, where "canopy" represents the space of all possible rollouts, making $\mathrm{CE}^\star$ naturally quantify the effective size of the generation space. $\mathrm{CE}^\star$ jointly captures uncertainty in both the output length $N$ and the generated sequence $Y_{1:N}$ -- indeed, we show that it equals total Shannon entropy $H(N, Y_{1:N}\mid X)$, where $X$ denotes the prompt. This formulation yields interpretable metrics, including a length-entropy correlation term $ρ(N, r_N)$, where $r_N$ is the entropy rate, quantifying information conveyance efficiency by indicating whether longer outputs are more or less informative per token. Empirically, across tasks and model families, we find that fine-tuned models predominantly exhibit more positive correlation $ρ(N, r_N)$, even when total entropy decreases. Furthermore, after controlling for model family, task, prompt, and output-length effects, we find that fine-tuning nearly triples the correlation strength between entropy rate and semantic diversity, suggesting that aligned models convert token uncertainty into semantic diversity more efficiently. Overall, these results demonstrate that fine-tuning does not simply reduce uncertainty, but fundamentally reorganizes it into more informative and semantically meaningful generations. Our code is available at https://github.com/WeiyiTian/canopy-entropy.

cs.CL · cs.AI · stat.ML

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2607.05552

Decomposing Yes-No Bias in LLM Moral Judgment: Stance, Position, Word, and Verdict

Haonan Huang

原文摘要

Large language models (LLMs) increasingly issue binary verdicts - safety gates, judge decisions, value probes - and the answer is read as the model's judgment. A settled judgment should be coherent across paired phrasings of the same question: yes to "do you approve?" should mirror no to "do you oppose?". LLMs often violate this coherence, and in a yes/no question the verdict, the answer word, and the printed position coincide, so a conventional flipped-question audit yields one contrast that cannot say what moved. We introduce a crossed psychometric battery that extends the flip to full 2x2 and 2x2x2 crossings of question polarity, printed order, and answer label, separating stance from position, word, label, and verdict-role effects; an independent graded instrument estimates each model's stance $θ$ with no binary format; and a two-parameter fit (framing susceptibility $m$, decisiveness $s$) separates robustness from saturation that hides susceptibility. Applying the battery to twenty moral dilemmas - 1.35 million trials over ten frontier models from the Claude, GPT, and Gemini families in 23 reasoning configurations - we find response-form bias widespread but never uniform. The largest apparent yes-no biases are almost purely surface (word and position), while several of the newest models show the reverse: near-zero flip contrasts concealing verdict attachment up to +0.42, a component no surface counterbalance can expose, replicated across label alphabets. Requested reasoning reshapes these channels rather than removing them. When subjects state decisions in free text and a separate model transcribes them, two judge families agree on 99.3% of jointly scored transcriptions with zero opposing committed labels, and the transcribed stance converges with the graded ruler (median $r=0.87$): response form is part of the instrument, and the judgment can be measured apart from it.

cs.CL · cs.AI · cs.CY

查看论文
arXiv ↗PDF ↗