Named after the hundred-eyed watchman of Greek myth, Argus watches the education landscape: spotting new opportunities, pressure-testing the ventures we're building, and tracing every read back to the real-world signals behind it.
The evidence library: the raw signals the pipeline is watching across the education ecosystem. Every idea is built from these.
arXiv:2603.19539v2 Announce Type: replace Abstract: We introduce an expert curated, real-world benchmark for evaluating document-grounded question-answering (QA) motivated by generic drug assessment, using the U.S. Food and Drug Administration (FDA) drug label documents. Drug labels contain rich but heterogeneous clinical and regulatory information, making accurate question answering difficult for current language models. In collaboration with FDA regulatory assessors, we introduce FDARxBench, and construct a multi-stage pipeline for generating high-quality, expert curated, QA examples spanning factual, multi-hop, and refusal tasks, and design evaluation protocols to assess both open-book and closed-book reasoning. Experiments across proprietary and open-weight models reveal substantial gaps in factual grounding, long-context retrieval, and safe refusal behavior. While motivated by FDA generic drug assessment needs, this benchmark also provides a substantial foundation for challenging
arXiv:2603.18593v2 Announce Type: replace Abstract: We propose a method that represents language models by log-likelihood vectors over prompt-response pairs and constructs model maps for comparing their conditional distributions. In this space, squared Euclidean distances between models are approximately proportional to the KL divergence between the corresponding conditional distributions. Experiments on a large collection of publicly available language models show that the maps capture meaningful global structure, including relationships to model attributes and task performance. The representation also captures systematic shifts induced by prompt modifications and their approximate additive compositionality; we use the vectors to predict downstream task scores and leverage their additive structure to approximate the effects of composite prompt operations without directly observing the corresponding log-likelihood vectors. We further introduce PMI vectors to reduce the influence of unc
arXiv:2603.18579v2 Announce Type: replace Abstract: Evaluating whether explanations faithfully reflect a model's reasoning remains an open problem. Existing benchmarks use single interventions without statistical testing, making it impossible to distinguish genuine faithfulness from chance-level performance. We show that faithfulness is not a fixed property but an operator-dependent quantity that changes with the intervention method used to measure it. We introduce ICE (Intervention-Consistent Explanation), a framework that evaluates explanations against random baselines of equal size under multiple operators. Evaluating 7 LLMs across 4 tasks with deletion and retrieval infill operators, we find that switching operators crosses the positive-evidence threshold in 18% of configurations (5 of 28 attention comparisons), with gaps reaching 44 percentage points. Randomized baselines detect anti-faithfulness (explanations worse than random) in nearly one-third of English deletion configuratio
arXiv:2603.16848v2 Announce Type: replace Abstract: The ``LLM-as-a-judge'' paradigm has become a standard method for evaluating open-ended generation. To address the quadratic scalability costs of pairwise comparisons, popular benchmarks like Arena-Hard and AlpacaEval compare all models against a single anchor. However, despite its widespread use, the impact of anchor selection on the reliability of the results remains largely unexplored. In this work, we systematically investigate the effect of anchor selection by evaluating 22 different anchors on the Arena-Hard-v2.0 dataset. We find that the choice of anchor is critical: a poor anchor can dramatically reduce correlation with human rankings. We identify that common anchor choices (best-performing and worst-performing models) make poor anchors. Because these extreme anchors are consistently better or worse than all other models, they are seldom indicative of the relative ranking of the models. We further quantify the effect size of an
arXiv:2603.16749v3 Announce Type: replace Abstract: Large language models (LLMs) are increasingly deployed in applications with societal impact, raising concerns about the cultural biases they encode. We probe these representations by evaluating whether LLMs can perform author profiling from song lyrics in a zero-shot setting, inferring singers' gender and ethnicity without task-specific fine-tuning. Across several open-source models evaluated on more than 10,000 lyrics, we find that LLMs achieve non-trivial profiling performance but demonstrate systematic cultural alignment: most models default toward North American ethnicity, while DeepSeek-1.5B aligns more strongly with Asian ethnicity. This finding emerges from both the models' prediction distributions and an analysis of their generated rationales. To quantify these disparities, we introduce two fairness metrics, Modality Accuracy Divergence (MAD) and Recall Divergence (RD), and show that Ministral-8B displays the strongest ethnici
arXiv:2603.12275v2 Announce Type: replace Abstract: Unlearning knowledge is a pressing and challenging task in Large Language Models (LLMs) because of their unprecedented capability to memorize and digest training data at scale, raising more significant issues regarding safety, privacy, and intellectual property. However, existing works, including parameter editing, fine-tuning, and distillation-based methods, are all focused on flat sentence-level data but overlook the relational, multi-hop, and reasoned knowledge in naturally structured data. In response to this gap, this paper introduces Graph Oblivion and Node Erasure (GONE), a benchmark for evaluating knowledge unlearning over structured knowledge graph (KG) facts in LLMs.This KG-based benchmark enables the disentanglement of three effects of unlearning: direct fact removal, reasoning-based leakage, and catastrophic forgetting. In addition, Neighborhood-Expanded Distribution Shaping (NEDS), a novel unlearning framework, is designe
arXiv:2602.22453v3 Announce Type: replace Abstract: Recent work has identified a subset of attention heads in Transformer as retrieval heads, which are responsible for retrieving information from the context. In this work, we first investigate retrieval heads in multilingual contexts. In multilingual language models, we find that retrieval heads are often shared across multiple languages. Expanding the study to cross-lingual setting, we identify Retrieval-Transition heads(RTH), which govern the transition to specific target-language output. Our experiments reveal that RTHs are distinct from retrieval heads and more vital for Chain-of-Thought reasoning in multilingual LLMs. Across four multilingual benchmarks (MMLU-ProX, MGSM, MLQA, and XQuaD) and two model families (Qwen-2.5 and Llama-3.1), we demonstrate that masking RTH induces bigger performance drop than masking Retrieval Heads (RH). Our work advances understanding of multilingual LMs by isolating the attention heads responsible fo
arXiv:2602.12639v2 Announce Type: replace Abstract: Legal text generated by large language models (LLMs) can usually achieve reasonable factual accuracy, but it frequently fails to adhere to the specialised stylistic norms and linguistic conventions of legal writing. In order to improve stylistic quality, a crucial first step is to establish a reliable evaluation method. However, having legal experts manually develop such a metric is impractical, as the implicit stylistic requirements in legal writing practice are difficult to formalise into explicit rubrics. Meanwhile, existing automatic evaluation methods also fall short: reference-based metrics conflate semantic accuracy with stylistic fidelity, and LLM-as-a-judge evaluations suffer from opacity and inconsistency. To address these challenges, we introduce CLASE (Chinese LegAlese Stylistic Evaluation), a hybrid evaluation method that focuses on the stylistic performance of legal text. The method incorporates a hybrid scoring mechanis
arXiv:2602.03619v3 Announce Type: replace Abstract: Nowadays, developing reliable DeepResearch-style long-form report generation remains challenging, as training and evaluation lack verifiable reward signals. Accordingly, rubric-based evaluation has become a common practice. However, existing approaches either rely on coarse, pre-defined rubrics that lack sufficient granularity or depend on manually constructed query-specific rubrics that are costly and difficult to scale. In this paper, we propose a pipeline to train preference-grounded query-specific rubric generators tailored for DeepResearch report generation. We first construct a dataset of DeepResearch-style queries annotated with human preferences over paired reports, and train rubric generators via reinforcement learning with a hybrid reward combining preference consistency, format validity, and LLM-based rubric evaluation. We evaluate the resulting rubric generators in two stages. First, on a held-out human-preference test set
arXiv:2601.12983v4 Announce Type: replace Abstract: Multimodal large language models (MLLMs) are increasingly used to automate chart generation from data tables, improving efficiency but introducing new misuse risks. We present ChartAttack, a framework for evaluating how MLLMs use design misleaders to generate charts that induce incorrect interpretations. We also introduce AttackViz, a chart question-answering (QA) dataset labeled with effective misleaders and their induced incorrect answers. ChartAttack reduces MLLM QA accuracy by 17.2 points in-domain and 11.9 points cross-domain. Conditional deception rates show targeted effects: correct answers shift to attacker-intended answers 11.2\% of the time in-domain and 11.7-14.9\% cross-domain, while originally incorrect answers rarely change. A controlled human study shows that ChartAttack-generated charts also reduce human QA performance. Finally, fine-tuning on AttackViz improves in-domain MLLM robustness to misleading charts. Our findi
arXiv:2601.08146v4 Announce Type: replace Abstract: Existing circuit discovery methods rely on templated tasks with clean counterfactuals, limiting their use on diverse natural text. We adapt Contextual Decomposition for Transformers (CD-T) for unstructured settings via label-balanced activation means and task-directional relevance scoring, enabling counterfactual-free circuit discovery. We leverage the discovered circuits for Circuit-Targeted Supervised Fine-Tuning (CT-SFT), restricting parameter updates to task-relevant heads and LayerNorm. Experiments on NusaX cross-lingual sentiment transfer show that CT-SFT is highly competitive for low-resource adaptation. While non-circuit sparse updates and full fine-tuning sometimes match target accuracy through capacity recruitment, CT-SFT most consistently avoids catastrophic forgetting, preserving source-language and related-task performance. Extensions to XNLI support the source-retention and intervention findings on a harder task and two
arXiv:2601.06536v3 Announce Type: replace Abstract: We present Expos\'ia, the first public dataset that connects writing and feedback in higher education, enabling research on educationally grounded computational approaches to teaching and evaluating academic writing. Expos\'ia includes student research project proposals and peer and instructor feedback consisting of comments and free-text reviews. The dataset was collected in the "Introduction to Scientific Work" course of the Computer Science. Expos\'ia reflects the multi-stage nature of the academic writing process that includes drafting, receiving feedback, and revising the writing based on the feedback received. Both the project proposals and peer feedback are accompanied by human assessment scores based on a fine-grained, pedagogically-grounded schema for writing and feedback assessment that we develop. We use Expos\'ia to benchmark state-of-the-art large language models (LLMs) on two tasks: automated scoring of (1) the proposals
arXiv:2601.05520v2 Announce Type: replace Abstract: Despite strong performance on many tasks, large language models (LLMs) show limited ability in historical and cultural reasoning, particularly in non-English contexts such as Chinese history. Taxonomic structures offer an effective mechanism to organize historical knowledge and improve understanding. However, manual taxonomy construction is costly and difficult to scale. Therefore, we propose \textbf{CHisAgent}, a multi-agent LLM framework for historical taxonomy construction in ancient Chinese contexts. CHisAgent decomposes taxonomy construction into three role-specialized stages: a bottom-up \textit{Inducer} that derives an initial hierarchy from raw historical corpora, a top-down \textit{Expander} that introduces missing intermediate concepts using LLM world knowledge, and an evidence-guided \textit{Enricher} that integrates external structured historical resources to ensure faithfulness. Using the \textit{Twenty-Four Histories}, w
arXiv:2511.08590v2 Announce Type: replace Abstract: Large Language Model (LLM) routing has demonstrated strong capability in balancing response quality with computational cost. As users exhibit diverse preferences, personalization has attracted increasing attention in LLM routing, since even identical queries may require different models to generate responses tailored to individual needs. However, existing approaches are not fully personalized and often fail to faithfully capture the complex interactions between users and LLMs. Moreover, user preference data is typically scarce and inconsistent in format, which limits the effectiveness of methods that directly leverage user-specific data. To address these challenges, we propose GMTRouter, which represents multi-turn user-LLM interactions as a heterogeneous graph with five node types: user, LLM, query, response and turn, thereby maximally preserving the rich relational structure of the interaction. Through a lightweight inductive graph
arXiv:2510.07177v2 Announce Type: replace Abstract: Aspect-based summarization has attracted significant attention for its ability to generate more fine-grained and user-aligned summaries. While most existing approaches assume a set of predefined aspects as input, real-world scenarios often present challenges where these given aspects may be incomplete, irrelevant, or entirely missing from the document. Users frequently expect systems to adaptively refine or filter the provided aspects based on the actual content. In this paper, we initiate this novel task setting, termed Content-Aware Refinement of Provided Aspects for Summarization (CARPAS), with the aim of dynamically adjusting the provided aspects based on the document context before summarizing. We construct three new datasets and conduct pilot experiments with four representative prompting strategies. Our analysis reveals that LLMs often over-generate aspects, resulting in excessively long and misaligned summaries. Building on th
arXiv:2509.00030v4 Announce Type: replace Abstract: Current sign language translation (SLT) systems attempt to learn all aspects of signing---manual gestures, high-speed fingerspelling, and asynchronous non-manual facial cues---within a single end-to-end network. Learning multiple tasks without detailed supervision leads to poor recognition of fingerspelled proper nouns and technical terms, and leaves rich disambiguating information from lip movements largely unexploited. We introduce SignBind-LLM, a modular framework that addresses these limitations through three dedicated expert streams: one for continuous signing, one for fingerspelling, and one for lipreading. Each expert is pre-trained independently using CTC on approximately two million automatically generated pseudo-gloss sequences, removing the need for manual gloss annotation. A lightweight transformer with learned temporal alignment fuses the expert outputs, and a pre-trained language model translates the resulting pseudo-glo
arXiv:2507.08468v2 Announce Type: replace Abstract: This paper provides an experimental evaluation of the capability of large language models (LLMs) to assist in legal decision-making within the framework of Austrian and European Union value-added tax (VAT) law. In tax consulting practice, clients often describe cases in natural language, making LLMs a prime candidate for supporting automated decision-making and reducing the workload of tax professionals. Given the requirement for legally grounded and well-justified analyses, the propensity of LLMs to hallucinate presents a considerable challenge. The experiments focus on two common methods for enhancing LLM performance: fine-tuning and retrieval-augmented generation (RAG). In this study, these methods are applied on both textbook cases and real-world cases from a tax consulting firm to systematically determine the best configurations of LLM-based systems and assess the legal-reasoning capabilities of LLMs. The findings highlight the p
arXiv:2506.00290v2 Announce Type: replace Abstract: This paper introduces DLM-One, a score-distillation-based framework for one-step sequence generation with continuous diffusion language models (DLMs). DLM-One eliminates iterative refinement by aligning the scores of a student model's outputs with the score function of a pretrained teacher DLM in the forward-diffused noisy space. We demonstrate that our framework is architecture-agnostic and robust across diverse continuous manifolds, including standard token embedding spaces and logit simplex spaces. Through experiments on multiple representative DLMs, we show that DLM-One achieves up to $\sim$2000$\times$ speedup in sampling steps and $\sim$500$\times$ in wall-clock time, while maintaining competitive performance on benchmark text generation tasks. We further analyze failure modes in language-domain diffusion distillation and propose an adversarially-regularized two-stage training scheme to prevent student degeneration. Our findings
arXiv:2505.23713v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly deployed in socially grounded applications, where success requires interpreting context, inferring others' mental states, and reasoning about unreliable information. Yet existing benchmarks rarely evaluate these demands jointly in complex, evolving settings. We introduce SocialMaze, a benchmark that organizes six tasks across social deduction games, daily-life interactions, and digital community platforms along three descriptive design axes: deep reasoning, dynamic interaction, and information uncertainty. These axes characterize intended sources of task difficulty rather than latent, factor-analytic dimensions of model capability. Automated checks and human validation support data quality. Evaluations of twelve proprietary and open-weight LLMs show substantial variation in the use of evolving interaction histories; stronger chain-of-thought reasoners perform better on tasks requiring deep
arXiv:2503.22411v2 Announce Type: replace Abstract: Is political incivility rising across democracies - and if so, why? Analysing approximately 13.8 million tweets from parliamentarians in 26 countries with a validated large language model, we find that the share of severe incivility nearly doubled between 2017 and 2022, from $\approx$2.7\% to $\approx$5.8\% of tweets. Populism is the strongest party-level predictor (OR~=~1.46, 95\%~CI: 1.29--1.65), and its association with incivility is amplified in opposition. Stronger liberal democracy is associated with lower severe incivility (OR~=~0.84 per SD, 95\%~CI: 0.72--0.98). The rise in incivility is overwhelmingly behavioral - party families across the spectrum becoming more aggressive - rather than compositional: the growing prominence of radical parties explains little of it. Nationalist and radical-right parties remain the most uncivil throughout, but the gap narrows as their 2020 peak partially recedes while mainstream parties become
arXiv:2503.17039v3 Announce Type: replace Abstract: Automatic text summarization relies on automatic evaluation to quickly determine the quality of summarization models via automatic metrics and LLM-as-a-Judge models. However, these techniques require meta-evaluation to ensure that they capture human judgments correctly. In this paper, we explore this meta-evaluation beyond English by generating a new multilingual summary meta-evaluation dataset (BASSE), which comprises human judgments on 2,040 abstractive summaries, generated either manually or by five Large Language Models (LLMs) with four different prompts. For each summary, annotators evaluate five criteria on a 5-point Likert scale: coherence, consistency, fluency, relevance, and 5W1H. We then benchmark automatic summarization metrics and LLM-as-a-Judge models. Our results show that currently proprietary judge LLMs have the highest correlation with human judgments, followed by criteria-specific automatic metrics, while open-source
arXiv:2410.10136v2 Announce Type: replace Abstract: In customer contact centers, human agents often struggle with long average handling times (AHT) due to the need to manually interpret queries and retrieve relevant knowledge base (KB) articles. While retrieval augmented generation (RAG) systems using large language models (LLMs) have been widely adopted in industry to assist with such tasks, RAG faces challenges in real-time conversations, such as inaccurate query formulation and redundant retrieval of frequently asked questions (FAQs). To address these limitations, we propose a decision support system that can look beyond RAG by first identifying customer questions in real time. If the query matches an FAQ, the system retrieves the answer directly from the FAQ database; otherwise, it generates answers via RAG. Our approach reduces reliance on manual queries, providing responses to agents within 2 seconds. Deployed in AI-powered human-agent assist solution at Minerva CQ, this system i
arXiv:2406.16893v3 Announce Type: replace Abstract: The emergence of Transformer-based Large Language Models (LLMs) has substantially augmented the capabilities of Natural Language Processing (NLP), thereby intensifying the demand for computational resources. Therefore, enhancing efficiency based on factors like computational requirements, energy consumption, carbon footprint and financial cost has become a vital area of research. This motivates us to conduct a survey on Transformer-based LLMs in NLP from the perspective of efficiency. In this survey of 312 articles, the efficiency-improvement endeavors have been systematically discussed targeting various aspects such as data curation, model design, model downsizing, and dynamic inferencing. This has been augmented with efficiency considerations in model adaptation strategies like pre-training, fine-tuning, prompt-engineering and Retrieval-Augmented Generation (RAG). Furthermore, a statistical analysis followed by an in-depth evaluatio
arXiv:2609.02849v1 Announce Type: cross Abstract: Competitive programming has become a key test of large language model reasoning, with international competitions such as IOI and ICPC representing its most challenging settings. We present an end-to-end specialization pipeline combining large-scale problem curation, synthetic reasoning traces, supervised fine-tuning (SFT), and reinforcement learning (RL). Using 22,000 curated problems, we train Nemotron-3-Nano-CC (30B-A3B) with SFT and RL and Nemotron-3-Ultra-CC (550B-A55B) with SFT alone. We further introduce GenCorrect, a feedback-driven test-time compute strategy that iteratively generates, evaluates, and refines diverse solutions. On IOI 2025, Nano-CC improves from 130 points to 291 after post-training and to 468 with GenCorrect, exceeding the gold threshold of 438.3 while Ultra-CC reaches 502. Guided by these results, we develop a competition-specific Ultra-CC system and evaluate it prospectively during IOI 2026. Under the same tim
arXiv:2609.02797v1 Announce Type: cross Abstract: People increasingly use language models to support life decisions. Many such decisions involve a probabilistic forecast: How likely is a major life event, a natural disaster, or an economic outcome? Users of language models may implicitly trust that these forecasts fall out of a coherent world model. In this paper, we evaluate the coherence of language model probabilistic forecasts through a procedure that builds on a theorem due to de Finetti. We elicit forecasts from language models across events generated from stock returns data. We then use linear programs to compute the largest Dutch-book profit - the profit an arbitrageur could guarantee by betting against model-generated probabilities - which we use as a measure of incoherence. Our procedure does not require outcome labels, so we can evaluate coherence even in settings where outcomes are not observed or have not yet resolved. We find substantial evidence of incoherence in languag
arXiv:2609.02780v1 Announce Type: cross Abstract: Streaming video understanding is a critical capability for real-world applications, including embodied intelligence, autonomous driving, industrial monitoring, surveillance and early warning, and wearable assistants. However, processing continuous video streams with multimodal large language models (MLLMs) is computationally expensive. Existing efforts have explored reducing streaming overhead through visual token pruning, token merging, quantization, on-demand frame retrieval, and context offloading. However, most existing methods overlook the dimension of model depth. Repeatedly executing full-depth MLLM prefill over incoming frames is prohibitively expensive, incurring substantial computational overhead and causing the KV cache to grow at a rate directly proportional to the prefill depth. To address these challenges, we propose ShallowStream, a novel framework that leverages the shallow layers of an MLLM to simultaneously perform fra
arXiv:2609.02749v1 Announce Type: cross Abstract: Autonomous agents are beginning to carry out machine-learning (ML) research end to end. These agents combine a model backbone with a harness for planning, execution, memory, and verification, but this architecture still leaves domain-specific know-how outside the agent. We call this missing layer operational knowledge, the know-how that separates knowing a method from making it work. That knowledge is not absent from the field. It appears in repositories and papers, but in forms written for human readers and too large to load during a task. Once distilled into compact, verified skills, this knowledge can be reused across tasks rather than rediscovered during each run. We present DisCo, a skill-powered research agent that creates skills and uses them during research. Its distillation runs in two complementary forms: task-agnostic, condensing the field's widely used repositories into reusable skills, and task-oriented, producing the skill
arXiv:2609.02745v1 Announce Type: cross Abstract: Selecting a retrieval model for a production RAG system requires reliable comparative evaluation, but obtaining relevance judgments at scale is expensive and difficult to repeat as new candidate systems arrive. We study pooled LLM evaluation, in which an LLM judges the union of documents retrieved by the current set of candidate systems, and the pool is then expanded incrementally as new systems are introduced by judging only the new documents they contribute. These judgments are reused to evaluate all systems on a common basis. We validate this approach on four retrieval benchmarks with 11 systems spanning dense, sparse, and hybrid configurations, and deploy it to compare 62 retrieval configurations for a financial news QA system. Pooled LLM rankings correlate strongly with gold-standard evaluation across datasets, and 97% of pairwise system orderings are preserved once bootstrap uncertainty in the qrels is taken into account. In produ
arXiv:2609.02707v1 Announce Type: cross Abstract: Does the door-in-the-face technique work on language models? In humans, a large request that is refused makes a smaller follow-up request more likely to be granted. We test this on nine production models from three providers: each model refuses a large request, then receives a smaller version of the same request, and we compare its compliance with asking directly. The answer depends on the model. On Anthropic's frontier models the technique works: Opus 5 answers the smaller request 65.8% of the time after refusing the larger one, against 29.3% when asked directly. On the frontier models of OpenAI and Google, and on Haiku 4.5, it backfires, lowering compliance by 15.5 to 23.0 points. A control locates the effect: a refused large request on an unrelated topic does less than the related one on all nine models, so the concession itself matters everywhere, while the reaction to having just refused something differs by model family. The techn
arXiv:2609.02649v1 Announce Type: cross Abstract: Aggregating noisy, conflicting textual hypotheses into a reliable consensus is a fundamental challenge when deploying NLP systems in real-world industrial settings. While monolithic Large Language Model (LLM) agents offer unbounded expressivity for tasks like Root Cause Analysis (RCA), they suffer from context limits, compounding hallucinations, and prohibitive inference latency. Traditional weak supervision offers statistical rigor but is mathematically restricted to discrete classes. We present Loom, a generative consensus framework deployed for real-world RCA that bridges these paradigms. Loom aggregates open-form hypotheses emitted by modular heuristics (diagnostic templates dynamically populated with episode-specific entities, times, and metrics) by projecting them into a continuous embedding space, and resolves conflicting signals with an iterative centroid-based reweighting algorithm. The resulting consensus weights ground a sing
arXiv:2609.02623v1 Announce Type: cross Abstract: Voice actors often re-read the same script while modifying their delivery in response to performance directions. We study this setting as direction-following TTS, where a system generates a new utterance that reflects a given direction relative to a reference utterance while preserving speaker identity and linguistic content. A key challenge is the lack of training data capturing such relative modifications. To address this, we propose a scalable pseudo-triplet construction pipeline that generates~(reference utterance, direction text, modified utterance) triplets. It generates controlled style variations using an impression-controllable TTS model and uses an LLM to produce natural language directions from estimated impression differences. Experimental results demonstrate that pseudo-triplets alone enable stable speaker-preserving modification, and that combining pseudo and recorded data further improves direction alignment while maintai
arXiv:2609.02486v1 Announce Type: cross Abstract: Document Visual Question Answering (DocVQA) often leverages Retrieval-Augmented Generation (RAG), where late-interaction encoders are commonly used to identify document pages relevant to a user query, before answer generation by a Large Vision-Language Model (LVLM). Existing approaches typically retrieve a fixed top-$k$ number of pages regardless of query complexity, which increases LVLM latency and may degrade answer accuracy. We introduce ViSAR (Visual Semantic Activation Retrieval), a training-free adaptive-$k$ retrieval method for late-interaction visual document retrieval. ViSAR operates directly in the embedding space to construct a query-conditioned page-level similarity matrix that highlights query-relevant semantics and dynamically determines the number of pages to retrieve. Across multiple encoders and LVLMs, ViSAR retrieves compact, query-adapted page sets that reduce RAG latency by up to 58.7\%, while maintaining or improvin
arXiv:2609.02451v1 Announce Type: cross Abstract: In this paper, we propose a scalable Kronecker-based approximation that captures cross-layer interactions without storing the entire Fisher matrix, enabling practical Hessian analysis for billion-parameter networks where full computation is infeasible. Our approach reveals consistent vulnerability patterns: value projection layers exhibit the highest sensitivity and strongest cross-layer correlations across multiple model families, while other components exhibit architecture-specific behaviors. Through extensive experiments on quantization, sparsification, inter-layer corruption, and post-corruption fine-tuning, we demonstrate that our approximation strongly correlates with both performance degradation and recovery. Our framework provides a practical, theoretically grounded tool for identifying fragile components in large models, opening new avenues for guided compression and optimization strategies, such as mixed-precision allocation,
arXiv:2609.02421v1 Announce Type: cross Abstract: Large Language Models (LLMs) have recently demonstrated strong capabilities in automated travel itinerary generation. However, real- world travel planning is inherently uncertain: transportation delays, crowd fluctuations, and unexpected stochastic delays frequently inval- idate otherwise feasible schedules. Existing benchmarks like TravelPlanner and TripCraft assume deterministic environments, evaluating only static constraint satisfaction and ignoring whether generated plans remain robust when such uncertainties arise. To address this limitation, we introduce UTP-Bench1 , a large-scale benchmark for uncertainty-aware travel planning. The dataset integrates real-world travel data spanning 504 cities of India, including attractions, restau- rants, accommodations, and multi-modal trans- portation networks. To model realistic disrup- tions, UTP-Bench incorporates empirical delay distributions and crowd-density patterns col- lected from ma
arXiv:2609.02343v1 Announce Type: cross Abstract: Recent advances in audio-language modeling have been driven by large-scale audio captioning datasets. However, existing datasets remain limited by low semantic diversity, generic descriptions lacking acoustic details, and one-to-one audio-caption mappings that poorly reflect the inherent ambiguity of auditory perception. We introduce SonicCaps, a large-scale audio captioning dataset comprising ~15M captions paired with ~700k audio clips, generated using a multi-modal large language model (Qwen3-Omni) conditioned on both audio and text. To explicitly promote diversity, we generate around 24 captions per audio via structured prompt engineering and few- shot generation, spanning main descriptions, rephrased variants (verbosity, style) and semantic tags. Human evaluation shows that SonicCaps is rated significantly higher than existing captioning datasets, with fine-grained analyses indicating that our captions are perceived as more descript
arXiv:2609.02336v1 Announce Type: cross Abstract: Effective in-context learning (ICL) for complex reasoning relies on selecting the right demonstrations. Traditional retrieval methods based on surface similarity fail to capture the underlying problem-solving logic. Recent logic-based methods address this by matching predefined reasoning steps, but the rigid rules and exact-match criteria is improper to handle flexible or diverse reasoning processes. To address the problem, we propose SALA, a Semantic-Aware Logical Alignment framework. Instead of relying on a fixed inventory, SALA automatically learns task-specific reasoning operations. It then embeds these operations into a continuous semantic space and uses dynamic time warping (DTW) to align the reasoning sequences. This approach allows for soft, flexible matching of reasoning logic while remaining highly interpretable. Experiments across four reasoning benchmarks and three LLMs demonstrate that SALA outperforms existing demonstratio
arXiv:2609.02316v1 Announce Type: cross Abstract: Generative engine optimization (GEO) enables content producers to increase the visibility of their web pages in generative search engines, but the same techniques can deliver targeted misinformation when adversaries publish ordinary-looking GEO-optimized documents that victim large language models (LLMs) retrieve and synthesize into distorted answers. No existing benchmark evaluates defenses against this threat under controlled conditions. Therefore, we present Counter-GEO-Bench, a defense benchmark that pairs 247 human-verified, quality-gated queries with information-preserving and information-distorting GEO rewrites, and evaluates defenses on attack success rate (ASR), false positive rate, and answer quality across three victim LLMs. Under Counter-GEO-Bench, three off-the-shelf defenses (Granite Guardian, Llama Guard 3, and NeMo Self-Check Fact-Checking) reduce ASR by at most 5.7% relative, while Granite Guardian's reduction is not st
arXiv:2609.02302v1 Announce Type: cross Abstract: A core obstacle to alignment evaluation is evaluation awareness: capable models can tell when they are being tested rather than deployed, weakening the conclusions a safety evaluation can support. We present two techniques that make simulated alignment evaluations harder to distinguish from real deployments. Our first technique, critique refinement, spends additional inference-time compute on each simulator action: the simulator generates multiple candidate actions, refines them using feedback from an instance of the target model on how to make them more realistic, and continues the evaluation with the most deployment-like candidate. Our second technique, DISH (Deployment-Imitating SWE-Agent Harness), wraps the target in an agent harness, reducing the gap between simulated and real deployment environments in coding settings. We test the techniques on multiple target models and find that they compose: applying both yields larger realism
arXiv:2609.02292v1 Announce Type: cross Abstract: The rapid proliferation of large language models (LLMs) and the growing diversity of their applications presents a unique optimization opportunity: selecting the right model for the task, while optimizing for speed, cost, and quality at a per-task level. However, inference endpoints can vary widely in quality, price, latency, context support, tool use, domain expertise, and reasoning behavior. This heterogeneity makes manual heuristics difficult to maintain and unlikely to achieve consistently favorable speed--cost--quality trade-offs on their own. We introduce \router{}, a lightweight GLiClass-based router that assigns a suitability score to each inference-time model label without autoregressive generation. The released 0.6B-parameter checkpoint combines a Qwen3 decoder with a shallow bidirectional scorer. Its decoder-KV execution path preserves a text-only key--value cache across a session, encodes only new dialogue turns, and evaluat
arXiv:2609.02285v1 Announce Type: cross Abstract: A long-held intuition in interpretability research is that representational entanglement, the sharing of structure between knowledge domains in a neural network, makes unlearning harder. While the intuition is widespread, it has never been directly tested in a controlled experiment. We present a way to do so: by repurposing Selective Gradient Masking (SGTM), we train a suite of six 254M-parameter language models on English Wikipedia with graded levels of disentanglement between biology and non-biology knowledge. Applying three standard unlearning methods to every model in the suite, we find that more disentangled models consistently achieve better retain-forget trade-offs: at a fixed level of forgetting, the most disentangled models incur roughly $4\times$ lower retain cost under two of the three methods, and $1.3\times$ lower under the third. Because our intervention changes only the model, not the data or the unlearning algorithm, thi
arXiv:2609.02273v1 Announce Type: cross Abstract: Model merging provides an efficient paradigm for constructing multi-task large language models (LLMs) without full model retraining, yet it remains challenged by parameter interference. While existing methods aim to preserve the capabilities of individual expert models and mitigate interference, they generally do not directly learn from the potentially degraded behaviors exposed by naive merging. In this paper, we propose a conflict-driven preference optimization framework for model merging (CoMerge), which reformulates model merging as a preference optimization problem. The approach utilizes a self-supervised, conflict-driven strategy that leverages the defects of naive merging methods (e.g., task arithmetic) as hard negative samples to construct preference pairs without external annotations. By applying preference optimization to refine lightweight, tensor-wise merging coefficients, CoMerge enables the model to mitigate parameter-spac
arXiv:2609.02262v1 Announce Type: cross Abstract: Ragebait refers to online content intentionally designed to provoke anger or outrage and thereby increase attention and engagement. However, reliable large-scale detection and systematic analysis of ragebait remain limited, hindering efforts to understand its prevalence, impact, and mitigation. This study aims to develop an effective ragebait detection framework and to clarify the characteristics of ragebait at scale, providing a basis for understanding and mitigating emotionally provocative content online. We constructed a labeled dataset with the assistance of a large language model (LLM) and trained several Japanese language models for ragebait detection. The resulting ensemble classifier was then applied to a large-scale dataset of Japanese-language posts on X. Our analysis shows that ragebait is more prevalent in politically and socially contentious topics, including politics, discrimination, public health, and interpersonal confli
arXiv:2609.02253v1 Announce Type: cross Abstract: Deep research agents augment large language models with external tools to answer complex, long-horizon questions through multi-turn reasoning. Learning from prior experience is crucial for continual improvement, yet existing methods either retrieve verbose task-specific traces that burden decision-making, or distill procedural skills that remain decoupled from downstream policy adaptation. We propose APEx, a hierarchical experience utilization framework that organizes interaction history into instance-level trajectory memories and category-level procedural skills, and couples them through a closed-loop architecture of Executor, Distiller, and Planner. The three modules are optimized via a three-stage alternating GRPO training paradigm, enabling reward-guided skill distillation rather than fixed-prompt generation. At test time, distilled skills serve as procedural priors for online Planner adaptation through skill-guided test-time reinfo
arXiv:2609.02250v1 Announce Type: cross Abstract: Ride-sharing, which allows multiple passengers with different origin-destination (OD) pairs to share a single vehicle, is a challenging operational problem, as it requires orders with different OD pairs to be efficiently bundled and assigned to vehicles under uncertain and varying scenarios. Although multi-agent reinforcement learning (MARL) solutions have achieved promising performance, they suffer from limited generalization (adapting to different environmental scenarios), low transferability (adapting to different platform objectives), and training difficulties in large-scale systems, such as the curse of dimensionality. Recently, motivated by the scaling of large language models (LLMs), several works have incorporated LLMs into ride-hailing systems, either by employing LLMs directly as decision-making agents or using them for automatic algorithm design. However, none of these approaches support vehicle sharing, which complicates the
arXiv:2609.02207v1 Announce Type: cross Abstract: Real-world personally identifiable information (PII) redaction often operates on document images---scans, screenshots, and PDF renderings---where OCR errors, layout structure, and visual noise determine whether sensitive information is actually removed. Existing PII benchmarks are mostly text-centric and do not measure document-level redaction risk: a page remains unsafe if even one identifier is missed. We introduce LeakageBench, a challenge set of 500 document images with 11,954 GDPR-aligned PII annotations spanning direct identifiers, linkage keys, and contextual re-identification surfaces. We evaluate generic OCR pipelines, commercial and task-adapted OCR-dependent detectors, and OCR-free vision-language models using entity-level F1, group-wise leakage, and document-level leakage metrics. Code Interpreter raises GPT-5.5 localization F1 from 0.090 to 0.249, but critical page-level leakage remains 0.968. These results show that strong
arXiv:2609.02133v1 Announce Type: cross Abstract: Empathetic response generation requires models to decide not only what to say, but also how to respond to the previous speaker's affective situation. We formulate this as response-side affective-orientation control and use multi-annotator emoji distributions as weak affective--attitudinal evidence, rather than as output symbols or gold labels, to induce a latent control space that operationally approximates listener stance. We construct EmojiDialogue, an utterance-level extension of EmpatheticDialogues with emoji votes and confidence scores, and propose EmoStance, which models source-side affective expression, predicts a soft response-side orientation from dialogue context and speaker roles, and steers a frozen instruction-tuned LLM through continuous prefix embeddings. In blind pairwise evaluation with 20 annotators and 800 judgments, EmoStance achieves a 62.2% decisive win rate, with the clearest gains in contextual specificity and pe
arXiv:2609.02094v1 Announce Type: cross Abstract: LLM-based multi-agent systems have shown strong performance on complex tasks, yet continual improvement from interaction experience remains challenging. Existing self-reflection methods build experience memories, but memories are mostly hard to invoke, refine, or scale, while agent skills offer a more actionable unit: structured procedural knowledge that specifies when to act, how to act, and which resources or tools to use. We introduce MASkills, a continual learning framework that optimizes multi-agent LLM systems through agent skills. MASkills presents a new agent-optimization pipeline that integrates skill-conditioned credit assignment, hierarchical credit aggregation, and momentum-smoothed optimization, enabling agent skill libraries to evolve through refinement, induction, consolidation, and pruning. Experiments on HotpotQA, LoCoMo, and GAIA demonstrate the effectiveness of MASkills across multiple agentic tasks. Our code is avail
arXiv:2609.02083v1 Announce Type: cross Abstract: Removing complete transformer layers preserves a standard serving architecture, but existing depth-compression methods can lose substantial quality, and the loss varies unpredictably across models. We introduce XMerge, a post-training method with two components. Cross-axis selection identifies a block with low relative-magnitude and angular hidden-state change, and local boundary reconstruction re-fits the adjacent surviving block to match the original two-block output. XMerge uses no task labels or end-to-end fine-tuning, and it introduces neither architectural changes nor additional inference-time parameters. Across seven Llama and Qwen backbones (0.5B-8B), five published baselines, and three layer-reduction levels, its advantage over baselines is largest at the most aggressive removal: at k=4 it ranks first on six of seven backbones on CORE (a 22-task aggregate) and, separately, on six of seven on MMLU (five of seven on both at once)
arXiv:2609.02082v1 Announce Type: cross Abstract: Visual modality enhances the capabilities of multimodal large language models (MLLMs) but also introduces a safety concern: a benign textual query may convey harmful intent when grounded in a visual image. We term this cross-modal safety drift and our pilot studies show that the safety response rate for such requests is substantially lower than that for requests containing explicitly unsafe text. This paper aims to systematically study this issue. First, we conduct an empirical analysis to identify representative unsafe response patterns. Building on these, we interpret model representations and attentions, revealing that visually risky cues receive limited attention and weakly trigger refusal. Motivated by the observation that safety signals from unsafe text processing can be transferred, we propose safety-awareness representation transfer (SRT), a lightweight direction-refinement method that mitigates cross-modal safety drift with a f
arXiv:2609.02049v1 Announce Type: cross Abstract: LLMs latent-state reasoning methods replace discrete intermediate tokens with continuous states, such as weighted mixtures of token embeddings, to retain multiple possible reasoning directions rather than committing to one. Yet pretrained language models often fail to preserve these mixtures. We study why through a combination of theoretical analysis and controlled empirical investigations on a variety of models. We identify three independent, distinct sources of failure. First, transformer architectures already distort mixture geometry, and training substantially amplifies this effect. Moreover, the failure can occur even if the model transports mixtures perfectly linearly: the softmax readout and autoregressive feedback form a dynamical system that either amplifies small differences until one component of the mixture dominates or contracts different mixtures until they become indistinguishable. We verify this theoretical prediction em