Papers

Deep learning-based automatic medical image segmentation plays a critical role in clinical diagnosis and treatment planning but remains challenging in few-shot scenarios due to the scarcity of annotated training data. Recently, self-supervised foundation models such as DINOv3, which were trained on large natural image datasets, have shown strong potential for dense feature extraction that can help with the few-shot learning challenge. Yet, their direct application to medical images is hindered by domain differences. In this work, we propose DINO-AugSeg, a novel framework that leverages DINOv3 features to address the few-shot medical image segmentation challenge. Specifically, we introduce WT-Aug, a wavelet-based feature-level augmentation module that enriches the diversity of DINOv3-extracted features by perturbing frequency components, and CG-Fuse, a contextual information-guided fusion module that exploits cross-attention to integrate semantic-rich low-resolution features with spatially detailed high-resolution features. Extensive experiments on six public benchmarks spanning five imaging modalities, including MRI, CT, ultrasound, endoscopy, and dermoscopy, demonstrate that DINO-AugSeg consistently outperforms existing methods under limited-sample conditions. The results highlight the effectiveness of incorporating wavelet-domain augmentation and contextual fusion for robust feature representation, suggesting DINO-AugSeg as a promising direction for advancing few-shot medical image segmentation. Code and data will be made available on https://github.com/apple1986/DINO-AugSeg.

Negative constraints (instructions of the form "do not use word X") represent a fundamental test of instruction-following capability in large language models. Despite their apparent simplicity, these constraints fail with striking regularity, and the conditions governing failure have remained poorly understood. This paper presents the first comprehensive mechanistic investigation of negative instruction failure. We introduce semantic pressure, a quantitative measure of the model's intrinsic probability of generating the forbidden token, and demonstrate that violation probability follows a tight logistic relationship with pressure ($p=\sigma(-2.40+2.27\cdot P_0)$; $n=40{,}000$ samples; bootstrap $95%$ CI for slope: $[2.21,,2.33]$). Through layer-wise analysis using the logit lens technique, we establish that the suppression signal induced by negative instructions is present but systematically weaker in failures: the instruction reduces target probability by only 5.2 percentage points in failures versus 22.8 points in successes – a $4.4\times$ asymmetry. We trace this asymmetry to two mechanistically distinct failure modes. In priming failure (87.5% of violations), the instruction's explicit mention of the forbidden word paradoxically activates rather than suppresses the target representation. In override failure (12.5%), late-layer feed-forward networks generate contributions of $+0.39$ toward the target probability – nearly $4\times$ larger than in successes – overwhelming earlier suppression signals. Activation patching confirms that layers 23–27 are causally responsible: replacing these layers' activations flips the sign of constraint effects. These findings reveal a fundamental tension in negative constraint design: the very act of naming a forbidden word primes the model to produce it.

Confidence estimation (CE) indicates how reliable the answers of large language models are and impacts user trust and decision-making. Existing evaluations mainly concern the alignment between confidence and correctness, but ignore the variability of language: confidence estimates should remain consistent under semantically equivalent prompts or answer variations, while changing when answer meaning differs, as this may indicate a change in correctness. Therefore, we introduce a novel evaluation framework based on three complementary properties: robustness to prompt perturbations, stability across semantically equivalent answers, and sensitivity to semantically different answers. We show that these metrics are largely independent from existing CE metrics, and that common CE methods often fail on them: while most methods achieve high robustness and stability, they struggle to distinguish semantically different answers, potentially because they do not effectively leverage generation-side information. Overall, our framework exposes overlooked limitations of current CE evaluations and provides guidance for selecting confidence estimators for real-world applications.

Language models now provide an interface to express and often solve general problems in natural language, yet their ultimate computational capabilities remain a major topic of scientific debate. Unlike a formal computer, a language model is trained to autoregressively predict successive elements in human-generated text. We prove that chaining a language model's autoregressive output is sufficient to perform universal computation. That is, a language model can simulate the execution of any algorithm on any input. The challenge of eliciting desired computational behaviour can thus be reframed in terms of programmability: the ease of finding a suitable prompt. Strikingly, we demonstrate that even randomly initialized language models are capable of universal computation before training. This implies that training does not give rise to computational expressiveness – rather, it improves programmability, enabling a natural language interface for accessing these intrinsic capabilities.

Chain-of-Thought (CoT) prompting often improves the reasoning performance of large language models (LLMs), but the internal signal that triggers this behavior remains poorly understood. Leveraging the sparse features captured by Sparse Autoencoders (SAEs), we propose a systematic framework to analyze and intervene on the internal representations of LLMs, identifying a small set of latent features that are linked to reasoning behavior and can be causally tested through targeted intervention. Across multiple model families and reasoning benchmarks, we show that steering one or a small number of reasoning-related latent features can substantially induce reasoning behavior without explicit CoT prompting, achieving accuracy comparable to CoT. We further show that the identified features are not tied to particular wording patterns or verbosity, and confirm their causal role in reasoning through suppression experiments that impair performance even under CoT prompting. These results suggest that CoT prompting activates specific latent features to trigger reasoning, and that targeted intervention on these features offers an alternative pathway to elicit efficient reasoning behavior without explicit CoT prompting. Code is available at https://github.com/Zhenghao-He/LatentCoT.

Scientific compound figures combine multiple labeled panels into a single image, and downstream pretraining and retrieval require panel-aligned visual-text pairs. However, in a PubMed Central (PMC)-scale crawl of 346,567 compound figures, 16.3% have no caption and are discarded by existing caption-decomposition pipelines. We propose FigEx2, a visual-conditioned framework that takes only a compound figure as input and jointly produces labeled panel boxes and panel-wise captions. FigEx2 introduces an Entity-Attention Kullback-Leibler (KL) regularizer that aligns the detector's cross-attention with scientific entities annotated for each panel, providing a stable conditioning signal that also improves localization, and applies Group Relative Policy Optimization (GRPO) with a panel-level Entity-F1 reward to optimize scientific faithfulness. We curate BioSci-Fig-Cap for in-domain supervision and contribute physics and chemistry test suites for cross-disciplinary evaluation. FigEx2 achieves 0.751 mAP@0.5:0.95 on BioSci-Fig-Cap, and outperforms Qwen3-VL-8B by 6.80 Entity-F1 on MedICaT for captioning. It also transfers zero-shot to out-of-distribution domains. The source code is available at https://github.com/Huang-AI4Medicine-Lab/FigEx2.

Large Language Models (LLMs) often struggle with creative generation, and multi-agent frameworks that improve reasoning through interaction can paradoxically hinder creativity by inducing content homogenization. We introduce LLM Review, a peer-review-inspired framework implementing Blind Peer Review: agents exchange targeted feedback while revising independently, preserving divergent creative trajectories. To enable rigorous evaluation, we propose SciFi-100, a science fiction writing dataset with a unified framework combining LLM-as-a-judge scoring, human annotation, and rule-based novelty metrics. Experiments demonstrate that LLM Review consistently outperforms multi-agent baselines, and smaller models with our framework can surpass larger single-agent models, suggesting interaction structure may substitute for model scale.

Use cases of sentiment analysis in the humanities often require contextualized, continuous scores. Concept Vector Projections (CVP) offer a recent solution: by modeling sentiment as a direction in embedding space, they produce continuous, multilingual scores that align closely with human judgments. Yet the method's portability across domains and underlying assumptions remain underexplored. We evaluate CVP across genres, historical periods, languages, and affective dimensions, finding that concept vectors trained on one corpus transfer well to others with minimal performance loss. To understand the patterns of generalization, we further examine the linearity assumption underlying CVP. Our findings suggest that while CVP is a portable approach that effectively captures generalizable patterns, its linearity assumption is approximate, pointing to potential for further development. Code available at: github.com/lauritswl/representation-transfer

Large Language Models (LLMs) increasingly operate over long-form dialogues with frequent topic shifts. While recent LLMs support extended context windows, efficient management of dialogue history in practice is needed due to inference cost and latency constraints. We present DyCP, a lightweight context management method implemented outside the LLM that dynamically identifies and retrieves relevant dialogue segments conditioned on the current turn, without offline memory construction. DyCP manages dialogue context while preserving the sequential nature of dialogue without predefined topic boundaries, enabling adaptive and efficient context selection. Across three long-form dialogue benchmarks-LoCoMo, MT-Bench+, and SCM4LLMs-and multiple LLM backends, DyCP achieves competitive answer quality in downstream generation, with more selective context usage and improved inference efficiency.

While NLP typically treats documents as independent and unordered samples, in longitudinal studies, this assumption rarely holds: documents are nested within authors and ordered in time, forming person-indexed, time-ordered $\textit{behavioral sequences}$. Here, we demonstrate the need for and propose a longitudinal modeling and evaluation paradigm that consequently updates four parts of the NLP pipeline: (1) evaluation splits aligned to generalization over people ($\textit{cross-sectional}$) and/or time ($\textit{prospective}$); (2) accuracy metrics separating between-person differences from within-person dynamics; (3) sequence inputs to incorporate history by default; and (4) model internals that support different $\textit{coarseness}$ of latent state over histories (pooled summaries, explicit dynamics, or interaction-based models). We demonstrate the issues ensued by traditional pipeline and our proposed improvements on a dataset of 17k daily diary transcripts paired with PTSD symptom severity from 238 participants, finding that traditional document-level evaluation can yield substantially different and sometimes reversed conclusions compared to our ecologically valid modeling and evaluation. We tie our results to a broader discussion motivating a shift from word-sequence evaluation toward $\textit{behavior-sequence}$ paradigms for NLP.

We introduce VULCA-Bench, a multicultural art-critique benchmark for evaluating Vision-Language Models' (VLMs) cultural understanding beyond surface-level visual perception. Existing VLM benchmarks predominantly measure L1-L2 capabilities (object recognition, scene description, and factual question answering) while under-evaluate higher-order cultural interpretation. VULCA-Bench contains 7,410 matched image-critique pairs spanning eight cultural traditions, with Chinese-English bilingual coverage. We operationalise cultural understanding using a five-layer framework (L1-L5, from Visual Perception to Philosophical Aesthetics), instantiated as 225 culture-specific dimensions and supported by expert-written bilingual critiques. Our pilot results indicate that higher-layer reasoning (L3-L5) is consistently more challenging than visual and technical analysis (L1-L2). The dataset, evaluation scripts, and annotation tools are available under CC BY 4.0 at https://github.com/yha9806/VULCA-Bench.

The rapid proliferation of misinformation across online platforms underscores the urgent need for robust, up-to-date, explainable, and multilingual fact-checking resources. However, existing datasets are limited in scope, often lacking multimodal evidence, structured annotations, and detailed links between claims, evidence, and verdicts. This paper introduces a comprehensive data collection and processing pipeline that constructs multimodal fact-checking datasets in French and German languages by aggregating ClaimReview feeds, scraping full debunking articles, normalizing heterogeneous claim verdicts, and enriching them with structured metadata and aligned visual content. We used state-of-the-art large language models (LLMs) and multimodal LLMs for (i) evidence extraction under predefined evidence categories and (ii) justification generation that links evidence to verdicts. Evaluation with G-Eval and human assessment demonstrates that our pipeline enables fine-grained comparison of fact-checking practices across different organizations or media markets, facilitates the development of more interpretable and evidence-grounded fact-checking models, and lays the groundwork for future research on multilingual, multimodal misinformation verification.

Vision-Language Models (VLMs) excel at visual description yet remain under-validated for cultural interpretation. Existing benchmarks assess perception without interpretation, and common evaluation proxies, such as automated metrics and LLM-judge averaging, are unreliable for culturally sensitive generative tasks. We address this measurement gap with a tri-tier evaluation framework grounded in art-theoretical constructs (Section 2). The framework operationalises cultural understanding through five levels (L1–L5) and 165 culture-specific dimensions across six traditions: Tier I computes automated quality indicators, Tier II applies rubric-based single-judge scoring, and Tier III calibrates the aggregate score to human expert ratings via sigmoid calibration. Applied to 15 VLMs across 294 evaluation pairs, the validated instrument reveals that (i) automated metrics and judge scoring measure different constructs, establishing single-judge calibration as the more reliable alternative; (ii) cultural understanding degrades from visual description (L1–L2) to cultural interpretation (L3–L5); and (iii) Western art samples consistently receive higher scores than non-Western ones. To our knowledge, this is the first cross-cultural evaluation instrument for generative art critique, providing a reproducible methodology for auditing VLM cultural competence. Framework code is available at https://github.com/yha9806/VULCA-Framework.

AI-text detectors achieve high accuracy on in-domain benchmarks, but often struggle to generalize across different generation conditions such as unseen prompts, model families, or domains. While prior work has reported these generalization gaps, there are limited insights about the underlying causes. In this work, we present a systematic study aimed at explaining generalization behavior through linguistic analysis. We construct a comprehensive benchmark that spans 6 prompting strategies, 7 large language models (LLMs), and 4 domain datasets, resulting in a diverse set of human- and AI-generated texts. Using this dataset, we fine-tune classification-based detectors on various generation settings and evaluate their cross-prompt, cross-model, and cross-dataset generalization. To explain the performance variance, we compute correlations between generalization accuracies and feature shifts of 80 linguistic features between training and test conditions. Our analysis reveals that generalization performance for specific detectors and evaluation conditions is significantly associated with linguistic features such as tense usage and pronoun frequency.

Generative AI models ought to be useful and safe across cross-cultural contexts. One critical step toward this goal is understanding how AI models adhere to sociocultural norms. While this challenge has gained attention in NLP, existing work lacks both nuance and coverage in understanding and evaluating models' norm adherence. We address these gaps by introducing a taxonomy of norms that clarifies their contexts (e.g., distinguishing between human-human norms that models should recognize and human-AI interactional norms that apply to the human-AI interaction itself), specifications (e.g., relevant domains), and mechanisms (e.g., modes of enforcement). We demonstrate how our taxonomy can be operationalized to automatically evaluate models' norm adherence in naturalistic, open-ended settings. Our exploratory analyses suggest that state-of-the-art models frequently violate norms, though violation rates vary by model, interactional context, and country. We further show that violation rates also vary by prompt intent and situational framing. Our taxonomy and demonstrative evaluation pipeline enable nuanced, context-sensitive evaluation of cultural norm adherence in realistic settings.

Value alignment is central to the development of safe and socially compatible artificial intelligence. However, how Large Language Models (LLMs) represent and enact human values in real-world decision contexts remains under-explored. We present ValAct-15k, a dataset of 3,000 advice-seeking scenarios derived from Reddit, designed to elicit ten values defined by Schwartz Theory of Basic Human Values. Using both the scenario-based questions and the traditional value questionnaire, we evaluate ten frontier LLMs (five from U.S. companies, five from Chinese ones) and human participants ($n = 55$). We find near-perfect cross-model consistency in scenario-based decisions (Pearson $r \approx 1.0$), contrasting sharply with the broad variability observed among humans ($r \in [-0.79, 0.98]$). Yet, both humans and LLMs show weak correspondence between self-reported and enacted values ($r = 0.4, 0.3$), revealing a systematic knowledge-action gap. When instructed to "hold" a specific value, LLMs' performance declines up to $6.6%$ compared to merely selecting the value, indicating a role-play aversion. These findings suggest that while alignment training yields normative value convergence, it does not eliminate the human-like incoherence between knowing and acting upon values.

Predicting treatment non-response for anxiety and depression is challenging, in part because of sparse symptom assessments in real-world care. We examined whether passively captured, fine-grained emotions serve as linguistic markers of treatment outcomes by analyzing 12 weeks of de-identified teletherapy transcripts from 12,043 U.S. patients with moderate-to-severe anxiety and depression symptoms. A transformer-based small language model extracted patients' emotions at the talk-turn level; a state-space model (VISTA-SSM) clustered subgroups based on emotion dynamics over time and produced temporal networks. Two groups emerged: an improving group (n=8,230) and a non-response group (n=3,813) showing increased odds of symptom deterioration, and lower likelihood of clinically significant improvement. Temporal networks indicated that sadness and fear exerted most influence on emotion dynamics in non-responders, whereas improving patients showed balanced joy, sadness, and neutral expressions. Findings suggest that linguistic markers of emotional inflexibility can serve as scalable, interpretable, and theoretically grounded indicators for treatment risk stratification.

Recent advances in large language models have enabled their application to a range of healthcare tasks. However, aligning LLMs with the nuanced demands of medical ethics, especially under complex real world scenarios, remains underexplored. In this work, we present MedES, a dynamic, scenario-centric benchmark specifically constructed from 260 authoritative Chinese medical, ethical, and legal sources to reflect the challenges in clinical decision-making. To facilitate model alignment, we introduce a guardian-in-the-loop framework that leverages a dedicated automated evaluator (trained on expert-labeled data and achieving over 97% accuracy within our domain) to generate targeted prompts and provide structured ethical feedback. Using this pipeline, we align a 7B-parameter LLM through supervised fine-tuning and domain-specific preference optimization. Experimental results, conducted entirely within the Chinese medical ethics context, demonstrate that our aligned model outperforms notably larger baselines on core ethical tasks, with observed improvements in both quality and composite evaluation metrics. Our work offers a practical and adaptable framework for aligning LLMs with medical ethics in the Chinese healthcare domain, and suggests that similar alignment pipelines may be instantiated in other legal and cultural environments through modular replacement of the underlying normative corpus.

The rapid evolution of Large Language Models (LLMs) has shifted focus from general-purpose capabilities to domain-specific expertise. However, adapting LLMs to specialized fields such as medicine presents two challenge: (1) the "Stability-Plasticity Dilemma", where the model must acquire complex clinical knowledge without suffering from catastrophic forgetting of general world knowledge; and (2) "Task Interference", where disparate sub-tasks, such as medical diagnosis, report summarization, and drug-drug interaction prediction, compete for limited low-rank parameter space. In this paper, we propose Med-MoE-LoRA, a novel framework that integrates Mixture-of-Experts (MoE) with Low-Rank Adaptation (LoRA) to enable efficient multi-task domain adaptation, especially for medical scenarios. Drawing inspiration from recent advances, our framework employs an asymmetric expert distribution where deeper layers are equipped with a higher density of LoRA experts to capture complex semantic abstractions. We further introduce a "Knowledge-Preservation Plugin", inspired by LoRA MoE, to isolate and protect general-purpose reasoning. By utilizing soft merging with adaptive routing and rank-wise decoupling, Med-MoE-LoRA achieves superior performance in medical benchmarks while reducing interference. Experimental results demonstrate that our approach consistently outperforms standard LoRA and conventional MoE architectures across multiple clinical NLP tasks while retaining the model's general cognitive capabilities.

Growing context lengths in transformer-based language models have made the key-value (KV) cache a critical inference bottleneck. While many KV cache pruning methods have been proposed, they have not yet been adopted in major inference engines due to speed–accuracy trade-offs. We introduce KVzap, a fast, input-adaptive approximation of KVzip that works in both prefilling and decoding. On Qwen3-8B, Llama-3.1-8B-Instruct, and Qwen3-32B across long-context and reasoning tasks, KVzap achieves $2$–$4\times$ KV cache compression with negligible accuracy loss and achieves state-of-the-art performance on the KVpress leaderboard. Code and models are available at https://github.com/NVIDIA/kvpress.

In human conversation, both interlocutors play an active role in maintaining mutual understanding. When listeners are uncertain about what speakers mean, for example, they can request clarification. It is an open question for language models whether they can assume a similar listener role, recognizing and expressing their own uncertainty through clarification. We argue that reference games are a suitable testbed to approach this question as they are controlled, self-contained, and make clarification needs explicit and measurable. To test this, we evaluate three vision-language models comparing a baseline reference resolution task to an experiment where the models are instructed to request clarification when uncertain. The results suggest that even in such simple tasks, models often struggle to recognize internal uncertainty and translate it into adequate clarification behavior. This demonstrates the value of reference games as testbeds for interaction qualities of (vision and) language models.

The increased use of Large Language Models (LLMs) in sensitive domains leads to growing interest in how their confidence scores correspond to fairness and bias. This study examines the alignment between LLM-predicted confidence and human-annotated bias judgments. Focusing on gender bias, the research investigates probability confidence calibration in contexts involving gendered pronoun resolution. The goal is to evaluate if calibration metrics based on predicted confidence scores effectively capture fairness-related disparities in LLMs. The results show that, among the six state-of-the-art models, Gemma-2 demonstrates the worst calibration according to the gender bias benchmark. The primary contribution of this work is a fairness-aware evaluation of LLMs' confidence calibration, offering guidance for ethical deployment. In addition, we introduce a new calibration metric, Gender-ECE, designed to measure gender disparities in resolution tasks.

Large language models (LLMs) are increasingly used as conversational partners for learning, yet the interactional dynamics supporting users' learning and engagement are understudied. We analyze the linguistic and interactional features from both LLM and participant chats across 397 human-LLM conversations about socio-political issues to identify the mechanisms and conditions under which LLM explanations shape changes in political knowledge and confidence. Mediation analyses reveal that LLM explanatory richness partially supports confidence by fostering users' reflective insight, whereas its effect on knowledge gain operates entirely through users' cognitive engagement. Moderation analyses show that these effects are highly conditional and vary by political efficacy. Confidence gains depend on how high-efficacy users experience and resolve uncertainty. Knowledge gains depend on high-efficacy users' ability to leverage extended interaction, with longer conversations benefiting primarily reflective users. In summary, we find that learning from LLMs is an interactional achievement, not a uniform outcome of better explanations. The findings underscore the importance of aligning LLM explanatory behavior with users' engagement states to support effective learning in designing Human-AI interactive systems.

Multi-hop kinship reasoning is a natural testbed for LLM compositionality, but existing benchmarks (notably CLUTRR) cover only the descriptive Eskimo system. We introduce KinshipQA, a procedurally-generated benchmark covering seven anthropologically-documented kinship systems (Eskimo, Sudanese, Hawaiian, Iroquois, Dravidian, Crow, Omaha) and up to six reasoning hops, with a tunable simulator horizon that eliminates exact-instance pretraining overlap. Evaluating six LLMs, we find a 40.9% accuracy drop when reasoning shifts from biological multi-hop to culturally-marked classification on the five non-descriptive systems. The drop holds for every non-descriptive system and is largest for the two skewing systems (Crow, Omaha), persists under chain-of-thought and few-shot prompting, and compounds with depth: at 5–6 hops cultural override falls to 10.6% while biological composition over the same chains remains at 58.6%. Under identical rule access humans reach 89.0% versus 50.7% for LLMs, so the questions are reliably solvable once the rule is supplied. Two follow-up experiments suggest distinct contributors. A fictional-rule control swapping system labels and kin terms for invented strings raises accuracy by 6.1%, implicating familiar English surface forms. An in-context-rule probe prepending the override rule helps skewing systems (+17.1%) but hurts non-skewing systems whose baseline already exceeds about 60% (-13.4%), consistent with a missing skewing prior alongside rule interference where the model already has a working approximation. Our code and data are publicly available on GitHub.

LLM agents operating over massive, dynamic tool libraries rely on effective retrieval, yet standard single-shot dense retrievers struggle with complex requests. These failures primarily stem from the disconnect between abstract user goals and technical documentation, and the limited capacity of fixed-size embeddings to model combinatorial tool compositions. To address these challenges, we propose TOOLQP, a lightweight framework that models retrieval as iterative query planning. Instead of single-shot matching, TOOLQP decomposes instructions into sub-tasks and dynamically generates queries to interact with the retriever, effectively bridging the semantic gap by targeting the specific sub-tasks required for composition. We train TOOLQP using synthetic query trajectories followed by optimization via Reinforcement Learning with Verifiable Rewards (RLVR). Experiments demonstrate that TOOLQP achieves state-of-the-art performance, exhibiting superior zero-shot generalization, robustness across diverse retrievers, and significant improvements in downstream agentic execution.

While Chain-of-Thought (CoT) prompting advances LLM reasoning, challenges persist in consistency, accuracy, and self-correction, especially for complex or ethically sensitive tasks. Existing single-dimensional reflection methods offer insufficient improvements. We propose MyGO Poly-Reflective Chain-of-Thought (PR-CoT), a novel methodology employing structured multi-perspective reflection. After initial CoT, PR-CoT guides the LLM to self-assess its reasoning across multiple predefined angles: logical consistency, information completeness, biases/ethics, and alternative solutions. Implemented purely via prompt engineering, this process refines the initial CoT into a more robust and accurate final answer without model retraining. Experiments across arithmetic, commonsense, ethical decision-making, and logical puzzles, using GPT-three point five and GPT-four models, demonstrate PR-CoT's superior performance. It significantly outperforms traditional CoT and existing reflection methods in logical consistency and error correction, with notable gains in nuanced domains like ethical decision-making. Ablation studies, human evaluations, and qualitative analyses further validate the contribution of each reflection perspective and the overall efficacy of our poly-reflective paradigm in fostering more reliable LLM reasoning.

While Vision-Language Models (VLMs) have significantly advanced Computer-Using Agents (CUAs), current frameworks struggle with robustness in long-horizon workflows and generalization in novel domains. These limitations stem from a lack of granular control over historical visual context curation and the absence of visual-aware tutorial retrieval. To bridge these gaps, we introduce OS-Symphony, a holistic framework that comprises an Orchestrator coordinating two key innovations for robust automation: (1) a Reflection-Memory Agent that utilizes milestone-driven long-term memory to enable trajectory-level self-correction, effectively mitigating visual context loss in long-horizon tasks; (2) Versatile Tool Agents featuring a Multimodal Searcher that adopts a SeeAct paradigm to navigate a browser-based sandbox to synthesize live, visually aligned tutorials, thereby resolving fidelity issues in unseen scenarios. Experimental results demonstrate that OS-Symphony delivers substantial performance gains across varying model scales, establishing new state-of-the-art results on three online benchmarks, notably achieving 65.84% on OSWorld.

Large Language Models (LLMs) can produce surprisingly sophisticated estimates of their own uncertainty. However, it remains unclear to what extent this expressed confidence is tied to the reasoning, knowledge, or decision making of the model. To test this, we introduce $\textbf{RiskEval}$: a framework designed to evaluate whether models adjust their abstention policies in response to varying error penalties. Our evaluation of several frontier models reveals a critical dissociation: models are neither cost-aware when articulating their verbal confidence, nor strategically responsive when deciding whether to engage or abstain under high-penalty conditions. Even when extreme penalties render frequent abstention the mathematically optimal strategy, models almost never abstain, resulting in utility collapse. This indicates that calibrated verbal confidence scores may not be sufficient to create trustworthy and interpretable AI systems, as current models lack the strategic agency to convert uncertainty signals into optimal and risk-sensitive decisions.

Understanding narratives requires identifying which events are most salient for a story's progression. We present a contrastive learning framework for modeling narrative salience that learns story embeddings from narrative twins: stories that share the same plot but differ in surface form. Our model is trained to distinguish a story from both its narrative twin and a distractor with similar surface features but different plot. Using the resulting embeddings, we evaluate four narratologically motivated operations for inferring salience (deletion, shifting, disruption, and summarization). Experiments on short narratives from the ROCStories corpus and longer Wikipedia plot summaries show that contrastively learned story embeddings outperform a masked-language-model baseline, and that summarization is the most reliable operation for identifying salient sentences. If narrative twins are not available, random dropout can be used to generate the twins from a single story. Effective distractors can be obtained either by prompting LLMs or, in long-form narratives, by using different parts of the same story.

Numerical reasoning is an important task in the analysis of financial documents. It helps in understanding and performing numerical predictions with logical conclusions for the given query seeking answers from financial texts. Recently, Large Language Models (LLMs) have shown promising results in multiple Question-Answering (Q-A) systems with the capability of logical reasoning. As documents related to finance often consist of long and complex financial contexts, LLMs appear well-suited for building high-quality automated financial question-answering systems. However, LLMs often face challenges in accurately processing the various numbers within financial reports. Extracting numerical data from unstructured text and semi-structured tables, and reliably performing accurate calculations, remains a significant bottleneck for numerical reasoning in most state-of-the-art LLMs. Recent studies have shown that structured data augmentations, such as Knowledge Graphs (KGs), have notably improved the predictions of LLMs along with logical explanations. Thus, it is an important requirement to consider inherent structured information in financial reports while using LLMs for various financial analytics. This paper proposes a framework to incorporate structured information using KGs along with LLM predictions for numerical reasoning tasks. The KGs are extracted using a proposed schema inherently from the document under processing. We evaluated our proposed framework over the benchmark data FinQA, using an open-source LLM, namely Llama 3.1 8B Instruct. We observed that the proposed framework improved execution accuracy by approximately 12% relative to the vanilla LLM.