Patients awaiting invasive procedures often have unanswered pre-procedural questions; however, time-pressured workflows and privacy constraints limit personalized counseling. We present LENOHA (Low Energy, No Hallucination, Leave No One Behind Architecture), a safety-first, local-first system that routes inputs with a high-precision sentence-transformer classifier and returns verbatim answers from a clinician-curated FAQ for clinical queries, eliminating free-text generation in the clinical path. We evaluated two domains (tooth extraction and gastroscopy) using expert-reviewed validation sets (n=400/domain) for thresholding and independent test sets (n=200/domain). Among the four encoders, E5-large-instruct (560M) achieved an overall accuracy of 0.983 (95% CI 0.964-0.991), AUC 0.996, and seven total errors, which were statistically indistinguishable from GPT-4o on this task; Gemini made no errors on this test set. Energy logging shows that the non-generative clinical path consumes ~1.0 mWh per input versus ~168 mWh per small-talk reply from a local 8B SLM, a ~170x difference, while maintaining ~0.10 s latency on a single on-prem GPU. These results indicate that near-frontier discrimination and generation-induced errors are structurally avoided in the clinical path by returning vetted FAQ answers verbatim, supporting privacy, sustainability, and equitable deployment in bandwidth-limited environments.

Computer-Use Agents (CUAs) are an increasingly deployed class of agents that take actions on GUIs to accomplish user goals. In this paper, we show that CUAs consistently exhibit Blind Goal-Directedness (BGD): a bias to pursue goals regardless of feasibility, safety, reliability, or context. We characterize three prevalent patterns of BGD: (i) lack of contextual reasoning, (ii) assumptions and decisions under ambiguity, and (iii) contradictory or infeasible goals. We develop BLIND-ACT, a benchmark of 90 tasks capturing these three patterns. Built on OSWorld, BLIND-ACT provides realistic environments and employs LLM-based judges to evaluate agent behavior, achieving 93.75% agreement with human annotations. We use BLIND-ACT to evaluate nine frontier models, including Claude Sonnet and Opus 4, Computer-Use-Preview, and GPT-5, observing high average BGD rates (80.8%) across them. We show that BGD exposes subtle risks that arise even when inputs are not directly harmful. While prompting-based interventions lower BGD levels, substantial risk persists, highlighting the need for stronger training- or inference-time interventions. Qualitative analysis reveals observed failure modes: execution-first bias (focusing on how to act over whether to act), thought-action disconnect (execution diverging from reasoning), and request-primacy (justifying actions due to user request). Identifying BGD and introducing BLIND-ACT establishes a foundation for future research on studying and mitigating this fundamental risk and ensuring safe CUA deployment.

This study demonstrates that GuruAgents, prompt-guided AI agents, can systematically operationalize the strategies of legendary investment gurus. We develop five distinct GuruAgents, each designed to emulate an iconic investor, by encoding their distinct philosophies into LLM prompts that integrate financial tools and a deterministic reasoning pipeline. In a backtest on NASDAQ-100 constituents from Q4 2023 to Q2 2025, the GuruAgents exhibit unique behaviors driven by their prompted personas. The Buffett GuruAgent achieves the highest performance, delivering a 42.2\% CAGR that significantly outperforms benchmarks, while other agents show varied results. These findings confirm that prompt engineering can successfully translate the qualitative philosophies of investment gurus into reproducible, quantitative strategies, highlighting a novel direction for automated systematic investing. The source code and data are available at https://github.com/yejining99/GuruAgents.

Multimodal Dialogue Summarization (MDS) is a critical task with wide-ranging applications. To support the development of effective MDS models, robust automatic evaluation methods are essential for reducing both cost and human effort. However, such methods require a strong meta-evaluation benchmark grounded in human annotations. In this work, we introduce MDSEval, the first meta-evaluation benchmark for MDS, consisting image-sharing dialogues, corresponding summaries, and human judgments across eight well-defined quality aspects. To ensure data quality and richfulness, we propose a novel filtering framework leveraging Mutually Exclusive Key Information (MEKI) across modalities. Our work is the first to identify and formalize key evaluation dimensions specific to MDS. We benchmark state-of-the-art modal evaluation methods, revealing their limitations in distinguishing summaries from advanced MLLMs and their susceptibility to various bias.

Autoregressive Large Language Models (LLMs) demonstrate exceptional performance in language understanding and generation. However, their application in text embedding tasks has been relatively slow, along with the analysis of their semantic representation in probing tasks, due to the constraints of the unidirectional attention mechanism. This paper aims to explore whether such constraints can be overcome by enabling bidirectional attention in LLMs. We tested different variants of the Llama architecture through additional training steps, progressively enabling bidirectional attention and unsupervised/supervised contrastive learning.

Neural network pruning is a promising technique to mitigate the excessive computational and memory requirements of large language models (LLMs). Despite its promise, however, progress in this area has diminished, as conventional methods are seemingly unable to surpass moderate sparsity levels (50-60%) without severely degrading model accuracy. This work breaks through the current impasse, presenting a principled and effective method called $\texttt{Elsa}$, which achieves extreme sparsity levels of up to 90% while retaining high model fidelity. This is done by identifying several limitations in current practice, all of which can be traced back to their reliance on a surrogate objective formulation. $\texttt{Elsa}$ tackles this issue directly and effectively via standard and well-established constrained optimization techniques based on ADMM. Our extensive experiments across a wide range of models and scales show that $\texttt{Elsa}$ achieves substantial improvements over existing methods; e.g., it achieves 7.8$\times$ less perplexity than the best existing method on LLaMA-2-7B at 90% sparsity. Furthermore, we present $\texttt{Elsa}_{\text{-L}}$, a quantized variant that scales to extremely large models (27B), and establish its theoretical convergence guarantees. These results highlight meaningful progress in advancing the frontier of LLM sparsity, while promising that significant opportunities for further advancement may remain in directions that have so far attracted limited exploration.

The discourse on privacy risks in Large Language Models (LLMs) has disproportionately focused on verbatim memorization of training data, while a constellation of more immediate and scalable privacy threats remain underexplored. This position paper argues that the privacy landscape of LLM systems extends far beyond training data extraction, encompassing risks from data collection practices, inference-time context leakage, autonomous agent capabilities, and the democratization of surveillance through deep inference attacks. We present a comprehensive taxonomy of privacy risks across the LLM lifecycle -- from data collection through deployment -- and demonstrate through case studies how current privacy frameworks fail to address these multifaceted threats. Through a longitudinal analysis of 1,322 AI/ML privacy papers published at leading conferences over the past decade (2016--2025), we reveal that while memorization receives outsized attention in technical research, the most pressing privacy harms lie elsewhere, where current technical approaches offer little traction and viable paths forward remain unclear. We call for a fundamental shift in how the research community approaches LLM privacy, moving beyond the narrow focus of current technical solutions and embracing interdisciplinary approaches that address the sociotechnical nature of these emerging threats.

The quadratic complexity of softmax attention remains a central bottleneck in scaling large language models (LLMs). [Alman and Song, NeurIPS 2023] proposed a sub-quadratic attention approximation algorithm, but it works only under the restrictive bounded-entry assumption. Since this assumption rarely holds in practice, its applicability to modern LLMs is limited. In this paper, we introduce support-basis decomposition, a new framework for efficient attention approximation beyond bounded entries. We empirically demonstrate that the entries of the query and key matrices exhibit sub-Gaussian behavior. Our approach uses this property to split large and small entries, enabling exact computation on sparse components and polynomial approximation on dense components. We establish rigorous theoretical guarantees, proving a sub-quadratic runtime, and extend the method to a multi-threshold setting that eliminates all distributional assumptions. Furthermore, we provide the first theoretical justification for the empirical success of polynomial attention [Kacham, Mirrokni, and Zhong, ICML 2024], showing that softmax attention can be closely approximated by a combination of multiple polynomial attentions with sketching.

We explore the geospatial reasoning capabilities of Large Language Models (LLMs), specifically, whether LLMs can read road network maps and perform navigation. Using road network as context, our prompting framework enables LLMs to generate valid paths without accessing any external navigation tools. Experiments show that LLMs outperform off-the-shelf baselines and specialized trajectory recovery models, with strong zero-shot generalization. Fine-grained analysis shows that LLMs have strong comprehension of the road network and coordinate systems, but also pose systematic biases with respect to regions and transportation modes. Finally, we demonstrate how LLMs can enhance navigation experiences by reasoning over maps in flexible ways to incorporate user preferences. Large Language Models (LLMs) are increasingly recognized as general-purpose systems, showing strong performance across domains ranging from mathematics and coding to vision and robotics. An emerging yet underex-plored question is whether these models possess geospa-tial understanding, the ability to reason about maps, paths, and spatial relationships. Such capabilities are fundamental to many real-world applications, e.g., autonomous vehicle navigation, logistics, and urban planning. While prior work has studied LLMs in contexts such as geographic knowledge retrieval (Manvi et al., 2024a;b) and map-based multiple-choice question answering (Dihan et al., 2025), the ability of LLMs to read road networks and plan paths has not been systematically evaluated. We investigate whether LLMs can perform navigation through the trajectory recovery task: reconstructing masked segments of GPS traces from the road network context, to bypass the restriction of relying on shortest path-type of ground truth which may not reflect human navigation pattern in practice (Golledge, 1995; Duckham & Kulik, 2003). Our dataset is framed in away that is harder than the traditional point-wise trajectory recovery task (Newson & Krumm, 2009; Song et al., 2017; Si et al., 2024), and closer to the higher-level navigation problem.

Large Language Models are profoundly changing work patterns in high-risk professional domains, yet their application also introduces severe and underexplored compliance risks. To investigate this issue, we conducted semi-structured interviews with 24 highly-skilled knowledge workers from industries such as law, healthcare, and finance. The study found that these experts are commonly concerned about sensitive information leakage, intellectual property infringement, and uncertainty regarding the quality of model outputs. In response, they spontaneously adopt various mitigation strategies, such as actively distorting input data and limiting the details in their prompts. However, the effectiveness of these spontaneous efforts is limited due to a lack of specific compliance guidance and training for Large Language Models. Our research reveals a significant gap between current NLP tools and the actual compliance needs of experts. This paper positions these valuable empirical findings as foundational work for building the next generation of Human-Centered, Compliance-Driven Natural Language Processing for Regulatory Technology (RegTech), providing a critical human-centered perspective and design requirements for engineering NLP systems that can proactively support expert compliance workflows.