As large language models (LLMs) evolve from conversational assistants into autonomous agents, evaluating the safety of their actions becomes critical. Prior safety benchmarks have primarily focused on preventing generation of harmful content, such as toxic text. However, they overlook the challenge of agents taking harmful actions when the most effective path to an operational goal conflicts with human safety. To address this gap, we introduce ManagerBench, a benchmark that evaluates LLM decision-making in realistic, human-validated managerial scenarios. Each scenario forces a choice between a pragmatic but harmful action that achieves an operational goal, and a safe action that leads to worse operational performance. A parallel control set, where potential harm is directed only at inanimate objects, measures a model's pragmatism and identifies its tendency to be overly safe. Our findings indicate that the frontier LLMs perform poorly when navigating this safety-pragmatism trade-off. Many consistently choose harmful options to advance their operational goals, while others avoid harm only to become overly safe and ineffective. Critically, we find this misalignment does not stem from an inability to perceive harm, as models' harm assessments align with human judgments, but from flawed prioritization. ManagerBench is a challenging benchmark for a core component of agentic behavior: making safe choices when operational goals and alignment values incentivize conflicting actions. Benchmark & code available at https://github.com/technion-cs-nlp/ManagerBench.

Trained on internet-scale video data, generative world models are increasingly recognized as powerful world simulators that can generate consistent and plausible dynamics over structure, motion, and physics. This raises a natural question: with the advent of strong video foundational models, might they supplant conventional vision encoder paradigms for general-purpose multimodal understanding? While recent studies have begun to explore the potential of world models on common vision tasks, these explorations typically lack a systematic investigation of generic, multimodal tasks. In this work, we strive to investigate the capabilities when world model priors are transferred into Vision-Language Models: we re-purpose a video diffusion model as a generative encoder to perform a single denoising step and treat the resulting latents as a set of visual embedding. We empirically investigate this class of models, which we refer to as World-Language Models (WorldLMs), and we find that generative encoders can capture latents useful for downstream understanding that show distinctions from conventional encoders. Naming our best-performing variant Dynamic Vision Aligner (DyVA), we further discover that this method significantly enhances spatial reasoning abilities and enables single-image models to perform multi-frame reasoning. Through the curation of a suite of visual reasoning tasks, we find DyVA to surpass both open-source and proprietary baselines, achieving state-of-the-art or comparable performance. We attribute these gains to WorldLM's inherited motion-consistency internalization from video pre-training. Finally, we systematically explore extensive model designs to highlight promising directions for future work. We hope our study can pave the way for a new family of VLMs that leverage priors from world models and are on a promising path towards generalist vision learners.

We study LLM routing, the problem of selecting the best model for each query while balancing user satisfaction, model expertise, and inference cost. We formulate routing as contextual dueling bandits, learning from pairwise preference feedback rather than absolute scores, thereby yielding label-efficient and dynamic adaptation. Building on this formulation, we introduce Category-Calibrated Fine-Tuning (CCFT), a representation-learning method that derives model embeddings from offline data using contrastive fine-tuning with categorical weighting. These embeddings enable the practical instantiation of Feel-Good Thompson Sampling for Contextual Dueling Bandits (FGTS.CDB), a theoretically grounded posterior-sampling algorithm. We propose four variants of the categorical weighting that explicitly integrate model quality and cost, and we empirically evaluate the proposed methods on the RouterBench and MixInstruct datasets. Across both benchmarks, our methods achieve lower cumulative regret and faster convergence, with better robustness and performance-cost balance than strong baselines built with a general-purpose OpenAI embedding model. The potential of large language models (LLMs) is so great that they have become a necessary part of daily life, with applications ranging from office assistance and fashion/dining suggestions to entertainment. LLM routing refers to a problem of dynamically selecting the most suitable LLM from a set of candidates for each query in a sequence of questions. Before the emergence of a universally dominant and affordable foundation model, routing is important because the choice of LLM must align with user traits, model expertise, and cost. To balance these three key factors, cascading algorithms such as FrugalGPT (Chen et al., 2024) and AutoMix (Aggarwal et al., 2024) were first proposed. The idea is to query a cheaper model first and advance the query to a more expensive one if the current response is unlikely to meet the user's expectation. A drawback of cascading is the accumulated cost and latency caused by calling multiple LLM candidates to generate the final response.

If two experts disagree on a test, we may conclude both cannot be 100 per cent correct. But if they completely agree, no possible evaluation can be excluded. This asymmetry in the utility of agreements versus disagreements is explored here by formalizing a logic of unsupervised evaluation for classifiers. Its core problem is computing the set of group evaluations that are logically consistent with how we observe them agreeing and disagreeing in their decisions. Statistical summaries of their aligned decisions are inputs into a Linear Programming problem in the integer space of possible correct or incorrect responses given true labels. Obvious logical constraints, such as, the number of correct responses cannot exceed the number of observed responses, are inequalities. But in addition, there are axioms, universally applicable linear equalities that apply to all finite tests. The practical and immediate utility of this approach to unsupervised evaluation using only logical consistency is demonstrated by building no-knowledge alarms that can detect when one or more LLMs-as-Judges are violating a minimum grading threshold specified by the user.

Robotic fabric manipulation in garment production for sewing, cutting, and ironing requires reliable flattening and alignment, yet remains challenging due to fabric deformability, effectively infinite degrees of freedom, and frequent occlusions from wrinkles, folds, and the manipulator's End-Effector (EE) and arm. To address these issues, this paper proposes the first Random-to-Target Fabric Flattening (RTFF) policy, which aligns a random wrinkled fabric state to an arbitrary wrinkle-free target state. The proposed policy adopts a hybrid Imitation Learning-Visual Servoing (IL-VS) framework, where IL learns with explicit fabric models for coarse alignment of the wrinkled fabric toward a wrinkle-free state near the target, and VS ensures fine alignment to the target. Central to this framework is a template-based mesh that offers precise target state representation, wrinkle-aware geometry prediction, and consistent vertex correspondence across RTFF manipulation steps, enabling robust manipulation and seamless IL-VS switching. Leveraging the power of mesh, a novel IL solution for RTFF-Mesh Action Chunking Transformer (MACT)-is then proposed by conditioning the mesh information into a Transformer-based policy. The RTFF policy is validated on a real dual-arm tele-operation system, showing zero-shot alignment to different targets, high accuracy, and strong generalization across fabrics and scales. Project website: https://kaitang98.github.io/RTFF_Policy/

We investigate how to adapt small, efficient LLMs to Faroese, a low-resource North Germanic language. Starting from English models, we continue pre-training on related Scandinavian languages, either individually or combined via merging, before fine-tuning on Faroese. We compare full fine-tuning with parameter-efficient tuning using LoRA, evaluating their impact on both linguistic accuracy and text comprehension. Due to the lack of existing Faroese evaluation data, we construct two new minimal-pair benchmarks from adapted and newly collected datasets and complement them with human evaluations by Faroese linguists. Our results demonstrate that transfer from related languages is crucial, though the optimal source language depends on the task: Icelandic enhances linguistic accuracy, whereas Danish boosts comprehension. Similarly, the choice between full fine-tuning and LoRA is task-dependent: LoRA improves linguistic acceptability and slightly increases human evaluation scores on the base model, while full fine-tuning yields stronger comprehension performance and better preserves model capabilities during downstream fine-tuning.

Audio descriptions (ADs) narrate important visual details in movies, enabling Blind and Low Vision (BLV) users to understand narratives and appreciate visual details. Existing works in automatic AD generation mostly focus on few-second trimmed clips, and evaluate them by comparing against a single ground-truth reference AD. However, writing ADs is inherently subjective. Through alignment and analysis of two independent AD tracks for the same movies, we quantify the subjectivity in when and whether to describe, and what and how to highlight. Thus, we show that working with trimmed clips is inadequate. We propose ADQA, a QA benchmark that evaluates ADs at the level of few-minute long, coherent video segments, testing whether they would help BLV users understand the story and appreciate visual details. ADQA features visual appreciation (VA) questions about visual facts and narrative understanding (NU) questions based on the plot. Through ADQA, we show that current AD generation methods lag far behind human-authored ADs. We conclude with several recommendations for future work and introduce a public leaderboard for benchmarking.

Building facades represent a significant untapped resource for solar energy generation in dense urban environments, yet assessing their photovoltaic (PV) potential remains challenging due to complex geometries and semantic com ponents. This study introduces SF-SPA (Semantic Facade Solar-PV Assessment), an automated framework that transforms street-view photographs into quantitative PV deployment assessments. The approach combines com puter vision and artificial intelligence techniques to address three key challenges: perspective distortion correction, semantic understanding of facade elements, and spatial reasoning for PV layout optimization. Our four-stage pipeline processes images through geometric rectification, zero-shot semantic segmentation, Large Language Model (LLM) guided spatial reasoning, and energy simulation. Validation across 80 buildings in four countries demonstrates ro bust performance with mean area estimation errors of 6.2% ± 2.8% compared to expert annotations. The auto mated assessment requires approximately 100 seconds per building, a substantial gain in efficiency over manual methods. Simulated energy yield predictions confirm the method's reliability and applicability for regional poten tial studies, urban energy planning, and building-integrated photovoltaic (BIPV) deployment. Code is available at: https:github.com/CodeAXu/Solar-PV-Installation

The emergence of agent-based systems represents a significant advancement in artificial intelligence, with growing applications in automated data extraction. However, chemical information extraction remains a formidable challenge due to the inherent heterogeneity of chemical data. Current agent-based approaches, both general-purpose and domain-specific, exhibit limited performance in this domain. To address this gap, we present ChemX, a comprehensive collection of 10 manually curated and domain-expert-validated datasets focusing on nanomaterials and small molecules. These datasets are designed to rigorously evaluate and enhance automated extraction methodologies in chemistry. To demonstrate their utility, we conduct an extensive benchmarking study comparing existing state-of-the-art agentic systems such as ChatGPT Agent and chemical-specific data extraction agents. Additionally, we introduce our own single-agent approach that enables precise control over document preprocessing prior to extraction. We further evaluate the performance of modern baselines, such as GPT-5 and GPT-5 Thinking, to compare their capabilities with agentic approaches. Our empirical findings reveal persistent challenges in chemical information extraction, particularly in processing domain-specific terminology, complex tabular and schematic representations, and context-dependent ambiguities. The ChemX benchmark serves as a critical resource for advancing automated information extraction in chemistry, challenging the generalization capabilities of existing methods, and providing valuable insights into effective evaluation strategies.

This study explores the integration of AI, particularly large language models (LLMs) like ChatGPT, into educational settings, focusing on the implications for teaching and learning. Through interviews with course coordinators from data science courses at Wageningen University, this research identifies both the benefits and challenges associated with AI in the classroom. While AI tools can streamline tasks and enhance learning, concerns arise regarding students' overreliance on these technologies, potentially hindering the development of essential cognitive and problem solving skills. The study highlights the importance of responsible AI usage, ethical considerations, and the need for adapting assessment methods to ensure educational outcomes are met. With careful integration, AI can be a valuable asset in education, provided it is used to complement rather than replace fundamental learning processes.