Document-level claim extraction remains an open challenge in the field of fact-checking, and subsequently, methods for evaluating extracted claims have received limited attention. In this work, we explore approaches to aligning two sets of claims pertaining to the same source document and computing their similarity through an alignment score. We investigate techniques to identify the best possible alignment and evaluation method between claim sets, with the aim of providing a reliable evaluation framework. Our approach enables comparison between model-extracted and human-annotated claim sets, serving as a metric for assessing the extraction performance of models and also as a possible measure of inter-annotator agreement. We conduct experiments on newly collected dataset--claims extracted from comments under Czech and Slovak news articles--domains that pose additional challenges due to the informal language, strong local context, and subtleties of these closely related languages. The results draw attention to the limitations of current evaluation approaches when applied to document-level claim extraction and highlight the need for more advanced methods--ones able to correctly capture semantic similarity and evaluate essential claim properties such as atomicity, checkworthiness, and decontextualization.

Constructed-response questions are crucial to encourage generative processing and test a learner's understanding of core concepts. However, the limited availability of instructor time, large class sizes, and other resource constraints pose significant challenges in providing timely and detailed evaluation, which is crucial for a holistic educational experience. In addition, providing timely and frequent assessments is challenging since manual grading is labor intensive, and automated grading is complex to generalize to every possible response scenario. This paper proposes a novel and practical approach to grade short-answer constructed-response questions. We discuss why this problem is challenging, define the nature of questions on which our method works, and finally propose a framework that instructors can use to evaluate their students' open-responses, utilizing near-domain data like data from similar questions administered in previous years. The proposed method outperforms the state of the art machine learning models as well as non-fine-tuned large language models like GPT 3.5, GPT 4, and GPT 4o by a considerable margin of over 10-20% in some cases, even after providing the LLMs with reference/model answers. Our framework does not require pre-written grading rubrics and is designed explicitly with practical classroom settings in mind. Our results also reveal exciting insights about learning from near-domain data, including what we term as accuracy and data advantages using human-labeled data, and we believe this is the first work to formalize the problem of automated short answer grading based on the near-domain data.

LLM-powered coding agents, which operate in iterative loops (turns) to solve software engineering tasks, are becoming increasingly powerful. However, their practical deployment is hindered by significant and unpredictable costs. This challenge arises from a combination of factors: quadratically growing token counts with each turn, the high price of models, the large number of turns required for real-world tasks, and the tendency of agents to take inefficient or unnecessary actions. While existing research focuses on optimizing individual turns, the strategic control of the total number of turns remains an underexplored area for managing agent performance and cost. To address this gap, we conduct a comprehensive empirical study on SWE-bench using three state-of-the-art models and evaluate the impact of three distinct turn-control strategies: an unrestricted baseline, a fixed-turn limit with reminders, and a novel dynamic-turn strategy that grants extensions on-demand. Our findings first reveal a fundamental trade-off in the unrestricted setting, where no single model excels across performance, cost, and turn efficiency. We then show that a fixed-turn limit, specifically at the 75th percentile of the baseline, serves as a "sweet spot", substantially reducing costs (by 24%-68%) with minimal impact on solve rates. Most significantly, the dynamic-turn strategy consistently outperforms fixed-limit approaches, achieving comparable or better solve rates while further reducing costs by an additional 12%-24% by intelligently allocating resources only to tasks that need them. This work provides the first systematic analysis of turn-control strategies, offering simple yet effective guidelines for developers to balance cost and efficacy. We demonstrate that dynamic resource allocation is a superior, easy-to-implement approach for deploying powerful yet economically viable coding agents.

Reliable handling of code diffs is central to agents that edit and refactor repositories at scale. We introduce Diff-XYZ, a compact benchmark for code-diff understanding with three supervised tasks: apply (old code $+$ diff $\rightarrow$ new code), anti-apply (new code $-$ diff $\rightarrow$ old code), and diff generation (new code $-$ old code $\rightarrow$ diff). Instances in the benchmark are triples $\langle \textit{old code}, \textit{new code}, \textit{diff} \rangle$ drawn from real commits in CommitPackFT, paired with automatic metrics and a clear evaluation protocol. We use the benchmark to do a focused empirical study of the unified diff format and run a cross-format comparison of different diff representations. Our findings reveal that different formats should be used depending on the use case and model size. For example, representing diffs in search-replace format performs best for larger models across most tasks, while structured udiff variants offer similar but slightly weaker performance. In contrast, smaller open models benefit little from any formatting choice. The Diff-XYZ benchmark is a reusable foundation for assessing and improving diff handling in LLMs that can aid future development of diff formats and models editing code. The dataset is published on HuggingFace Hub: https://huggingface.co/datasets/JetBrains-Research/diff-xyz.

Increasingly, more and more people are turning to large language models (LLMs) for healthcare advice and consultation, making it important to gauge the efficacy and accuracy of the responses of LLMs to such queries. While there are pre-existing medical benchmarks literature which seeks to accomplish this very task, these benchmarks are almost universally in English, which has led to a notable gap in existing literature pertaining to multilingual LLM evaluation. Within this work, we seek to aid in addressing this gap with Cancer-Myth-Indic, an Indic language benchmark built by translating a 500-item subset of Cancer-Myth, sampled evenly across its original categories, into five under-served but widely used languages from the subcontinent (500 per language; 2,500 translated items total). Native-speaker translators followed a style guide for preserving implicit presuppositions in translation; items feature false presuppositions relating to cancer. We evaluate several popular LLMs under this presupposition stress.

Aligning AI with human values is a pressing unsolved problem. To address the lack of quantitative metrics for value alignment, we propose EigenBench: a black-box method for comparatively benchmarking language models' values. Given an ensemble of models, a constitution describing a value system, and a dataset of scenarios, our method returns a vector of scores quantifying each model's alignment to the given constitution. To produce these scores, each model judges the outputs of other models across many scenarios, and these judgments are aggregated with EigenTrust (Kamvar et al., 2003), yielding scores that reflect a weighted consensus judgment of the whole ensemble. EigenBench uses no ground truth labels, as it is designed to quantify subjective traits for which reasonable judges may disagree on the correct label. Hence, to validate our method, we collect human judgments on the same ensemble of models and show that EigenBench's judgments align closely with those of human evaluators. We further demonstrate that EigenBench can recover model rankings on the GPQA benchmark without access to objective labels, supporting its viability as a framework for evaluating subjective values for which no ground truths exist.

The widespread deployment of LLM-based agents is likely to introduce a critical privacy threat: malicious agents that proactively engage others in multi-turn interactions to extract sensitive information. However, the evolving nature of such dynamic dialogues makes it challenging to anticipate emerging vulnerabilities and design effective defenses. To tackle this problem, we present a search-based framework that alternates between improving attack and defense strategies through the simulation of privacy-critical agent interactions. Specifically, we employ LLMs as optimizers to analyze simulation trajectories and iteratively propose new agent instructions. To explore the strategy space more efficiently, we further utilize parallel search with multiple threads and cross-thread propagation. Through this process, we find that attack strategies escalate from direct requests to sophisticated tactics, such as impersonation and consent forgery, while defenses evolve from simple rule-based constraints to robust identity-verification state machines. The discovered attacks and defenses transfer across diverse scenarios and backbone models, demonstrating strong practical utility for building privacy-aware agents.

The rise of misinformation underscores the need for scalable and reliable fact-checking solutions. Large language models (LLMs) hold promise in automating fact verification, yet their effectiveness across global contexts remains uncertain. We systematically evaluate nine established LLMs across multiple categories (open/closed-source, multiple sizes, diverse architectures, reasoning-based) using 5,000 claims previously assessed by 174 professional fact-checking organizations across 47 languages. Our methodology tests model generalizability on claims postdating training cutoffs and four prompting strategies mirroring both citizen and professional fact-checker interactions, with over 240,000 human annotations as ground truth. Findings reveal a concerning pattern resembling the Dunning-Kruger effect: smaller, accessible models show high confidence despite lower accuracy, while larger models demonstrate higher accuracy but lower confidence. This risks systemic bias in information verification, as resource-constrained organizations typically use smaller models. Performance gaps are most pronounced for non-English languages and claims originating from the Global South, threatening to widen existing information inequalities. These results establish a multilingual benchmark for future research and provide an evidence base for policy aimed at ensuring equitable access to trustworthy, AI-assisted fact-checking.

Large Language Models (LLMs) agents augmented with domain tools promise to autonomously execute complex tasks requiring human-level intelligence, such as customer service and digital assistance. However, their practical deployment is often limited by their low success rates under complex real-world environments. To tackle this, prior research has primarily focused on improving the agents themselves, such as developing strong agentic LLMs, while overlooking the role of the system environment in which the agent operates. In this paper, we study a complementary direction: improving agent success rates by optimizing the system environment in which the agent operates. We collect 142 agent traces (3,656 turns of agent-environment interactions) across 5 state-of-the-art agentic benchmarks. By analyzing these agent failures, we propose a taxonomy for agent-environment interaction failures that includes 6 failure modes. Guided by these findings, we design Aegis, a set of targeted environment optimizations: 1) environment observability enhancement, 2) common computation offloading, and 3) speculative agentic actions. These techniques improve agent success rates on average by 6.7-12.5%, without any modifications to the agent and underlying LLM.

The potential of large language models (LLMs) in specialized domains such as legal risk analysis remains underexplored. In response to growing interest in locally deploying open-source LLMs for legal tasks while preserving data confidentiality, this paper introduces ContractEval, the first benchmark to thoroughly evaluate whether open-source LLMs could match proprietary LLMs in identifying clause-level legal risks in commercial contracts. Using the Contract Understanding Atticus Dataset (CUAD), we assess 4 proprietary and 15 open-source LLMs. Our results highlight five key findings: (1) Proprietary models outperform open-source models in both correctness and output effectiveness, though some open-source models are competitive in certain specific dimensions. (2) Larger open-source models generally perform better, though the improvement slows down as models get bigger. (3) Reasoning ("thinking") mode improves output effectiveness but reduces correctness, likely due to over-complicating simpler tasks. (4) Open-source models generate "no related clause" responses more frequently even when relevant clauses are present. This suggests "laziness" in thinking or low confidence in extracting relevant content. (5) Model quantization speeds up inference but at the cost of performance drop, showing the tradeoff between efficiency and accuracy. These findings suggest that while most LLMs perform at a level comparable to junior legal assistants, open-source models require targeted fine-tuning to ensure correctness and effectiveness in high-stakes legal settings. ContractEval offers a solid benchmark to guide future development of legal-domain LLMs.