Large language models are capable of leveraging both contextual and parametric knowledge but how they prioritize and integrate these sources remains underexplored. We introduce CoPE, a novel evaluation framework that systematically measures contextual knowledge (CK) and parametric knowledge (PK) across models and languages. Using our MultiWikiAtomic dataset in English, Spanish, and Danish, we analyze how large language models (LLMs) integrate context, prioritize information, and incorporate PK in open-ended question answering. Our analysis uncovers a phenomenon we call lost-in-the-later, where LLMs tend to overlook or deprioritize information that appears later in a given context, revealing a strong positional bias that affects contextual grounding. We further find that reasoning models, as well as non-reasoning models prompted with chain-of-thought (CoT), use context even less than non-reasoning models without CoT and fail to mitigate the lost-in-the-later effect. CoT prompting, in particular, results in lower recall and shorter responses, leading to degraded contextual grounding. Based on these insights, we design prompt-based methods to effectively leverage input context. A case study applying CoPE to summarization demonstrates that CK-informed prompting improves factual grounding and reduces hallucination.

Recent advances in large language models (LLMs) have enabled new applications in e-commerce customer service. However, their capabilities remain constrained in complex, multimodal scenarios. We present MindFlow, the first open-source multimodal LLM agent tailored for e-commerce. Built on the CoALA framework, it integrates memory, decision-making, and action modules, and adopts a modular "MLLM-as-Tool" strategy for effect visual-textual reasoning. Evaluated via online A/B testing and simulation-based ablation, MindFlow demonstrates substantial gains in handling complex queries, improving user satisfaction, and reducing operational costs, with a 93.53% relative improvement observed in real-world deployments.

The integration of Large Language Models (LLMs) into browser extensions has revolutionized web browsing, enabling sophisticated functionalities like content summarization, intelligent translation, and context-aware writing assistance. However, these AI-powered extensions introduce unprecedented challenges in testing and reliability assurance. Traditional browser extension testing approaches fail to address the non-deterministic behavior, context-sensitivity, and complex web environment integration inherent to LLM-powered extensions. Similarly, existing LLM testing methodologies operate in isolation from browser-specific contexts, creating a critical gap in effective evaluation frameworks. To bridge this gap, we present ASSURE, a modular automated testing framework specifically designed for AI-powered browser extensions. ASSURE comprises three principal components: (1) a modular test case generation engine that supports plugin-based extension of testing scenarios, (2) an automated execution framework that orchestrates the complex interactions between web content, extension processing, and AI model behavior, and (3) a configurable validation pipeline that systematically evaluates behavioral consistency and security invariants rather than relying on exact output matching. Our evaluation across six widely-used AI browser extensions demonstrates ASSURE's effectiveness, identifying 531 distinct issues spanning security vulnerabilities, metamorphic relation violations, and content alignment problems. ASSURE achieves 6.4x improved testing throughput compared to manual approaches, detecting critical security vulnerabilities within 12.4 minutes on average. This efficiency makes ASSURE practical for integration into development pipelines, offering a comprehensive solution to the unique challenges of testing AI-powered browser extensions.

Large, high-quality annotated corpora remain scarce in document-level entity and relation extraction in zero-shot or few-shot settings. In this paper, we present a fully automatic, LLM-based pipeline for synthetic data generation and in-context learning for document-level entity and relation extraction. In contrast to existing approaches that rely on manually annotated demonstrations or direct zero-shot inference, our method combines synthetic data generation with retrieval-based in-context learning, using a reasoning-optimized language model. This allows us to build a high-quality demonstration database without manual annotation and to dynamically retrieve relevant examples at inference time. Based on our approach we produce a synthetic dataset of over $5k$ Wikipedia abstracts with approximately $59k$ entities and $30k$ relation triples. Finally, we evaluate in-context learning performance on the DocIE shared task, extracting entities and relations from long documents in a zero-shot setting. We find that in-context joint entity and relation extraction at document-level remains a challenging task, even for state-of-the-art large language models.

The emergence of Large Language Models (LLMs) with increasingly sophisticated natural language understanding and generative capabilities has sparked interest in the Agent-based Modelling (ABM) community. With their ability to summarize, generate, analyze, categorize, transcribe and translate text, answer questions, propose explanations, sustain dialogue, extract information from unstructured text, and perform logical reasoning and problem-solving tasks, LLMs have a good potential to contribute to the modelling process. After reviewing the current use of LLMs in ABM, this study reflects on the opportunities and challenges of the potential use of LLMs in ABM. It does so by following the modelling cycle, from problem formulation to documentation and communication of model results, and holding a critical stance.

LLM-powered applications are highly susceptible to the quality of user prompts, and crafting high-quality prompts can often be challenging especially for domain-specific applications. This paper presents a novel dynamic context-aware prompt recommendation system for domain-specific AI applications. Our solution combines contextual query analysis, retrieval-augmented knowledge grounding, hierarchical skill organization, and adaptive skill ranking to generate relevant and actionable prompt suggestions. The system leverages behavioral telemetry and a two-stage hierarchical reasoning process to dynamically select and rank relevant skills, and synthesizes prompts using both predefined and adaptive templates enhanced with few-shot learning. Experiments on real-world datasets demonstrate that our approach achieves high usefulness and relevance, as validated by both automated and expert evaluations.

We prove that perfect hallucination control in large language models is mathematically impossible. No LLM inference mechanism can simultaneously achieve truthful response generation, semantic information conservation, relevant knowledge revelation, and knowledge-constrained optimality. This impossibility is fundamental, arising from the mathematical structure of information aggregation itself rather than engineering limitations. The proof spans three mathematical frameworks: auction theory, proper scoring theory for probabilistic predictions, and log-sum-exp analysis for transformer architectures. In each setting, we demonstrate that information aggregation creates unavoidable violations of conservation principles. The Jensen gap in transformer probability aggregation provides a direct measure of this impossibility. These results reframe hallucination from an engineering bug to an inevitable mathematical feature of distributed intelligence. There are fundamental trade-offs between truthfulness, knowledge utilization, and response completeness, providing principled foundations for managing rather than eliminating hallucination. This work reveals deep connections between neural network inference, philosophy of knowledge and reasoning, and classical results in game theory and information theory, opening new research directions for developing beneficial AI systems within mathematical constraints.

A simple yet effective method for inference-time alignment of generative models is Best-of-$N$ (BoN), where $N$ outcomes are sampled from a reference policy, evaluated using a proxy reward model, and the highest-scoring one is selected. While prior work argues that BoN is almost optimal in reward vs KL tradeoffs, the effectiveness of BoN depends critically on the quality of the proxy reward model used for selection. For this purpose, we study BoN through a smooth version known as Soft Best-of-N (SBoN) and develop a theoretical framework to address this gap. We analyze the scaling behaviour of BoN by providing bounds on the KL divergence between the SBoN policy and the reference policy, offering insights into how performance varies with the number of samples. We also study the regret gap, i.e., the gap between the expected true reward under the optimal policy and the SBoN policy. Our theoretical and empirical findings show that smoothing helps SBoN mitigate reward overoptimization, especially when the quality of the proxy reward is low.

Despite the remarkable performance of Large Language Models (LLMs) in automated discharge summary generation, they still suffer from hallucination issues, such as generating inaccurate content or fabricating information without valid sources. In addition, electronic medical records (EMRs) typically consist of long-form data, making it challenging for LLMs to attribute the generated content to the sources. To address these challenges, we propose LCDS, a Logic-Controlled Discharge Summary generation system. LCDS constructs a source mapping table by calculating textual similarity between EMRs and discharge summaries to constrain the scope of summarized content. Moreover, LCDS incorporates a comprehensive set of logical rules, enabling it to generate more reliable silver discharge summaries tailored to different clinical fields. Furthermore, LCDS supports source attribution for generated content, allowing experts to efficiently review, provide feedback, and rectify errors. The resulting golden discharge summaries are subsequently recorded for incremental fine-tuning of LLMs. Our project and demo video are in the GitHub repository https://github.com/ycycyc02/LCDS.

Biomedical semantic question answering rooted in information retrieval can play a crucial role in keeping up to date with vast, rapidly evolving and ever-growing biomedical literature. A robust system can help researchers, healthcare professionals and even layman users access relevant knowledge grounded in evidence. The BioASQ 2025 Task13b Challenge serves as an important benchmark, offering a competitive platform for advancement of this space. This paper presents the methodologies and results from our participation in this challenge where we built a Retrieval-Augmented Generation (RAG) system that can answer biomedical questions by retrieving relevant PubMed documents and snippets to generate answers. For the retrieval task, we generated dense embeddings from biomedical articles for initial retrieval, and applied an ensemble of finetuned cross-encoders and large language models (LLMs) for re-ranking to identify top relevant documents. Our solution achieved an MAP@10 of 0.1581, placing 10th on the leaderboard for the retrieval task. For answer generation, we employed few-shot prompting of instruction-tuned LLMs. Our system achieved macro-F1 score of 0.95 for yes/no questions (rank 12), Mean Reciprocal Rank (MRR) of 0.64 for factoid questions (rank 1), mean-F1 score of 0.63 for list questions (rank 5), and ROUGE-SU4 F1 score of 0.29 for ideal answers (rank 11).