Multilingual language models (LMs) are expected to recall factual knowledge consistently across languages, yet they often fail to transfer knowledge between languages even when they possess the correct information in one of the languages. For example, we find that an LM may correctly identify Rashed Al Shashai as being from Saudi Arabia when asked in Arabic, but consistently fails to do so when asked in English or Swahili. To systematically investigate this limitation, we introduce a benchmark of 10,000 country-related facts across 13 languages and propose three novel metrics: Factual Recall Score, Knowledge Transferability Score, and Cross-Lingual Factual Knowledge Transferability Score-to quantify factual recall and knowledge transferability in LMs across different languages. Our results reveal fundamental weaknesses in today's state-of-the-art LMs, particularly in cross-lingual generalization where models fail to transfer knowledge effectively across different languages, leading to inconsistent performance sensitive to the language used. Our findings emphasize the need for LMs to recognize language-specific factual reliability and leverage the most trustworthy information across languages. We release our benchmark and evaluation framework to drive future research in multilingual knowledge transfer.

Purpose: To evaluate the accuracy and reasoning ability of DeepSeek-R1 and three other recently released large language models (LLMs) in bilingual complex ophthalmology cases. Methods: A total of 130 multiple-choice questions (MCQs) related to diagnosis (n = 39) and management (n = 91) were collected from the Chinese ophthalmology senior professional title examination and categorized into six topics. These MCQs were translated into English using DeepSeek-R1. The responses of DeepSeek-R1, Gemini 2.0 Pro, OpenAI o1 and o3-mini were generated under default configurations between February 15 and February 20, 2025. Accuracy was calculated as the proportion of correctly answered questions, with omissions and extra answers considered incorrect. Reasoning ability was evaluated through analyzing reasoning logic and the causes of reasoning error. Results: DeepSeek-R1 demonstrated the highest overall accuracy, achieving 0.862 in Chinese MCQs and 0.808 in English MCQs. Gemini 2.0 Pro, OpenAI o1, and OpenAI o3-mini attained accuracies of 0.715, 0.685, and 0.692 in Chinese MCQs (all P<0.001 compared with DeepSeek-R1), and 0.746 (P=0.115), 0.723 (P=0.027), and 0.577 (P<0.001) in English MCQs, respectively. DeepSeek-R1 achieved the highest accuracy across five topics in both Chinese and English MCQs. It also excelled in management questions conducted in Chinese (all P<0.05). Reasoning ability analysis showed that the four LLMs shared similar reasoning logic. Ignoring key positive history, ignoring key positive signs, misinterpretation medical data, and too aggressive were the most common causes of reasoning errors. Conclusion: DeepSeek-R1 demonstrated superior performance in bilingual complex ophthalmology reasoning tasks than three other state-of-the-art LLMs. While its clinical applicability remains challenging, it shows promise for supporting diagnosis and clinical decision-making.

Large Language Models have garnered significant attention for their capabilities in multilingual natural language processing, while studies on risks associated with cross biases are limited to immediate context preferences. Cross-language disparities in reasoning-based recommendations remain largely unexplored, with a lack of even descriptive analysis. This study is the first to address this gap. We test LLM's applicability and capability in providing personalized advice across three key scenarios: university applications, travel, and relocation. We investigate multilingual bias in state-of-the-art LLMs by analyzing their responses to decision-making tasks across multiple languages. We quantify bias in model-generated scores and assess the impact of demographic factors and reasoning strategies (e.g., Chain-of-Thought prompting) on bias patterns. Our findings reveal that local language bias is prevalent across different tasks, with GPT-4 and Sonnet reducing bias for English-speaking countries compared to GPT-3.5 but failing to achieve robust multilingual alignment, highlighting broader implications for multilingual AI agents and applications such as education.

Legal case documents play a critical role in judicial proceedings. As the number of cases continues to rise, the reliance on manual drafting of legal case documents is facing increasing pressure and challenges. The development of large language models (LLMs) offers a promising solution for automating document generation. However, existing benchmarks fail to fully capture the complexities involved in drafting legal case documents in real-world scenarios. To address this gap, we introduce CaseGen, the benchmark for multi-stage legal case documents generation in the Chinese legal domain. CaseGen is based on 500 real case samples annotated by legal experts and covers seven essential case sections. It supports four key tasks: drafting defense statements, writing trial facts, composing legal reasoning, and generating judgment results. To the best of our knowledge, CaseGen is the first benchmark designed to evaluate LLMs in the context of legal case document generation. To ensure an accurate and comprehensive evaluation, we design the LLM-as-a-judge evaluation framework and validate its effectiveness through human annotations. We evaluate several widely used general-domain LLMs and legal-specific LLMs, highlighting their limitations in case document generation and pinpointing areas for potential improvement. This work marks a step toward a more effective framework for automating legal case documents drafting, paving the way for the reliable application of AI in the legal field. The dataset and code are publicly available at https://github.com/CSHaitao/CaseGen.

Modern language models rely on static vocabularies, fixed before pretraining, in contrast to the adaptive vocabulary acquisition observed in human language learning. To bridge this gap, we introduce vocabulary curriculum learning, an approach that improves pretraining efficiency with log-linear scaling gains relative to vocabulary size. Our method alternates between entropy-guided vocabulary expansion and model optimization, enabling models to learn transferable representations across diverse tokenization granularities. This approach naturally gives rise to an optimal computation allocation pattern: longer tokens capture predictable content, while shorter tokens focus on more complex, harder-to-predict contexts. Experiments on small-scale GPT models demonstrate improved scaling efficiency, reinforcing the effectiveness of dynamic tokenization. We release our code to support further research and plan to extend our experiments to larger models and diverse domains.

With the proliferation of data across various domains, there is a critical demand for tools that enable non-experts to derive meaningful insights without deep data analysis skills. To address this need, existing automatic fact sheet generation tools offer heuristic-based solutions to extract facts and generate stories. However, they inadequately grasp the semantics of data and struggle to generate narratives that fully capture the semantics of the dataset or align the fact sheet with specific user needs. Addressing these shortcomings, this paper introduces \tool, a novel tool designed for the automatic generation and customisation of fact sheets. \tool applies the concept of collaborative AI workers to transform raw tabular dataset into comprehensive, visually compelling fact sheets. We define effective taxonomy to profile AI worker for specialised tasks. Furthermore, \tool empowers users to refine these fact sheets through intuitive natural language commands, ensuring the final outputs align closely with individual preferences and requirements. Our user evaluation with 18 participants confirms that \tool not only surpasses state-of-the-art baselines in automated fact sheet production but also provides a positive user experience during customization tasks.

Improvements in the area of large language models have shifted towards the construction of models capable of using external tools and interpreting their outputs. These so-called web agents have the ability to interact autonomously with the internet. This allows them to become powerful daily assistants handling time-consuming, repetitive tasks while supporting users in their daily activities. While web agent research is thriving, the sustainability aspect of this research direction remains largely unexplored. We provide an initial exploration of the energy and CO2 cost associated with web agents. Our results show how different philosophies in web agent creation can severely impact the associated expended energy. We highlight lacking transparency regarding the disclosure of model parameters and processes used for some web agents as a limiting factor when estimating energy consumption. As such, our work advocates a change in thinking when evaluating web agents, warranting dedicated metrics for energy consumption and sustainability.

Recent advances in multimodal ECG representation learning center on aligning ECG signals with paired free-text reports. However, suboptimal alignment persists due to the complexity of medical language and the reliance on a full 12-lead setup, which is often unavailable in under-resourced settings. To tackle these issues, we propose **K-MERL**, a knowledge-enhanced multimodal ECG representation learning framework. **K-MERL** leverages large language models to extract structured knowledge from free-text reports and employs a lead-aware ECG encoder with dynamic lead masking to accommodate arbitrary lead inputs. Evaluations on six external ECG datasets show that **K-MERL** achieves state-of-the-art performance in zero-shot classification and linear probing tasks, while delivering an average **16%** AUC improvement over existing methods in partial-lead zero-shot classification.

We present a comprehensive evaluation framework for assessing Large Language Models' (LLMs) capabilities in suicide prevention, focusing on two critical aspects: the Identification of Implicit Suicidal ideation (IIS) and the Provision of Appropriate Supportive responses (PAS). We introduce \ourdata, a novel dataset of 1,308 test cases built upon psychological frameworks including D/S-IAT and Negative Automatic Thinking, alongside real-world scenarios. Through extensive experiments with 8 widely used LLMs under different contextual settings, we find that current models struggle significantly with detecting implicit suicidal ideation and providing appropriate support, highlighting crucial limitations in applying LLMs to mental health contexts. Our findings underscore the need for more sophisticated approaches in developing and evaluating LLMs for sensitive psychological applications.

Automated planning is traditionally the domain of experts, utilized in fields like manufacturing and healthcare with the aid of expert planning tools. Recent advancements in LLMs have made planning more accessible to everyday users due to their potential to assist users with complex planning tasks. However, LLMs face several application challenges within end-user planning, including consistency, accuracy, and user trust issues. This paper introduces VeriPlan, a system that applies formal verification techniques, specifically model checking, to enhance the reliability and flexibility of LLMs for end-user planning. In addition to the LLM planner, VeriPlan includes three additional core features -- a rule translator, flexibility sliders, and a model checker -- that engage users in the verification process. Through a user study (n=12), we evaluate VeriPlan, demonstrating improvements in the perceived quality, usability, and user satisfaction of LLMs. Our work shows the effective integration of formal verification and user-control features with LLMs for end-user planning tasks.