Leaderboards showcase the current capabilities and limitations of Large Language Models (LLMs). To motivate the development of LLMs that represent the linguistic and cultural diversity of the Spanish-speaking community, we present La Leaderboard, the first open-source leaderboard to evaluate generative LLMs in languages and language varieties of Spain and Latin America. La Leaderboard is a community-driven project that aims to establish an evaluation standard for everyone interested in developing LLMs for the Spanish-speaking community. This initial version combines 66 datasets in Basque, Catalan, Galician, and different Spanish varieties, showcasing the evaluation results of 50 models. To encourage community-driven development of leaderboards in other languages, we explain our methodology, including guidance on selecting the most suitable evaluation setup for each downstream task. In particular, we provide a rationale for using fewer few-shot examples than typically found in the literature, aiming to reduce environmental impact and facilitate access to reproducible results for a broader research community.

As autonomous agents powered by large language models (LLMs) continue to demonstrate potential across various assistive tasks, ensuring their safe and reliable behavior is crucial for preventing unintended consequences. In this work, we introduce CIP, a novel technique that leverages causal influence diagrams (CIDs) to identify and mitigate risks arising from agent decision-making. CIDs provide a structured representation of cause-and-effect relationships, enabling agents to anticipate harmful outcomes and make safer decisions. Our approach consists of three key steps: (1) initializing a CID based on task specifications to outline the decision-making process, (2) guiding agent interactions with the environment using the CID, and (3) iteratively refining the CID based on observed behaviors and outcomes. Experimental results demonstrate that our method effectively enhances safety in both code execution and mobile device control tasks.

Recent Large Vision-Language Models (LVLMs) have introduced a new paradigm for understanding and reasoning about image input through textual responses. Although they have achieved remarkable performance across a range of multi-modal tasks, they face the persistent challenge of hallucination, which introduces practical weaknesses and raises concerns about their reliable deployment in real-world applications. Existing work has explored contrastive decoding approaches to mitigate this issue, where the output of the original LVLM is compared and contrasted with that of a perturbed version. However, these methods require two or more queries that slow down LVLM response generation, making them less suitable for real-time applications. T o overcome this limitation, we propose ONLY, a training-free decoding approach that requires only a single query and a one-layer intervention during decoding, enabling efficient real-time deployment. Specifically, we enhance textual outputs by selectively amplifying crucial textual information using a text-to-visual entropy ratio for each token. Extensive experimental results demonstrate that our proposed ONLY consistently outperforms state-of-the-art methods across various benchmarks while requiring minimal implementation effort and computational cost. Code is available at https://github.com/zifuwan/ONLY .

As multimodal language models advance, their application to 3D scene understanding is a fast-growing frontier, driving the development of 3D Vision-Language Models (VLMs). Current methods show strong dependence on object detectors, introducing processing bottlenecks and limitations in taxonomic flexibility. To address these limitations, we propose a scene-centric 3D VLM for 3D Gaussian splat scenes that employs language- and task-aware scene representations. Our approach directly embeds rich linguistic features into the 3D scene representation by associating language with each Gaussian primitive, achieving early modality alignment. To process the resulting dense representations, we introduce a dual sparsifier that distills them into compact, task-relevant tokens via task-guided and location-guided pathways, producing sparse, task-aware global and local scene tokens. Notably, we present the first Gaussian splatting-based VLM, leveraging photorealistic 3D representations derived from standard RGB images, demonstrating strong generalization: it improves performance of prior 3D VLM five folds, in out-of-the-domain settings.

Video Multimodal Large Language Models (V-MLLMs) have shown impressive capabilities in temporal reasoning and cross-modal understanding, yet their vulnerability to adversarial attacks remains underexplored due to unique challenges: complex cross-modal reasoning mechanisms, temporal dependencies, and computational constraints. We present CAVALRY-V (Cross-modal Language-Vision Adversarial Yielding for Videos), a novel framework that directly targets the critical interface between visual perception and language generation in V-MLLMs. Our approach introduces two key innovations: (1) a dual-objective semantic-visual loss function that simultaneously disrupts the model's text generation logits and visual representations to undermine cross-modal integration, and (2) a computationally efficient two-stage generator framework that combines large-scale pre-training for cross-model transferability with specialized fine-tuning for spatiotemporal coherence. Empirical evaluation on comprehensive video understanding benchmarks demonstrates that CAVALRY-V significantly outperforms existing attack methods, achieving 22.8% average improvement over the best baseline attacks on both commercial systems (GPT-4.1, Gemini 2.0) and open-source models (QwenVL-2.5, InternVL-2.5, Llava-Video, Aria, MiniCPM-o-2.6). Our framework achieves flexibility through implicit temporal coherence modeling rather than explicit regularization, enabling significant performance improvements even on image understanding (34.4% average gain). This capability demonstrates CAVALRY-V's potential as a foundational approach for adversarial research across multimodal systems.

Hallucinations are one of the major problems of LLMs, hindering their trustworthiness and deployment to wider use cases. However, most of the research on hallucinations focuses on English data, neglecting the multilingual nature of LLMs. This paper describes our submission to the SemEval-2025 Task-3 - Mu-SHROOM, the Multilingual Shared-task on Hallucinations and Related Observable Overgeneration Mistakes. We propose a two-part pipeline that combines retrieval-based fact verification against Wikipedia with a BERT-based system fine-tuned to identify common hallucination patterns. Our system achieves competitive results across all languages, reaching top-10 results in eight languages, including English. Moreover, it supports multiple languages beyond the fourteen covered by the shared task. This multilingual hallucination identifier can help to improve LLM outputs and their usefulness in the future.

Large-scale pretraining has transformed modeling of language and other data types, but its potential remains underexplored in healthcare with structured electronic health records (EHRs). We present a novel generative pretraining strategy for sequential EHR data using next-visit event prediction. Our model learns to autoregressively generate various tokenized clinical events for the next visit based on patient history and inherently handles the joint prediction of heterogeneous data types. Additionally, we introduce regularization on predicting repeated events and highlight a key pitfall in EHR-based foundation model evaluations: repeated event tokens can inflate performance metrics when new onsets are not distinguished from subsequent occurrences. Our model is evaluated via zero-shot prediction for forecasting dementia and knee osteoarthritis incidence within 2 and 5 years, and the model performance rivals a fully fine-tuned masked pretrained Transformer baseline, demonstrating that our approach captures complex clinical dependencies without requiring costly task-specific fine-tuning.

As conversational search engines increasingly adopt generation-based paradigms powered by Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG), the integration of advertisements into generated responses presents both commercial opportunities and challenges for user experience. Unlike traditional search, where advertisements are clearly delineated, generative systems blur the boundary between informational content and promotional material, raising concerns around transparency and trust. In this work, we propose a modular pipeline for advertisement management in RAG-based conversational systems, consisting of an ad-rewriter for seamless ad integration and a robust ad-classifier for detection. We leverage synthetic data to train high-performing classifiers, which are then used to guide two complementary ad-integration strategies: supervised fine-tuning of the ad-rewriter and a best-of-N sampling approach that selects the least detectable ad-integrated response among multiple candidates. Our evaluation focuses on two core questions: the effectiveness of ad classifiers in detecting diverse ad integration strategies, and the training methods that best support coherent, minimally intrusive ad insertion. Experimental results show that our ad-classifier, trained on synthetic advertisement data inspired by marketing strategies and enhanced through curriculum learning, achieves robust detection performance. Additionally, we demonstrate that classifier-guided optimization, through both fine-tuning and best-of-N sampling, significantly improves ad stealth, enabling more seamless integration. These findings contribute an adversarial co-evolution framework for developing more sophisticated ad-aware generative search systems and robust ad classifiers.

Frontier vision-language models (VLMs) have made remarkable improvements in video understanding tasks. However, real-world videos typically exist as continuously evolving data streams (e.g., dynamic scenes captured by wearable glasses), necessitating models to continually adapt to shifting data distributions and novel scenarios. Considering the prohibitive computational costs of fine-tuning models on new tasks, usually, a small subset of parameters is updated while the bulk of the model remains frozen. This poses new challenges to existing continual learning frameworks in the context of large multimodal foundation models, i.e., catastrophic forgetting and update conflict. While the foundation models struggle with parameter-efficient continual learning, the hippocampus in the human brain has evolved highly efficient mechanisms for memory formation and consolidation. Inspired by the rapid Binding and pattern separation mechanisms in the hippocampus, in this work, we propose Bisecle for video-language continual learning, where a multi-directional supervision module is used to capture more cross-modal relationships and a contrastive prompt learning scheme is designed to isolate task-specific knowledge to facilitate efficient memory storage. Binding and separation processes further strengthen the ability of VLMs to retain complex experiences, enabling robust and efficient continual learning in video understanding tasks. We perform a thorough evaluation of the proposed Bisecle, demonstrating its ability to mitigate forgetting and enhance cross-task generalization on several VideoQA benchmarks.

We present a Transformer architecture for long-context language modeling that combines global attention with two biologically inspired components: chunked local attention and a gated FIFO memory mechanism. This unified attention block allows the model to efficiently handle both short-range and long-range dependencies without increasing attention cost quadratically. The memory module persistently stores past token representations using a gated update mechanism inspired by recurrent networks. Rotary positional encoding is applied per attention head to enable directionally disentangled, scale-invariant positional signals. The architecture is implemented entirely from scratch in PyTorch, with no reliance on high-level libraries, enabling transparent and modular experimentation. Our model offers a lightweight and extensible design for tasks such as dialogue modeling, code completion, and document understanding.