When Xi Jinping, China's leader, made his entrance at a symposium with a group of top entrepreneurs this past week, he seemed to be in good spirits. China has had a few good weeks. Artificial intelligence models by the startup DeepSeek sent U.S. stocks tumbling and Western commentators screaming, "Sputnik moment." Then, an animated film based on Chinese mythology raked in nearly 2 billion. Xi signaled that he stood behind the private sector at the meeting on Monday, pushing the Hong Kong stock market to its highest point in three years.

Recent Multi-Party Conversation (MPC) models typically rely on graph-based approaches to capture dialogue structures. However, these methods have limitations, such as information loss during the projection of utterances into structural embeddings and constraints in leveraging pre-trained language models directly. In this paper, we propose \textbf{SS-MPC}, a response generation model for MPC that eliminates the need for explicit graph structures. Unlike existing models that depend on graphs to analyze conversation structures, SS-MPC internally encodes the dialogue structure as a sequential input, enabling direct utilization of pre-trained language models. Experimental results show that \textbf{SS-MPC} achieves \textbf{15.60\% BLEU-1} and \textbf{12.44\% ROUGE-L} score, outperforming the current state-of-the-art MPC response generation model by \textbf{3.91\%p} in \textbf{BLEU-1} and \textbf{0.62\%p} in \textbf{ROUGE-L}. Additionally, human evaluation confirms that SS-MPC generates more fluent and accurate responses compared to existing MPC models.

Understanding the uncertainty in large language model (LLM) explanations is important for evaluating their faithfulness and reasoning consistency, and thus provides insights into the reliability of LLM's output regarding a question. In this work, we propose a novel framework that quantifies uncertainty in LLM explanations through a reasoning topology perspective. By designing a structural elicitation strategy, we guide the LLMs to frame the explanations of an answer into a graph topology. This process decomposes the explanations into the knowledge related sub-questions and topology-based reasoning structures, which allows us to quantify uncertainty not only at the semantic level but also from the reasoning path. It further brings convenience to assess knowledge redundancy and provide interpretable insights into the reasoning process. Our method offers a systematic way to interpret the LLM reasoning, analyze limitations, and provide guidance for enhancing robustness and faithfulness. This work pioneers the use of graph-structured uncertainty measurement in LLM explanations and demonstrates the potential of topology-based quantification.

Impact of Compression on Model Performance: Through comprehensive experiments, the study demonstrates that moderate pruning can enhance model efficiency, but excessive compression leads to substantial performance degradation in both language and multimodal models. Sustainable AI Development: The findings emphasize the need for optimized AI models to reduce the environmental impact, addressing critical issues like carbon footprint, electricity, and water consumption associated with training and deploying large-scale AI systems. Systematic Weight Evaluation for Pruning Large Language Models: Enhancing Performance and Sustainability Ashhadul Islam a, Samir Brahim Belhaouari a, Amine Bermak a a College Of Science & Engineering, Hamad Bin Khalifa University, Education City, Doha, 34110, QatarAbstract The exponential growth of large language models (LLMs) like ChatGPT has revolutionized artificial intelligence, offering unprecedented capabilities in natural language processing. However, the extensive computational resources required for training these models have significant environmental implications, including high carbon emissions, energy consumption, and water usage. This research presents a novel approach to LLM pruning, focusing on the systematic evaluation of individual weight importance throughout the training process. By monitoring parameter evolution over time, we propose a method that effectively reduces model size without compromising performance. Extensive experiments with both a scaled-down LLM and a large multimodal model reveal that moderate pruning enhances efficiency and reduces loss, while excessive pruning drastically deteriorates model performance. These findings highlight the critical need for optimized AI models to ensure sustainable development, balancing technological advancement with environmental responsibility.

Scaling pre-training compute has proven effective for achieving mulitlinguality, but does the same hold for test-time scaling? In this work, we introduce MCLM, a multilingual math benchmark featuring competition-level problems in 55 languages. We test three test-time scaling methods-Outcome Reward Modeling (ORM), Process Reward Modeling (ORM), and Budget Forcing (BF)-on both Qwen2.5-1.5B Math and MR1-1.5B, a multilingual LLM we trained for extended reasoning. Our experiments show that using Qwen2.5-1.5B Math with ORM achieves a score of 35.8 on MCLM, while BF on MR1-1.5B attains 35.2. Although "thinking LLMs" have recently garnered significant attention, we find that their performance is comparable to traditional scaling methods like best-of-N once constrained to similar levels of inference FLOPs. Moreover, while BF yields a 20-point improvement on English AIME, it provides only a 1.94-point average gain across other languages-a pattern consistent across the other test-time scaling methods we studied-higlighting that test-time scaling may not generalize as effectively to multilingual tasks. To foster further research, we release MCLM, MR1-1.5B, and evaluation results.

Large multimodal models (LMMs) often struggle to recognize novel concepts, as they rely on pre-trained knowledge and have limited ability to capture subtle visual details. Domain-specific knowledge gaps in training also make them prone to confusing visually similar, commonly misrepresented, or low-resource concepts. To help LMMs better align nuanced visual features with language, improving their ability to recognize and reason about novel or rare concepts, we propose a Contrastive visual Data Augmentation (CoDA) strategy. CoDA extracts key contrastive textual and visual features of target concepts against the known concepts they are misrecognized as, and then uses multimodal generative models to produce targeted synthetic data. Automatic filtering of extracted features and augmented images is implemented to guarantee their quality, as verified by human annotators. We show the effectiveness and efficiency of CoDA on low-resource concept and diverse scene recognition datasets including INaturalist and SUN. We additionally collect NovelSpecies, a benchmark dataset consisting of newly discovered animal species that are guaranteed to be unseen by LMMs. LLaVA-1.6 1-shot updating results on these three datasets show CoDA significantly improves SOTA visual data augmentation strategies by 12.3% (NovelSpecies), 5.1% (SUN), and 6.0% (iNat) absolute gains in accuracy.

Question answering on the hybrid context of tables and text (TATQA) is a critical task, with broad applications in data-intensive domains. However, existing TATQA datasets are limited to English, leading to several drawbacks: (i) They overlook the challenges of multilingual TAT-QA and cannot assess model performance in the multilingual setting. (ii) They do not reflect real-world scenarios where tables and texts frequently appear in non-English languages. To address the limitations, we propose the first multilingual TATQA dataset (MULTITAT). Specifically, we sample data from 3 mainstream TATQA datasets and translate it into 10 diverse languages. To align the model TATQA capabilities in English with other languages, we develop a baseline, Ours. Experimental results reveal that the performance on non-English data in MULTITAT drops by an average of 19.4% compared to English, proving the necessity of MULTITAT. We further analyze the reasons for this performance gap. Furthermore, Ours outperforms other baselines by an average of 3.3, demonstrating its effectiveness.

Recently, the Muon optimizer based on matrix orthogonalization has demonstrated strong results in training small-scale language models, but the scalability to larger models has not been proven. We identify two crucial techniques for scaling up Muon: (1) adding weight decay and (2) carefully adjusting the per-parameter update scale. These techniques allow Muon to work out-of-the-box on large-scale training without the need of hyper-parameter tuning. Scaling law experiments indicate that Muon achieves $\sim\!2\times$ computational efficiency compared to AdamW with compute optimal training. Based on these improvements, we introduce Moonlight, a 3B/16B-parameter Mixture-of-Expert (MoE) model trained with 5.7T tokens using Muon. Our model improves the current Pareto frontier, achieving better performance with much fewer training FLOPs compared to prior models. We open-source our distributed Muon implementation that is memory optimal and communication efficient. We also release the pretrained, instruction-tuned, and intermediate checkpoints to support future research.

Tabular data contains rich structural semantics and plays a crucial role in organizing and manipulating information. To better capture these structural semantics, this paper introduces the HybrId-modal Preference oPtimizatiOn (HIPPO) model, which represents tables using both text and image, and optimizes MLLMs to effectively learn more comprehensive table information from these multiple modalities. Specifically, HIPPO samples model responses from hybrid-modal table representations and designs a modality-consistent sampling strategy to enhance response diversity and mitigate modality bias during DPO training. Experimental results on table question answering and table fact verification tasks demonstrate the effectiveness of HIPPO, achieving a 4% improvement over various table reasoning models. Further analysis reveals that HIPPO not only enhances reasoning abilities based on unimodal table representations but also facilitates the extraction of crucial and distinct semantics from different modal representations. All data and codes are available at https://github.com/NEUIR/HIPPO.

Understanding the effects of economic shocks on firms is critical for analyzing economic growth and resilience. We introduce a Web-Based Affectedness Indicator (W AI), a general-purpose tool for real-time monitoring of economic disruptions across diverse contexts. By leveraging Large Language Model (LLM) assisted classification and information extraction on texts from over five million company websites, W AI quantifies the degree and nature of firms' responses to external shocks. Using the COVID-19 pandemic as a specific application, we show that W AI is highly correlated with pandemic containment measures and reliably predicts firm performance. Unlike traditional data sources, W AI provides timely firm-level information across industries and geographies worldwide that would otherwise be unavailable due to institutional and data availability constraints. This methodology offers significant potential for monitoring and mitigating the impact of technological, political, financial, health or environmental crises, and represents a transformative tool for adaptive policy-making and economic resilience. Economic shocks, whether driven by public health crises, technological disruptions, geopolitical conflicts, or climate events, pose significant challenges to businesses and policymakers alike. Timely and accurate monitoring of these shocks is critical for crafting effective responses and enhancing economic resilience. However, traditional methods for measuring the impacts of such disruptions - such as surveys and administrative data - are often limited by costs, time lags, and coverage. In this study, we introduce the Web-Based Affectedness Indicator (W AI), a scalable and cost-effective tool for real-time monitoring of economic disruptions at the firm level. By analyzing textual data from millions of company websites, W AI provides granular insights into how firms experience and respond to external shocks. This 1 methodology overcomes traditional limitations by leveraging ubiquitous online content and state-of-the-art natural language processing (NLP) models to generate a dynamic and comprehensive view of economic affectedness. W AI can provide information on a wide range of challenges, including supply chain disruptions, financial crises, and climate-related shocks.