With the rapid advancement of natural language processing (NLP) technologies, the demand for high-quality Chinese document question-answering datasets is steadily growing. To address this issue, we present the Chinese Multi-Document Question Answering Dataset(ChiMDQA), specifically designed for downstream business scenarios across prevalent domains including academic, education, finance, law, medical treatment, and news. ChiMDQA encompasses long-form documents from six distinct fields, consisting of 6,068 rigorously curated, high-quality question-answer (QA) pairs further classified into ten fine-grained categories. Through meticulous document screening and a systematic question-design methodology, the dataset guarantees both diversity and high quality, rendering it applicable to various NLP tasks such as document comprehension, knowledge extraction, and intelligent QA systems. Additionally, this paper offers a comprehensive overview of the dataset's design objectives, construction methodologies, and fine-grained evaluation system, supplying a substantial foundation for future research and practical applications in Chinese QA. The code and data are available at: https://anonymous.4open.science/r/Foxit-CHiMDQA/.

To foster trustworthy Artificial Intelligence (AI) within the European Union, the AI Act requires providers to mark and detect the outputs of their general-purpose models. The Article 50 and Recital 133 call for marking methods that are ''sufficiently reliable, interoperable, effective and robust''. Yet, the rapidly evolving and heterogeneous landscape of watermarks for Large Language Models (LLMs) makes it difficult to determine how these four standards can be translated into concrete and measurable evaluations. Our paper addresses this challenge, anchoring the normativity of European requirements in the multiplicity of watermarking techniques. Introducing clear and distinct concepts on LLM watermarking, our contribution is threefold. (1) Watermarking Categorisation: We propose an accessible taxonomy of watermarking methods according to the stage of the LLM lifecycle at which they are applied - before, during, or after training, and during next-token distribution or sampling. (2) Watermarking Evaluation: We interpret the EU AI Act's requirements by mapping each criterion with state-of-the-art evaluations on robustness and detectability of the watermark, and of quality of the LLM. Since interoperability remains largely untheorised in LLM watermarking research, we propose three normative dimensions to frame its assessment. (3) Watermarking Comparison: We compare current watermarking methods for LLMs against the operationalised European criteria and show that no approach yet satisfies all four standards. Encouraged by emerging empirical tests, we recommend further research into watermarking directly embedded within the low-level architecture of LLMs.

Zero-Shot Stance Detection (ZSSD) identifies the attitude of the post toward unseen targets. Existing research using contrastive, meta-learning, or data augmentation suffers from generalizability issues or lack of coherence between text and target. Recent works leveraging large language models (LLMs) for ZSSD focus either on improving unseen target-specific knowledge or generating explanations for stance analysis. However, most of these works are limited by their over-reliance on explicit reasoning, provide coarse explanations that lack nuance, and do not explicitly model the reasoning process, making it difficult to interpret the model's predictions. To address these issues, in our study, we develop a novel interpretable ZSSD framework, IRIS. We provide an interpretable understanding of the attitude of the input towards the target implicitly based on sequences within the text (implicit rationales) and explicitly based on linguistic measures (explicit rationales). IRIS considers stance detection as an information retrieval ranking task, understanding the relevance of implicit rationales for different stances to guide the model towards correct predictions without requiring the ground-truth of rationales, thus providing inherent interpretability. In addition, explicit rationales based on communicative features help decode the emotional and cognitive dimensions of stance, offering an interpretable understanding of the author's attitude towards the given target. Extensive experiments on the benchmark datasets of VAST, EZ-STANCE, P-Stance, and RFD using 50%, 30%, and even 10% training data prove the generalizability of our model, benefiting from the proposed architecture and interpretable design.

Large language models (LLMs) achieve strong performance across benchmarks--from knowledge quizzes and math reasoning to web-agent tasks--but these tests occur in static settings, lacking real dynamics and uncertainty. Consequently, they evaluate isolated reasoning or problem-solving rather than decision-making under uncertainty. To address this, we introduce LiveTradeBench, a live trading environment for evaluating LLM agents in realistic and evolving markets. LiveTradeBench follows three design principles: (i) Live data streaming of market prices and news, eliminating dependence on offline backtesting and preventing information leakage while capturing real-time uncertainty; (ii) a portfolio-management abstraction that extends control from single-asset actions to multi-asset allocation, integrating risk management and cross-asset reasoning; and (iii) multi-market evaluation across structurally distinct environments--U.S. stocks and Polymarket prediction markets--differing in volatility, liquidity, and information flow. At each step, an agent observes prices, news, and its portfolio, then outputs percentage allocations that balance risk and return. Using LiveTradeBench, we run 50-day live evaluations of 21 LLMs across families. Results show that (1) high LMArena scores do not imply superior trading outcomes; (2) models display distinct portfolio styles reflecting risk appetite and reasoning dynamics; and (3) some LLMs effectively leverage live signals to adapt decisions. These findings expose a gap between static evaluation and real-world competence, motivating benchmarks that test sequential decision making and consistency under live uncertainty.

We evaluate how visualizations can influence the judgment of MLLMs about the presence or absence of bridges in a network. We show that the inclusion of visualization improves confidence over a structured text-based input that could theoretically be helpful for answering the question. On the other hand, we observe that standard visualization techniques create a strong bias towards accepting or refuting the presence of a bridge -- independently of whether or not a bridge actually exists in the network. While our results indicate that the inclusion of visualization techniques can effectively influence the MLLM's judgment without compromising its self-reported confidence, they also imply that practitioners must be careful of allowing users to include visualizations in generative AI applications so as to avoid undesired hallucinations.

The increasing complexity of machine learning models and the proliferation of diverse hardware architectures (CPUs, GPUs, accelerators) make achieving optimal performance a significant challenge. Heterogeneity in instruction sets, specialized kernel requirements for different data types and model features (e.g., sparsity, quantization), and architecture-specific optimizations complicate performance tuning. Manual optimization is resource-intensive, while existing automatic approaches often rely on complex hardware-specific heuristics and uninterpretable intermediate representations, hindering performance portability. We introduce PerfLLM, a novel automatic optimization methodology leveraging Large Language Models (LLMs) and Reinforcement Learning (RL). Central to this is PerfDojo, an environment framing optimization as an RL game using a human-readable, mathematically-inspired code representation that guarantees semantic validity through transformations. This allows effective optimization without prior hardware knowledge, facilitating both human analysis and RL agent training. We demonstrate PerfLLM's ability to achieve significant performance gains across diverse CPU (x86, Arm, RISC-V) and GPU architectures.

We study LLMs for tabular prediction with mixed text, numeric, and categorical fields. We introduce TabGemma, a schema-agnostic in-context learner that treats rows as sequences and tackles two practical hurdles when adapting pretrained LLMs for tabular predictions: unstable numeric tokenization and limited context size. We propose to canonicalize numbers via signed scientific notation and continue pretraining of a 12B Gemma 3 model with a target imputation objective using a large-scale real world dataset. For inference, we use a compact n-gram-based retrieval to select informative exemplars that fit within a 128k-token window. On semantically rich benchmarks, TabGemma establishes a new state of the art on classification across low- and high-data regimes and improves monotonically with more context rows. For regression, it is competitive at small sample sizes but trails conventional approaches as data grows. Our results show that LLMs can be effective tabular in-context learners on highly semantic tasks when paired with dedicated numeric handling and context retrieval, while motivating further advances in numeric modeling and long-context scaling.

In this study, we explore the fine-tuning of Large Language Models (LLMs) to better support policymakers in their crucial work of understanding, analyzing, and crafting legal regulations. To equip the model with a deep understanding of legal texts, we curated a supervised dataset tailored to the specific needs of the legal domain. Additionally, we integrated the Retrieval-Augmented Generation (RAG) method, enabling the LLM to access and incorporate up-to-date legal knowledge from external sources. This combination of fine-tuning and RAG-based augmentation results in a tool that not only processes legal information but actively assists policymakers in interpreting regulations and drafting new ones that align with current needs. The results demonstrate that this approach can significantly enhance the effectiveness of legal research and regulation development, offering a valuable resource in the ever-evolving field of law.

This paper presents the first empirical demonstration of controllable locality in transformer language models, a novel architectural framework that enables continuous control over the degree of representation localization through a tunable locality dial parameter. Unlike traditional language models that rely exclusively on distributed representations, our approach allows dynamic interpolation between highly interpretable localist encodings and efficient distributed representations without requiring model retraining. We conducted experiments on the WikiText corpus using a two-layer transformer architecture, systematically varying the locality parameter λ across the full spectrum from 1.0 (fully localist) to 0.0 (fully distributed). Our results demonstrate that localist configurations achieve dramatically lower attention entropy, with λ = 1.0 yielding 5.36 bits compared to 7.18 bits at λ = 0.0, while maintaining substantially higher pointer fidelity scores reflecting stronger alignment with rule-specified targets. Prediction experiments reveal that intermediate locality values optimize the tradeoff between interpretability and performance, with λ = 0.6 achieving test perplexity of 4.65 and accuracy of 84.7%. These findings establish that localist language models provide a practical framework for applications in regulated domains requiring both transparency and capability, offering precise mathematical control over the interpretability-performance spectrum through explicit penalty thresholds and information-theoretic design principles.

Measuring the full abilities of large language models (LLMs) requires benchmarks representing multiple tasks. We aim to create large, high-quality datasets for comparison of logical reasoning skills across several languages and of suitable difficulty for LLMs of various reasoning ability. We explore multiple ways of increasing difficulty. We generate zebra puzzles in multiple languages, themes, sizes and including 14 different clue types and 8 red herring types (uninformative clues). We find puzzle sizes 2x3 and 4x5 are sufficiently challenging for GPT-4o mini (a non-reasoning model) and o3-mini (a reasoning model), respectively. Including 5 red herrings decreases o3-mini puzzle-level accuracy on 4x5 puzzles by 15$\pm$7 %. Scores of o3-mini on 4x5 puzzles are not significantly affected by use of English vs. Danish or the common houses theme vs. the country-specific smoerrebroed theme. We find no correlation between difficulty and the selected clue types. Datasets of 128+1024 puzzles are published as MultiZebraLogic in each of nine Germanic languages for sizes 2x3 and 4x5. We publish code for puzzle generation, designed for adaptablity into more languages and themes.