Increasing test-time compute for LLMs shows promise across domains but remains underexplored in code generation, despite extensive study in math. In this paper, we propose S*, the first hybrid test-time scaling framework that substantially improves the coverage and selection accuracy of generated code. S* augments the existing parallel scaling approach with sequential scaling to further increase the performance. It further leverages a novel selection mechanism that adaptively generates distinguishing inputs for pairwise comparison, combined with execution-grounded information to robustly identify correct solutions.We evaluate S* across 12 Large Language Models and Large Reasoning Models and show that: (1) S* consistently improves performance across model families and sizes, enabling a 3B model to outperform GPT-4o-mini; (2) S* enables non-reasoning models to surpass reasoning models—GPT-4o-mini with S* outperforms o1-preview by 3.7% on LiveCodeBench; (3) S* further boosts state-of-the-art reasoning models—DeepSeek-R1-Distill-Qwen-32B with S* achieves 85.7% on LiveCodeBench, approaching o1 (high) at 88.5%. Codes, model generations and intermediate experiments results are available under Codes, model generations and intermediate ex-periments results are available under https://github.com/NovaSky-AI/SkyThought.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
HetGCoT: Heterogeneous Graph-Enhanced Chain-of-Thought LLM Reasoning for Academic Question Answering
PDF ↗Academic question answering (QA) in heterogeneous scholarly networks presents unique challenges requiring both structural understanding and interpretable reasoning. While graph neural networks (GNNs) capture structured graph information and large language models (LLMs) demonstrate strong capabilities in semantic comprehension, current approaches lack integration at the reasoning level. We propose HetGCoT, a framework enabling LLMs to effectively leverage and learn information from graphs to reason interpretable academic QA results. Our framework introduces three technical contributions: (1) a framework that transforms heterogeneous graph structural information into LLM-processable reasoning chains, (2) an adaptive metapath selection mechanism identifying relevant subgraphs for specific queries, and (3) a multi-step reasoning strategy systematically incorporating graph contexts into the reasoning process. Experiments on OpenAlex and DBLP datasets show our approach outperforms all sota baselines. The framework demonstrates adaptability across different LLM architectures and applicability to various scholarly question answering tasks.
Parsing the Switch: LLM-Based UD Annotation for Complex Code-Switched and Low-Resource Languages
PDF ↗Code-switching presents a complex challenge for syntactic analysis, especially in low-resource language settings where annotated data is scarce. While recent work has explored the use of large language models (LLMs) for sequence-level tagging, few approaches systematically investigate how well these models capture syntactic structure in code-switched contexts. Moreover, existing parsers trained on monolingual treebanks often fail to generalize to multilingual and mixed-language input. To address this gap, we introduce the BiLingua Pipeline, an LLM-based annotation pipeline designed to produce Universal Dependencies (UD) annotations for code-switched text. First, we develop a prompt-based framework for Spanish-English and Spanish-Guaraní data, combining few-shot LLM prompting with expert review. Second, we release two annotated datasets, including the first Spanish-Guaraní UD-parsed corpus. Third, we conduct a detailed syntactic analysis of switch points across language pairs and communicative contexts. Experimental results show that BiLingua Pipeline achieves up to 95.29% LAS after expert revision, significantly outperforming prior baselines and multilingual parsers. These results show that LLMs, when carefully guided, can serve as practical tools for bootstrapping syntactic resources in under-resourced, code-switched environments.
Existing metrics often lack the granularity and interpretability to capture nuanced clinical differences between candidate and ground-truth radiology reports, resulting in suboptimal evaluation. We introduce a **Cl**inically grounded tabular framework with **E**xpert-curated labels and **A**ttribute-level comparison for **R**adiology report evaluation (**CLEAR**). CLEAR not only examines whether a report can accurately identify the presence or absence of medical conditions, but it also assesses whether the report can precisely describe each positively identified condition across five key attributes: first occurrence, change, severity, descriptive location, and recommendation. Compared with prior works, CLEAR’s multi-dimensional, attribute-level outputs enable a more comprehensive and clinically interpretable evaluation of report quality. Additionally, to measure the clinical alignment of CLEAR, we collaborated with five board-certified radiologists to develop **CLEAR-Bench**, a dataset of 100 chest radiograph reports from MIMIC-CXR, annotated across 6 curated attributes and 13 CheXpert conditions. Our experiments demonstrated that CLEAR achieves high accuracy in extracting clinical attributes and provides automated metrics that are strongly aligned with clinical judgment.
Modern causal language models stack many attention blocks to improve performance, but not all blocks are necessary for every task. We propose Hopscotch, a simple yet effective method that identifies and skips attention blocks with least contributions to a task and adapts to preserve output quality. Hopscotch jointly optimizes which blocks to skip and how to scale the outputs of the remaining layers. By introducing lightweight, trainable scaling parameters to attention and MLP blocks, it mitigates distribution shifts in hidden states caused by removing attention blocks. Hopscotch does not modify model weights or require access to pretraining or instruction-tuning data, and is compatible with existing model compression techniques. When applied to \texttt{Llama-3.1-8B} and \texttt{Qwen-2.5-7B}, Hopscotch achieves less than a 2% drop in performance even after skipping four attention blocks.
Mind the Dialect: NLP Advancements Uncover Fairness Disparities for Arabic Users in Recommendation Systems
PDF ↗Recommendation systems play a critical role in shaping user experiences and access to digital content. However, these systems can exhibit unfair behavior when their performance varies across user groups, especially in linguistically diverse populations. Recent advances in NLP have enabled the identification of user dialects, allowing for more granular analysis of such disparities. In this work, we investigate fairness disparities in recommendation quality among Arabic-speaking users, a population whose dialectal diversity is underrepresented in recommendation system research. By uncovering performance gaps across dialectal variation, we highlight the intersection of NLP and recommendation system and underscore the broader social impact of NLP. Our findings emphasize the importance of interdisciplinary approaches in building fair recommendation systems, particularly for global and local platforms serving diverse Arabic-speaking communities. The source code is available at https://github.com/alshabae/FairArRecSys.
SIMBA UQ: Similarity-Based Aggregation for Uncertainty Quantification in Large Language Models
PDF ↗When does a large language model (LLM) know what it does not know? Uncertainty quantification (UQ) provides measures of uncertainty, such as an estimate of the confidence in an LLM’s generated output, and is therefore increasingly recognized as a crucial component of trusted AI systems. Black-box UQ methods do not require access to internal model information from the generating LLM and therefore have numerous real-world advantages, such as robustness to system changes, adaptability to choice of LLM, reduced costs, and computational tractability. In this paper, we investigate the effectiveness of UQ techniques that are primarily but not necessarily entirely black- box, where the consistency between a generated output and other sampled generations is used as a proxy for confidence in its correctness. We propose a high-level non-verbalized similarity-based aggregation framework that subsumes a broad swath of UQ approaches suitable for complex generative tasks, as well as introduce specific novel techniques from the framework that train confidence estimation models using small training sets. Through an empirical study with datasets spanning the diverse tasks of question answering, summarization, and text-to-SQL, we demonstrate that our proposed similarity-based methods can yield better calibrated confidences than baselines.
MRFD: Multi-Region Fusion Decoding with Self-Consistency for Mitigating Hallucinations in LVLMs
PDF ↗Large Vision-Language Models (LVLMs) have shown strong performance across multimodal tasks. However, they often produce hallucinations—text that is inconsistent with visual input, due to the limited ability to verify information in different regions of the image. To address this, we propose **Multi-Region Fusion Decoding (MRFD)**, a training-free decoding method that improves factual grounding by modeling inter-region consistency. MRFD identifies salient regions using cross-attention, generates initial responses for each, and computes reliability weights based on Jensen-Shannon Divergence (JSD) among the responses. These weights guide a consistency-aware fusion of per-region predictions, using region-aware prompts inspired by Chain-of-Thought reasoning. Experiments across multiple LVLMs and benchmarks show that MRFD significantly reduces hallucinations and improves response factuality without requiring model updates.
Post-hoc Study of Climate Microtargeting on Social Media Ads with LLMs: Thematic Insights and Fairness Evaluation
PDF ↗Climate change communication on social media increasingly employs microtargeting strategies to effectively reach and influence specific demographic groups. This study presents a *post-hoc* analysis of microtargeting practices within climate campaigns by leveraging large language models (LLMs) to examine Meta (previously known as Facebook) advertisements. Our analysis focuses on two key aspects: **demographic targeting** and **fairness**. We evaluate the ability of LLMs to accurately predict the intended demographic targets, such as gender and age group. Furthermore, we instruct the LLMs to generate explanations for their classifications, providing transparent reasoning behind each decision. These explanations reveal the specific thematic elements used to engage different demographic segments, highlighting distinct strategies tailored to various audiences. Our findings show that ***young adults*** are primarily targeted through messages emphasizing *activism and environmental consciousness*, while **women** are engaged through themes related to *caregiving roles and social advocacy*. Additionally, we conduct a comprehensive fairness analysis to uncover biases in model predictions. We assess disparities in accuracy and error rates across demographic groups using established fairness metrics such as Demographic Parity, Equal Opportunity, and Predictive Equality. Our findings indicate that while LLMs perform well overall, certain biases exist, particularly in the classification of **male** audiences. The analysis of thematic explanations uncovers recurring patterns in messaging strategies tailored to various demographic groups, while the fairness analysis underscores the need for more inclusive targeting methods. This study provides a valuable framework for future research aimed at enhancing transparency, accountability, and inclusivity in social media-driven climate campaigns.
Responsible use of Authorship Verification (AV) systems not only requires high accuracy but also interpretable solutions. More importantly, for systems to be used to make decisions with real-world consequences requires the model’s prediction to be explainable using interpretable features that can be traced to the original texts. Neural methods achieve high accuracies, but their representations lack direct interpretability. Furthermore, LLM predictions cannot be explained faithfully – if there is an explanation given for a prediction, it doesn’t represent the reasoning process behind the model’s prediction. In this paper, we introduce Residualized Similarity (RS), a novel method that supplements systems using interpretable features with a neural network to improve their performance while maintaining interpretability. Authorship verification is fundamentally a similarity task, where the goal is to measure how alike two documents are. The key idea is to use the neural network to predict a similarity residual, i.e. the error in the similarity predicted by the interpretable system. Our evaluation across four datasets shows that not only can we match the performance of state-of-the-art authorship verification models, but we can show how and to what degree the final prediction is faithful and interpretable.
Large Language Models (LLMs) have demonstrated impressive capabilities in text generation but raise concerns regarding potential copyright infringement. While prior research has explored mitigation strategies like content filtering and alignment, the impact of adversarial persuasion techniques in eliciting copyrighted content remains underexplored. This paper investigates how structured persuasion strategies, including logical appeals, emotional framing, and compliance techniques, can be used to manipulate LLM outputs and potentially increase copyright risks. We introduce a structured persuasion workflow, incorporating query mutation, intention-preserving filtering, and few-shot prompting, to systematically analyze the influence of persuasive prompts on LLM responses. Through experiments on state-of-the-art LLMs, including GPT-4o-mini and Claude-3-haiku, we quantify the effectiveness of different persuasion techniques and assess their implications for AI safety. Our results highlight the vulnerabilities of LLMs to adversarial persuasion and provide empirical evidence of the increased risk of generating copyrighted content under such influence. We conclude with recommendations for strengthening model safeguards and future directions for enhancing LLM robustness against manipulation. Code is available at https://github.com/Rongite/Persuasion.
Large Language Models (LLMs) excel in Natural Language Processing (NLP) tasks but often propagate societal biases from their training data, leading to discriminatory outputs. These biases are amplified by the models’ self-attention mechanisms, which disproportionately emphasize biased correlations with sensitive tokens, like “he” or “she”, reflecting the sensitive attributes such as gender and race. To address this issue, we propose a novel fine-tuning method, called Cross-Attention-based Weight Decay (CrAWD), which modifies the LLM architecture to mitigate bias. CrAWD introduces a cross-attention mechanism between an input sequence and a sensitive token sequence, enabling the model to identify and selectively decay the attention weights of tokens associated with sensitive tokens. This reduces the influence of biased association on the model’s generation while maintaining task performance. Evaluations on real-world datasets demonstrate the effectiveness of our proposed CrAWD method. Notably, our method can handle multiple sensitive attributes by adjusting the sensitive token sequence, and it does not require full knowledge of sensitive tokens presented in the dataset, underscoring CrAWD’s versatility in promoting fair LLMs across various applications.
As Large Language Models (LLMs) gain expertise across diverse domains and modalities, scalable oversight becomes increasingly challenging, particularly when their capabilities may surpass human evaluators. Debate has emerged as a promising mechanism for enabling such oversight. We extend the debate paradigm to a multimodal setting, exploring its potential for blind models to supervise and enhance the performance of sighted ones. We focus on visual question answering (VQA), where two “sighted” expert vision-language models debate an answer, while a “blind” (text-only) judge adjudicates based solely on the quality of the arguments. In our framework, the experts only defend answers aligned with their beliefs, thereby obviating the need for explicit role-playing and concentrating the debate on instances of expert disagreement. Experiments on several multimodal tasks demonstrate that the debate framework consistently outperforms individual expert models. Moreover, judgments from blind LLMs can be used to instil reasoning capabilities in vision-language models through fine-tuning.
In our era of widespread false information, human fact-checkers often face the challenge of duplicating efforts when verifying claims that may have already been addressed in other countries or languages. As false information transcends linguistic boundaries, the ability to automatically detect previously fact-checked claims across languages has become an increasingly important task. This paper presents the first comprehensive evaluation of large language models (LLMs) for multilingual previously fact-checked claim detection. We assess seven LLMs across 20 languages in both monolingual and cross-lingual settings. Our results show that while LLMs perform well for high-resource languages, they struggle with low-resource languages. Moreover, translating original texts into English proved to be beneficial for low-resource languages. These findings highlight the potential of LLMs for multilingual previously fact-checked claim detection and provide a foundation for further research on this promising application of LLMs.
MFTCXplain: A Multilingual Benchmark Dataset for Evaluating the Moral Reasoning of LLMs through Multi-hop Hate Speech Explanation
PDF ↗Ensuring the moral reasoning capabilities of Large Language Models (LLMs) is a growing concern as these systems are used in socially sensitive tasks. Nevertheless, current evaluation benchmarks present two major shortcomings: a lack of annotations that justify moral classifications, which limits transparency and interpretability; and a predominant focus on English, which constrains the assessment of moral reasoning across diverse cultural settings. In this paper, we introduce MFTCXplain, a multilingual benchmark dataset for evaluating the moral reasoning of LLMs via multi-hop hate speech explanations using the Moral Foundations Theory. MFTCXplain comprises 3,000 tweets across Portuguese, Italian, Persian, and English, annotated with binary hate speech labels, moral categories, and text span-level rationales. Our results show a misalignment between LLM outputs and human annotations in moral reasoning tasks. While LLMs perform well in hate speech detection (F1 up to 0.836), their ability to predict moral sentiments is notably weak (F1 < 0.35). Furthermore, rationale alignment remains limited mainly in underrepresented languages. Our findings show the limited capacity of current LLMs to internalize and reflect human moral reasoning.
Through a controlled study, we identify a systematic deficiency in the multimodal grounding of Vision Language Models (VLMs). While VLMs can recall factual associations when provided a textual reference to an entity, their ability to do so is significantly diminished when the reference is visual instead. Forcing VLMs to rely on image representations of an entity halves their ability to recall factual knowledge, suggesting that VLMs struggle to link their internal knowledge of an entity with its image representation. We show that such linking failures are correlated with the expression of distinct patterns in model internal states, and that probes on these internal states achieve over 92% accuracy at flagging cases where the VLM response is unreliable. These probes can be applied, without retraining, to identify when a VLM will fail to correctly answer a question that requires an understanding of multimodal input. When used to facilitate selective prediction on a visual question answering task, the probes increase coverage by 7.87% (absolute) while also reducing the risk of error by 0.9% (absolute). Addressing the systematic, detectable deficiency is an important avenue in language grounding, and we provide informed recommendations for future directions.
We propose XRAG, a novel benchmark designed to evaluate the generation abilities of LLMs in cross-lingual Retrieval-Augmented Generation (RAG) settings where the user language does not match the retrieval results. XRAG is constructed from recent news articles to ensure that its questions require external know-ledge to be answered. It covers the real-world scenarios of monolingual and multilingual retrieval, and provides relevancy annotations for each retrieved document. Our novel dataset construction pipeline results in questions that require complex reasoning, as evidenced by the significant gap between human and LLM performance. Consequently, XRAG serves as a valuable benchmark for studying LLM reasoning abilities, even before considering the additional cross-lingual complexity. Experimental results on five LLMs uncover two previously unreported challenges in cross-lingual RAG: 1) in the monolingual retrieval setting, all evaluated models struggle with response language correctness; 2) in the multilingual retrieval setting, the main challenge lies in reasoning over retrieved information across languages rather than generation of non-English text.
Although large language models (LLMs) have demonstrated outperforming human experts in medical examinations, it remains challenging to adopt LLMs in real-world clinical decision-making that typically involves multi-hop medical reasoning. Common practices include prompting commercial LLMs and fine-tuning LLMs on medical data. However, in the clinical domain, using commercial LLMs raises privacy concerns regarding sensitive patient data. Fine-tuning competitive medical LLMs for different tasks usually requires extensive data and computing resources, which are difficult to acquire, especially in medical institutions with limited infrastructure. We propose DrAgent, which can build LLMs as agents to deliver accurate medical decision-making and reasoning. In implementation, we take a lightweight LLM as the backbone to collaborate with diverse clinical tools. To make efficient use of data, DrAgent introduces recursive curriculum learning to optimize the LLM in an easy-to-hard progression. The results show that our approach achieves competitive performance on diverse datasets.
Testing is essential to modern software engineering for building reliable software.Given the high costs of manually creating test cases,automated test case generation, particularly methods utilizing large language models,has become increasingly popular.These neural approaches generate semantically meaningful tests that are more maintainable compared with traditional automated testing methods such as fuzzing.However, the diversity and volume of unit tests in current datasets are limited, especially for newer but important languages.In this paper, we present a novel data augmentation technique, *FuzzAug*,that brings the benefits of fuzzing to large language models by incorporating valid testing semantics and providing diverse coverage-guided inputs.Doubling the size of training datasets,FuzzAug improves performance over the baselines significantly.This technique demonstrates the potential of introducing prior knowledge from dynamic software analysisto improve neural test generation,offering significant enhancements in this task.Our code is open-sourced at https://github.com/SecurityLab-UCD/FuzzAug.
Not Lost After All: How Cross-Encoder Attribution Challenges Position Bias Assumptions in LLM Summarization
PDF ↗Position bias, the tendency of Large Language Models (LLMs) to select content based on its structural position in a document rather than its semantic relevance, has been viewed as a key limitation in automatic summarization. To measure position bias, prior studies rely heavily on n-gram matching techniques, which fail to capture semantic relationships in abstractive summaries where content is extensively rephrased. To address this limitation, we apply a cross-encoder-based alignment method that jointly processes summary-source sentence pairs, enabling more accurate identification of semantic correspondences even when summaries substantially rewrite the source. Experiments with five LLMs across six summarization datasets reveal significantly different position bias patterns than those reported by traditional metrics. Our findings suggest that these patterns primarily reflect rational adaptations to document structure and content rather than true model limitations. Through controlled experiments and analyses across varying document lengths and multi-document settings, we show that LLMs use content from all positions more effectively than previously assumed, challenging common claims about “lost-in-the-middle” behaviour.