Designing effective task-level prompts is crucial for improving the performance of Large Language Models (LLMs). While prior work on instruction induction demonstrates that LLMs can infer better instructions with limited examples, existing approaches often rely on input-output pairs, where obtaining labeled answers can be difficult or costly. To address this limitation, we propose Strategy-Induct, a framework that derives task-level instructions solely from a small set of example questions without requiring labeled answers. Our approach first prompts the model to generate explicit reasoning strategies for each question, forming (strategy, question) pairs. These pairs are then used to induce a task instruction that guides reasoning. Experiments across multiple tasks and model scales demonstrate that Strategy-Induct outperforms state-of-the-art methods in question-only settings. Furthermore, we observe that jointly utilizing LLMs and Large Reasoning Models for both task instruction generation and inference can lead to further performance improvements.
论文检索
输入标题、作者或关键词，从 9,460 篇学术成果中精准定位
Large Language Models (LLMs) have fundamentally transformed natural language processing (NLP), demonstrating remarkable capabilities across a wide spectrum of tasks. However, when applied to instruction-based text editing, LLMs continue to exhibit some limitations. Different from free-form generation, instruction-based editing requires precise, targeted modifications that respect two essential properties: faithfully implementing the specific instruction and local fidelity. Existing approaches often overlook these properties, treating editing as a generic text generation problem. As a result, they either over-edit or fail to apply modifications consistently. To address this gap, we propose HyperEdit, a framework that adaptively processes each editing request to best align with it. To achieve this, HyperEdit generates request-specific dynamic weights that guide the editing process. The computational overhead of producing these weights is minimized through a carefully designed hypernetwork. With this design, HyperEdit achieves a relatively 9% improvement over the state-of-the-art editing model.
Confidence in LLMs is a useful indicator of model uncertainty and answer reliability. Existing work mainly focused on single-turn scenarios, while research on confidence in complex multi-turn interactions is limited. In this paper, we investigate whether LLM-based search agents have the ability to communicate their own confidence through verbalized confidence scores after long sequences of actions, a significantly more challenging task compared to outputting confidence in a single interaction. Experimenting on open-source agentic models, we first find that models exhibit much higher task accuracy at high confidence while having near-zero accuracy when confidence is low. Based on this observation, we propose Test-Time Scaling (TTS) methods that use confidence scores to determine answer quality, encourage the model to try again until reaching a satisfactory confidence level. Results show that our proposed methods significantly reduce token consumption while demonstrating competitive performance compared to baseline fixed budget TTS methods.
As large language models (LLMs) are increasingly deployed, ensuring their safe use is paramount. Jailbreaking, adversarial prompts that bypass model alignment to trigger harmful outputs, present significant risks, with existing studies reporting high success rates in evading common LLMs. However, previous evaluations have focused solely on the models, neglecting the full deployment pipeline, which typically incorporates additional safety mechanisms like content moderation filters. To address this gap, we present a systematic evaluation of jailbreak attacks targeting LLM safety alignment, assessing their success across the full inference pipeline, including both input and output filtering stages. Our findings yield two key insights: first, nearly all evaluated jailbreak techniques can be detected by at least one safety filter, suggesting that prior assessments may have overestimated the practical success of these attacks; second, while safety filters are effective in detection, there remains room to better balance recall and precision to further optimize protection and user experience.We highlight critical gaps and call for further refinement of detection accuracy and usability in LLM safety systems.
Wikipedia is a critical resource for modern NLP, serving as a rich repository of up-to-date and citation-backed information on a wide variety of subjects. The reliability of Wikipedia—its groundedness in its cited sources—is vital to this purpose. This work analyzes both how grounded Wikipedia is and how readily fine-grained grounding evidence can be retrieved. To this end, we introduce PeopleProfiles—a large-scale, multi-level dataset of claim support annotations on biographical Wikipedia articles. We show that: ~22% of claims in Wikipedia *lead* sections are unsupported by the article body; ~30% of annotated claims in the article *body* are unsupported by their (publicly accessible) sources; and real-world Wikipedia citation practices often differ from documented standards. Finally, we show that complex evidence retrieval remains a challenge—even for recent reasoning rerankers.
State-of-the-art large multi-modal models (LMMs) face challenges when processing high-resolution images, as these inputs are converted into enormous visual tokens, many of which are irrelevant to the downstream task. In this paper, we propose Multi-turn Grounding-based Policy Optimization (MGPO), an end-to-end reinforcement learning (RL) framework that enables LMMs to iteratively focus on key visual regions by automatically cropping sub-images, based on model-predicted grounding coordinates within a multi-turn conversation framework. Compared to supervised fine-tuning (SFT), which requires costly additional grounding annotations, our approach highlights that LMMs can emerge robust grounding abilities during the RL training process, leveraging only a binary reward function derived from the correctness of the final answer. Additionally, we observe that LMMs struggle to autonomously trigger visual grounding during the rollout process. To address this cold start problem, we design a multi-turn conversational template and restrict policy loss computation to model outputs generated across multiple dialogue rounds, thereby promoting stable optimization. Extensive experiments demonstrate that, when trained on standard visual-question-short answering data without grounding annotations, MGPO effectively elicits stronger grounding capabilities compared to GRPO, leading to 5.4% improvement on in-distribution MME-Realworld and 5.2% improvement on the challenging out-of-distribution (OOD) V* Bench. Notably, MGPO post-training on Qwen2.5-VL-7B with 21K samples surpasses OpenAI’s o1 and GPT-4o models on the OOD V* Bench.
PUMA: Projected Universal Multilingual ASR for Low-Resource Settings. Application to Diverse African Languages
PDF ↗Multilingual ASR systems often fail to generalize to low-resource and linguistically diverse languages while remaining costly to scale. We introduce PUMA, a unified multilingual ASR model that improves low-resource performance with reduced model complexity. PUMA employs a Universal Language Projection (ULP) module that integrates a learnable language token with acoustic representations, enabling language-aware processing through shared parameters. Experiments on diverse African languages show consistent word error rate reductions over strong multilingual baselines, highlighting improved robustness and generalization. Our code is available at the following GitHub URL: https://github.com/ilyes-okd/PUMA
Visual text grounding provides interpretable evidence for document question answering. Due to the complex layouts and mixed visual-text contents in text-rich images, effective visual text grounding requires strong visual and spatial reasoning to localize multiple referenced regions. Existing multimodal large language model (MLLM) approaches often struggle to align query tokens with visual–text patches, heavily relying on lengthy OCR inputs. To tackle this problem, we propose Doc-AGround, an OCR-free approach that leverages the MLLM’s inherent multi-head attention for multi-patch grounding. Doc-AGround extracts a patch-wise attention map as the grounding prediction. Concurrently, it introduces an effective multi-head weighting mechanism to amplify the attention heads’ intrinsic role in connecting vision and text. Empirical results of Doc-AGround show state-of-the-art performance on challenging document grounding benchmarks, demonstrating the effectiveness of the proposed attention-based grounding design.
Despite the rapid progress of large language models (LLMs) in code generation, existing evaluations focus on functional correctness or syntactic validity, overlooking how LLMs make critical design choices such as which library or programming language to use.To fill this gap, we perform the first empirical study of LLMs’ preferences for libraries and programming languages when generating code, covering eight diverse LLMs.We observe a strong tendency to overuse widely adopted libraries such as NumPy; in up to 45% of cases, this usage is not required and deviates from the ground-truth solutions.The LLMs we study also show a significant preference toward Python as their default language.For high-performance project initialisation tasks where Python is not the optimal language, it remains the dominant choice in 58% of cases, and Rust is not used once.These results highlight how LLMs prioritise familiarity and popularity over suitability and task-specific optimality;underscoring the need for targeted fine-tuning, data diversification, and evaluation benchmarks that explicitly measure language and library selection fidelity.
The adoption of large language models (LLMs) is transforming the peer review process, from assisting reviewers in writing detailed evaluations to generating entire reviews automatically. While these capabilities offer new opportunities, they also raise concerns about fairness and reliability. In this paper, we investigate bias in LLM-generated peer reviews through controlled interventions on author metadata, including affiliation, gender, seniority, and publication history. Our analysis consistently shows a strong affiliation bias favoring authors from highly ranked institutions. We also identify directional preferences associated with seniority and prior publication record, which can influence acceptance decisions for borderline papers. Gender effects are smaller but present in several models. Notably, implicit biases become more pronounced when examining token-level soft ratings, suggesting that alignment may mask but not fully eliminate underlying preferences.
Diversity Collapse in Multi-Agent LLM Systems: Structural Coupling and Collective Failure in Open-Ended Idea Generation
PDF ↗Multi-agent systems (MAS) are increasingly used for open-ended idea generation, driven by the expectation that collective interaction will broaden the exploration diversity. However, when and why such collaboration truly expands the solution space remains unclear. We present a systematic empirical study of diversity in MAS-based ideation across three bottom-up levels: model intelligence, agent cognition, and system dynamics. At the model level, we identify a compute efficiency paradox, where stronger, highly aligned models yield diminishing marginal diversity despite higher per-sample quality. At the cognition level, authority-driven dynamics suppress semantic diversity compared to junior-dominated groups. At the system level, group-size scaling yields diminishing returns and dense communication topologies accelerate premature convergence. We characterize these outcomes as collective failures emerging from structural coupling, a process where interaction inadvertently contracts agent exploration and triggers diversity collapse. Our analysis shows that this collapse arises primarily from the interaction structure rather than inherent model insufficiency, highlighting the importance of preserving independence and disagreement when designing MAS for creative tasks. Our code is available at https://github.com/Xtra-Computing/MAS_Diversity.
Recent video generation models can produce smooth and visually appealing clips, but they often struggle to synthesize complex dynamics with a coherent chain of consequences. Accurately modeling visual outcomes and state transitions over time remains a core challenge. In contrast, large language and multimodal models (e.g., GPT-4o) exhibit strong visual state reasoning and future prediction capabilities. To bridge these strengths, we introduce VChain, a novel inference-time chain-of-visual-thought framework that injects visual reasoning signals from multimodal models into video generation. Specifically, VChain contains a dedicated pipeline that leverages large multimodal models to generate a sparse set of critical keyframes as snapshots, which are then used to guide the sparse inference-time visual-state adaptation of a pre-trained video generator only at these key moments. Our approach is tuning-efficient, introduces minimal overhead and avoids dense supervision. Extensive experiments on complex, multi-step scenarios show that VChain significantly enhances the quality of generated videos.
Recently, large language models (LLMs) have demonstrated remarkable problem-solving capabilities by autonomously integrating with external tools for collaborative reasoning. However, due to the inherently complex and diverse nature of multimodal information, enabling multimodal large language models (MLLMs) to flexibly and efficiently utilize external tools during reasoning remains an underexplored challenge. In this work, we introduce ToolScope, an agentic framework designed to unify global planning with local multimodal perception, adopting a specialized Perceive tool to mitigates visual context degradation in long-horizon VQA task. ToolScope comprises three primary components: the Global Navigator, the Agentic Executor, and the Response Synthesizer. The Global Navigator functions as a "telescope”, offering high-level strategic guidance. The Agentic Executor operates iteratively to augment MLLM with local perception through the integration of external tools—Search, Code, and Perceive. Finally, the Response Synthesizer consolidates and organizes the reasoning process into a coherent, user-friendly output. We evaluate ToolScope on four VQA benchmarks across diverse domains, including VQA 2.0, ScienceQA, MAT-Search and MathVista. It demonstrates strong generalization capabilities, achieving an average performance improvement of up to +6.69% across all datasets. Our code is available at https://github.com/dengmengjie/ToolScope.
Large language models (LLMs) have emerged as a powerful tool for synthetic data generation. A particularly important use case is producing synthetic replicas of private text, which requires carefully balancing privacy and utility. We propose Realistic and Privacy-Preserving Synthetic Data Generation (RPSG), which uses private seeds and integrates privacy-preserving strategies, including a formal differential privacy (DP) mechanism in the candidate selection, to generate realistic synthetic data. Comprehensive experiments against state-of-the-art private synthetic data generation methods demonstrate that RPSG achieves high fidelity to private data while providing strong privacy protection.
As large language models (LLMs) continue to advance, significant effort is spent on manually crafting prompts to unlock their full potential. While existing prompt optimization methods automate this process, they often underperform due to their reliance on learning exclusively from incorrect samples. We propose the Learning from Contrastive Prompts (LCP) framework, which leverages contrastive prompts to distinguish between high- and low-performing cases. By identifying and amplifying the differences that make prompts effective, LCP systematically extracts principles underlying successful prompt design. On the Big-Bench Hard benchmark, LCP achieves an 87.5% win rate on Claude-3-Sonnet and 75.7% on Claude-4-Sonnet. Experiments on DeepSeek-R1 (88.2% win rate) and SuperGLUE further confirm that LCP generalizes across both proprietary and open-source models and diverse NLU benchmarks.The framework offers a principled and scalable foundation for automated prompt engineering, reducing manual intervention in adapting LLMs to diverse applications.
Video-LLMs face a fundamental tension in long-video reasoning: static, sparse frame sampling either dilutes evidence across task-irrelevant segments at significant cost or misses fine-grained temporal semantics altogether. We propose a novel, cognitively-inspired task — Endomorphic Multimodal Compression (EMC) — as a structurally-constrained sufficient-statistic problem for VideoQA, and formulate it as an endomorphic transformation F_EMC : (V, Q) → (v, q) that compresses the multimodal input while preserving answer invariance across reasonable downstream models. The endomorphic form keeps the compressed output in the downstream pipeline’s native task space — a structural mirror of the filter-then-reason mechanism in the cognitive literature motivating EMC — distinguishing it from latent-code compression (IB / VIB) and making the formulation extensible to other multimodal settings. Under the Markov chain A → (V, Q) → (v, q), EMC realizes the classical sufficiency condition I((v, q); A) = I((V, Q); A) in its VideoQA-natural form. As a modular front-end, EMC plugs into both Video Instruction Tuning and Video Question Answering pipelines. We release the first dedicated benchmark and propose ReSimplifyIt, an EMC baseline surpassing prior methods by 0.40 F-1 with competitive query rewriting. Integrating EMC yields relative gains of 7.33% in training and 33.7% in inference for video-language understanding.
MALLM-GAN: Multi-Agent Large Language Model as Generative Adversarial Network for Synthesizing Tabular Data
PDF ↗In the era of big data, access to abundant data is crucial to driving research forward. However, such data are often inaccessible due to privacy concerns or high costs, particularly in the healthcare domain. Generating synthetic (tabular) data can address this, but existing models typically require substantial amounts of data to train effectively, contradicting our objective of solving data scarcity. To address this challenge, we propose a novel framework to generate synthetic tabular data, powered by large language models (LLMs) that emulates the architecture of a Generative Adversarial Network (GAN). By incorporating the data generation process as contextual information and utilizing LLM as the optimizer, our approach significantly enhances the quality of synthetic data generation in common scenarios with small sample sizes. Our experimental results on public and private datasets demonstrate that our model outperforms several state-of-art models regarding generating higher quality synthetic data for downstream tasks while keeping the privacy of the real data in low data regime. Code is available at https://github.com/yling1105/MALLM-GAN.
Code secrets are sensitive assets for software developers, and their leakage poses significant cybersecurity risks. While the rapid development of AI code assistants powered by Code Large Language Models (CLLMs), CLLMs are shown to inadvertently leak such secrets due to a notorious memorization phenomenon. This study first reveals that Byte-Pair Encoding (BPE) tokenization leads to unexpected behavior of secret memorization, which we term as gibberish bias. Specifically, we identified that some secrets are among the easiest for CLLMs to memorize. These secrets yield high character-level entropy, but low token-level entropy. Then, this paper supports the biased claim with numerical data. We identified that the roots of the bias are the token distribution shift between the CLLM training data and the secret data. We further discuss how gibberish bias manifests under the “larger vocabulary” trend. To conclude the paper, we discuss potential mitigation strategies and the broader implications on current tokenizer design.
Incorporating Large Language Models (LLMs) for downstream tasks has recently garnered considerable attention, where fine-tuning plays a key role in LLMs’ adaptation. These LLMs, often consisting of billions of parameters, require vast amounts of computational resources when customizing them for new tasks. To mitigate this, researchers have proposed the parameter-efficient fine-tuning (PEFT) as a practical solution by adjusting fewer parameters of a pre-trained LLM. However, these methods heavily rely on their own structural modifications that fail to establish an efficient knowledge-sharing mechanism to distill rich knowledge from other expert models, which may lead to inefficient fine-tuning. In this paper, we propose Pen2Sword, a lightweight fine-tuning framework for domain adaptation which efficiently transfers knowledge from a small expert model to a target large model via embedding layers, significantly enhancing the fine-tuning efficiency of large models. Specifically, we first selects optimal expert models via a preserving function, then facilitates knowledge transfer through vocabulary alignment and embedding expansion, and finally accelerates domain adaptation with a fast fine-tuning paradigm. Extensive empirical evaluations across multiple domains demonstrate that our Pen2Sword framework consistently accelerates domain-specific fine-tuning, improves model performance (e.g., +13.6% in code and +20.1% in math), and remains robust across diverse model families and PEFT methods. The codes and data are available at https://github.com/pengmeishu/Pen2Sword.
Large Language Models Are Effective Human Annotation Assistants, But Not Good Independent Annotators
PDF ↗Event annotation is important for identifying, monitoring, and understanding sociological trends. Although expert annotators set the gold standard, they are expensive and inefficient. While state-of-the-art NLP models are an attractive alternative, they are often evaluated on standalone subtasks rather than entire workflows. Thus, we evaluate a holistic workflow that summarizes news with event coreference resolution and argument extraction in three modes: AI-only, AI assistance, and human only. Although AI’s recall is seven times higher than the tf-idf baseline at coreference resolution, it is far from replacing experts. However, experts adopt AI-extracted arguments 60% of the time, reducing extraction time by 25%. Our code and data are in https://github.com/Obertura777/gtd-data.