Authors: Mingyan Liu
Abstract: Generative models, such as GPT and BERT, have significantly improved performance in tasks like text generation and summarization. However, hallucinations "where models generate non-factual or misleading content" are especially problematic in smaller-scale architectures, limiting their real-world applicability.In this paper, we propose a unified Virtual Mixture-of-Experts (MoE) fusion strategy that enhances inference performance and mitigates hallucinations in a single Qwen 1.5 0.5B model without increasing the parameter count. Our method leverages multiple domain-specific expert prompts (with the number of experts being adjustable) to guide the model from different perspectives. We apply a statistical outlier truncation strategy based on the mean and standard deviation to filter out abnormally high probability predictions, and we inject noise into the embedding space to promote output diversity. To clearly assess the contribution of each module, we adopt a fixed voting mechanism rather than a dynamic gating network, thereby avoiding additional confounding factors. We provide detailed theoretical derivations from both statistical and ensemble learning perspectives to demonstrate how our method reduces output variance and suppresses hallucinations. Extensive ablation experiments on dialogue generation tasks show that our approach significantly improves inference accuracy and robustness in small models. Additionally, we discuss methods for evaluating the orthogonality of virtual experts and outline the potential for future work involving dynamic expert weight allocation using gating networks.
Authors: Sifan Li, Yujun Cai, Bryan Hooi, Nanyun Peng, Yiwei Wang
Abstract: Traditional Chinese Medicine (TCM) has seen increasing adoption in healthcare, with specialized Large Language Models (LLMs) emerging to support clinical applications. A fundamental requirement for these models is accurate identification of TCM drug ingredients. In this paper, we evaluate how general and TCM-specialized LLMs perform when identifying ingredients of Chinese drugs. Our systematic analysis reveals consistent failure patterns: models often interpret drug names literally, overuse common herbs regardless of relevance, and exhibit erratic behaviors when faced with unfamiliar formulations. LLMs also fail to understand the verification task. These findings demonstrate that current LLMs rely primarily on drug names rather than possessing systematic pharmacological knowledge. To address these limitations, we propose a Retrieval Augmented Generation (RAG) approach focused on ingredient names. Experiments across 220 TCM formulations show our method significantly improves accuracy from approximately 50% to 82% in ingredient verification tasks. Our work highlights critical weaknesses in current TCM-specific LLMs and offers a practical solution for enhancing their clinical reliability.
Authors: Gon\c{c}alo Faria, Noah A. Smith
Abstract: Increasing test-time computation has emerged as a promising direction for improving language model performance, particularly in scenarios where model finetuning is impractical or impossible due to computational constraints or private model weights. However, existing test-time search methods using a reward model (RM) often degrade in quality as compute scales, due to the over-optimization of what are inherently imperfect reward proxies. We introduce QAlign, a new test-time alignment approach. As we scale test-time compute, QAlign converges to sampling from the optimal aligned distribution for each individual prompt. By adopting recent advances in Markov chain Monte Carlo for text generation, our method enables better-aligned outputs without modifying the underlying model or even requiring logit access. We demonstrate the effectiveness of QAlign on mathematical reasoning benchmarks (GSM8K and GSM-Symbolic) using a task-specific RM, showing consistent improvements over existing test-time compute methods like best-of-n and majority voting. Furthermore, when applied with more realistic RMs trained on the Tulu 3 preference dataset, QAlign outperforms direct preference optimization (DPO), best-of-n, majority voting, and weighted majority voting on a diverse range of datasets (GSM8K, MATH500, IFEval, MMLU-Redux, and TruthfulQA). A practical solution to aligning language models at test time using additional computation without degradation, our approach expands the limits of the capability that can be obtained from off-the-shelf language models without further training.
Authors: Liangwei Yang, Yuhui Xu, Juntao Tan, Doyen Sahoo, Silvio Savarese, Caiming Xiong, Huan Wang, Shelby Heinecke
Abstract: As large language models continue to scale, their growing computational and storage demands pose significant challenges for real-world deployment. In this work, we investigate redundancy within Transformer-based models and propose an entropy-based pruning strategy to enhance efficiency while maintaining performance. Empirical analysis reveals that the entropy of hidden representations decreases in the early blocks but progressively increases across most subsequent blocks. This trend suggests that entropy serves as a more effective measure of information richness within computation blocks. Unlike cosine similarity, which primarily captures geometric relationships, entropy directly quantifies uncertainty and information content, making it a more reliable criterion for pruning. Extensive experiments demonstrate that our entropy-based pruning approach surpasses cosine similarity-based methods in reducing model size while preserving accuracy, offering a promising direction for efficient model deployment.
Authors: Sander Noels, Guillaume Bied, Maarten Buyl, Alexander Rogiers, Yousra Fettach, Jefrey Lijffijt, Tijl De Bie
Abstract: Large Language Models (LLMs) are increasingly deployed as gateways to information, yet their content moderation practices remain underexplored. This work investigates the extent to which LLMs refuse to answer or omit information when prompted on political topics. To do so, we distinguish between hard censorship (i.e., generated refusals, error messages, or canned denial responses) and soft censorship (i.e., selective omission or downplaying of key elements), which we identify in LLMs' responses when asked to provide information on a broad range of political figures. Our analysis covers 14 state-of-the-art models from Western countries, China, and Russia, prompted in all six official United Nations (UN) languages. Our analysis suggests that although censorship is observed across the board, it is predominantly tailored to an LLM provider's domestic audience and typically manifests as either hard censorship or soft censorship (though rarely both concurrently). These findings underscore the need for ideological and geographic diversity among publicly available LLMs, and greater transparency in LLM moderation strategies to facilitate informed user choices. All data are made freely available.
Authors: Wei-Lin Chen, Zhepei Wei, Xinyu Zhu, Shi Feng, Yu Meng
Abstract: Large language models (LLMs) are increasingly used as automatic evaluators in applications such as benchmarking, reward modeling, and self-refinement. Prior work highlights a potential self-preference bias where LLMs favor their own generated responses, a tendency often intensifying with model size and capability. This raises a critical question: Is self-preference detrimental, or does it simply reflect objectively superior outputs from more capable models? Disentangling these has been challenging due to the usage of subjective tasks in previous studies. To address this, we investigate self-preference using verifiable benchmarks (mathematical reasoning, factual knowledge, code generation) that allow objective ground-truth assessment. This enables us to distinguish harmful self-preference (favoring objectively worse responses) from legitimate self-preference (favoring genuinely superior ones). We conduct large-scale experiments under controlled evaluation conditions across diverse model families (e.g., Llama, Qwen, Gemma, Mistral, Phi, GPT, DeepSeek). Our findings reveal three key insights: (1) Better generators are better judges -- LLM evaluators' accuracy strongly correlates with their task performance, and much of the self-preference in capable models is legitimate. (2) Harmful self-preference persists, particularly when evaluator models perform poorly as generators on specific task instances. Stronger models exhibit more pronounced harmful bias when they err, though such incorrect generations are less frequent. (3) Inference-time scaling strategies, such as generating a long Chain-of-Thought before evaluation, effectively reduce the harmful self-preference. These results provide a more nuanced understanding of LLM-based evaluation and practical insights for improving its reliability.
Authors: Abhilekh Borah, Hasnat Md Abdullah, Kangda Wei, Ruihong Huang
Abstract: The rise of Large Language Models (LLMs) has raised questions about their ability to understand climate-related contexts. Though climate change dominates social media, analyzing its multimodal expressions is understudied, and current tools have failed to determine whether LLMs amplify credible solutions or spread unsubstantiated claims. To address this, we introduce CliME (Climate Change Multimodal Evaluation), a first-of-its-kind multimodal dataset, comprising 2579 Twitter and Reddit posts. The benchmark features a diverse collection of humorous memes and skeptical posts, capturing how these formats distill complex issues into viral narratives that shape public opinion and policy discussions. To systematically evaluate LLM performance, we present the Climate Alignment Quotient (CAQ), a novel metric comprising five distinct dimensions: Articulation, Evidence, Resonance, Transition, and Specificity. Additionally, we propose three analytical lenses: Actionability, Criticality, and Justice, to guide the assessment of LLM-generated climate discourse using CAQ. Our findings, based on the CAQ metric, indicate that while most evaluated LLMs perform relatively well in Criticality and Justice, they consistently underperform on the Actionability axis. Among the models evaluated, Claude 3.7 Sonnet achieves the highest overall performance. We publicly release our CliME dataset and code to foster further research in this domain.
Authors: Zixuan Ke, Yifei Ming, Shafiq Joty
Abstract: This tutorial on adaptation of LLMs is designed to address the growing demand for models that go beyond the static capabilities of generic LLMs by providing an overview of dynamic, domain-specific, and task-adaptive LLM adaptation techniques. While general LLMs have demonstrated strong generalization across a variety of tasks, they often struggle to perform well in specialized domains such as finance, healthcare, and code generation for underrepresented languages. Additionally, their static nature limits their ability to evolve with the changing world, and they are often extremely large in size, making them impractical and costly to deploy at scale. As a result, the adaptation of LLMs has drawn much attention since the birth of LLMs and is of core importance, both for industry, which focuses on serving its targeted users, and academia, which can greatly benefit from small but powerful LLMs. To address this gap, this tutorial aims to provide an overview of the LLM adaptation techniques. We start with an introduction to LLM adaptation, from both the data perspective and the model perspective. We then emphasize how the evaluation metrics and benchmarks are different from other techniques. After establishing the problems, we explore various adaptation techniques. We categorize adaptation techniques into two main families. The first is parametric knowledge adaptation, which focuses on updating the parametric knowledge within LLMs. Additionally, we will discuss real-time adaptation techniques, including model editing, which allows LLMs to be updated dynamically in production environments. The second kind of adaptation is semi-parametric knowledge adaptation, where the goal is to update LLM parameters to better leverage external knowledge or tools through techniques like retrieval-augmented generation (RAG) and agent-based systems.
Authors: Dongsuk Jang, Alan Li, Arman Cohan
Abstract: Automated summarization of healthcare community question-answering forums is challenging due to diverse perspectives presented across multiple user responses to each question. The PerAnsSumm Shared Task was therefore proposed to tackle this challenge by identifying perspectives from different answers and then generating a comprehensive answer to the question. In this study, we address the PerAnsSumm Shared Task using two complementary paradigms: (i) a training-based approach through QLoRA fine-tuning of LLaMA-3.3-70B-Instruct, and (ii) agentic approaches including zero- and few-shot prompting with frontier LLMs (LLaMA-3.3-70B-Instruct and GPT-4o) and a Mixture-of-Agents (MoA) framework that leverages a diverse set of LLMs by combining outputs from multi-layer feedback aggregation. For perspective span identification/classification, GPT-4o zero-shot achieves an overall score of 0.57, substantially outperforming the 0.40 score of the LLaMA baseline. With a 2-layer MoA configuration, we were able to improve LLaMA performance up by 28 percent to 0.51. For perspective-based summarization, GPT-4o zero-shot attains an overall score of 0.42 compared to 0.28 for the best LLaMA zero-shot, and our 2-layer MoA approach boosts LLaMA performance by 32 percent to 0.37. Furthermore, in few-shot setting, our results show that the sentence-transformer embedding-based exemplar selection provides more gain than manually selected exemplars on LLaMA models, although the few-shot prompting is not always helpful for GPT-4o. The YaleNLP team's approach ranked the overall second place in the shared task.
Authors: Samuele Marro, Davide Evangelista, X. Angelo Huang, Emanuele La Malfa, Michele Lombardi, Michael Wooldridge
Abstract: Language is typically modelled with discrete sequences. However, the most successful approaches to language modelling, namely neural networks, are continuous and smooth function approximators. In this work, we show that Transformer-based language models implicitly learn to represent sentences as continuous-time functions defined over a continuous input space. This phenomenon occurs in most state-of-the-art Large Language Models (LLMs), including Llama2, Llama3, Phi3, Gemma, Gemma2, and Mistral, and suggests that LLMs reason about language in ways that fundamentally differ from humans. Our work formally extends Transformers to capture the nuances of time and space continuity in both input and output space. Our results challenge the traditional interpretation of how LLMs understand language, with several linguistic and engineering implications.
Authors: Simon A. Lee, Anthony Wu, Jeffrey N. Chiang
Abstract: We introduce Clinical ModernBERT, a transformer based encoder pretrained on large scale biomedical literature, clinical notes, and medical ontologies, incorporating PubMed abstracts, MIMIC IV clinical data, and medical codes with their textual descriptions. Building on ModernBERT the current state of the art natural language text encoder featuring architectural upgrades such as rotary positional embeddings (RoPE), Flash Attention, and extended context length up to 8,192 tokens our model adapts these innovations specifically for biomedical and clinical domains. Clinical ModernBERT excels at producing semantically rich representations tailored for long context tasks. We validate this both by analyzing its pretrained weights and through empirical evaluation on a comprehensive suite of clinical NLP benchmarks.
Authors: Amit K Verma, Zhisong Zhang, Junwon Seo, Robin Kuo, Runbo Jiang, Emma Strubell, Anthony D Rollett
Abstract: With the advent of large language models (LLMs), the vast unstructured text within millions of academic papers is increasingly accessible for materials discovery, although significant challenges remain. While LLMs offer promising few- and zero-shot learning capabilities, particularly valuable in the materials domain where expert annotations are scarce, general-purpose LLMs often fail to address key materials-specific queries without further adaptation. To bridge this gap, fine-tuning LLMs on human-labeled data is essential for effective structured knowledge extraction. In this study, we introduce a novel annotation schema designed to extract generic process-structure-properties relationships from scientific literature. We demonstrate the utility of this approach using a dataset of 128 abstracts, with annotations drawn from two distinct domains: high-temperature materials (Domain I) and uncertainty quantification in simulating materials microstructure (Domain II). Initially, we developed a conditional random field (CRF) model based on MatBERT, a domain-specific BERT variant, and evaluated its performance on Domain I. Subsequently, we compared this model with a fine-tuned LLM (GPT-4o from OpenAI) under identical conditions. Our results indicate that fine-tuning LLMs can significantly improve entity extraction performance over the BERT-CRF baseline on Domain I. However, when additional examples from Domain II were incorporated, the performance of the BERT-CRF model became comparable to that of the GPT-4o model. These findings underscore the potential of our schema for structured knowledge extraction and highlight the complementary strengths of both modeling approaches.
Authors: Siddharth Srikanth, Varun Bhatt, Boshen Zhang, Werner Hager, Charles Michael Lewis, Katia P. Sycara, Aaquib Tabrez, Stefanos Nikolaidis
Abstract: Understanding how humans collaborate and communicate in teams is essential for improving human-agent teaming and AI-assisted decision-making. However, relying solely on data from large-scale user studies is impractical due to logistical, ethical, and practical constraints, necessitating synthetic models of multiple diverse human behaviors. Recently, agents powered by Large Language Models (LLMs) have been shown to emulate human-like behavior in social settings. But, obtaining a large set of diverse behaviors requires manual effort in the form of designing prompts. On the other hand, Quality Diversity (QD) optimization has been shown to be capable of generating diverse Reinforcement Learning (RL) agent behavior. In this work, we combine QD optimization with LLM-powered agents to iteratively search for prompts that generate diverse team behavior in a long-horizon, multi-step collaborative environment. We first show, through a human-subjects experiment (n=54 participants), that humans exhibit diverse coordination and communication behavior in this domain. We then show that our approach can effectively replicate trends from human teaming data and also capture behaviors that are not easily observed without collecting large amounts of data. Our findings highlight the combination of QD and LLM-powered agents as an effective tool for studying teaming and communication strategies in multi-agent collaboration.
Authors: Essential AI, :, Darsh J Shah, Peter Rushton, Somanshu Singla, Mohit Parmar, Kurt Smith, Yash Vanjani, Ashish Vaswani, Adarsh Chaluvaraju, Andrew Hojel, Andrew Ma, Anil Thomas, Anthony Polloreno, Ashish Tanwer, Burhan Drak Sibai, Divya S Mansingka, Divya Shivaprasad, Ishaan Shah, Karl Stratos, Khoi Nguyen, Michael Callahan, Michael Pust, Mrinal Iyer, Philip Monk, Platon Mazarakis, Ritvik Kapila, Saurabh Srivastava, Tim Romanski
Abstract: A language model's ability to reflect on its own reasoning provides a key advantage for solving complex problems. While most recent research has focused on how this ability develops during reinforcement learning, we show that it actually begins to emerge much earlier - during the model's pre-training. To study this, we introduce deliberate errors into chains-of-thought and test whether the model can still arrive at the correct answer by recognizing and correcting these mistakes. By tracking performance across different stages of pre-training, we observe that this self-correcting ability appears early and improves steadily over time. For instance, an OLMo2-7B model pre-trained on 4 trillion tokens displays self-correction on our six self-reflection tasks.
Authors: Kaung Lwin Thant, Kwankamol Nongpong, Ye Kyaw Thu, Thura Aung, Khaing Hsu Wai, Thazin Myint Oo
Abstract: Named Entity Recognition (NER) involves identifying and categorizing named entities within textual data. Despite its significance, NER research has often overlooked low-resource languages like Myanmar (Burmese), primarily due to the lack of publicly available annotated datasets. To address this, we introduce myNER, a novel word-level NER corpus featuring a 7-tag annotation scheme, enriched with Part-of-Speech (POS) tagging to provide additional syntactic information. Alongside the corpus, we conduct a comprehensive evaluation of NER models, including Conditional Random Fields (CRF), Bidirectional LSTM (BiLSTM)-CRF, and their combinations with fastText embeddings in different settings. Our experiments reveal the effectiveness of contextualized word embeddings and the impact of joint training with POS tagging, demonstrating significant performance improvements across models. The traditional CRF joint-task model with fastText embeddings as a feature achieved the best result, with a 0.9818 accuracy and 0.9811 weighted F1 score with 0.7429 macro F1 score. BiLSTM-CRF with fine-tuned fastText embeddings gets the best result of 0.9791 accuracy and 0.9776 weighted F1 score with 0.7395 macro F1 score.
Authors: Kepu Zhang, Weijie Yu, Zhongxiang Sun, Jun Xu
Abstract: Syllogistic reasoning is a fundamental aspect of legal decision-making, enabling logical conclusions by connecting general legal principles with specific case facts. Although existing large language models (LLMs) can generate responses to legal questions, they fail to perform explicit syllogistic reasoning, often producing implicit and unstructured answers that lack explainability and trustworthiness. To address this limitation, we propose SyLeR, a novel framework that empowers LLMs to engage in explicit syllogistic legal reasoning. SyLeR integrates a tree-structured hierarchical retrieval mechanism to effectively combine relevant legal statutes and precedent cases, forming comprehensive major premises. This is followed by a two-stage fine-tuning process: supervised fine-tuning warm-up establishes a foundational understanding of syllogistic reasoning, while reinforcement learning with a structure-aware reward mechanism refines the ability of the model to generate diverse logically sound and well-structured reasoning paths. We conducted extensive experiments across various dimensions, including in-domain and cross-domain user groups (legal laypersons and practitioners), multiple languages (Chinese and French), and different LLM backbones (legal-specific and open-domain LLMs). The results show that SyLeR significantly improves response accuracy and consistently delivers explicit, explainable, and trustworthy legal reasoning.
Authors: Kang Xue, Ming Dong, Xinhui Tu, Tingting He
Abstract: The rapid growth in the parameter size of Large Language Models (LLMs) has led to the development of Parameter-Efficient Fine-Tuning (PEFT) methods to alleviate the computational costs of fine-tuning. Among these, Fisher Induced Sparse uncHanging (FISH) Mask is a selection-based PEFT technique that identifies a subset of pre-trained parameters for fine-tuning based on approximate Fisher information. However, the integration of FISH Mask with other PEFT methods, such as LoRA and Adapters, remains underexplored. In this paper, we propose FISH-Tuning, a novel approach that incorporates FISH Mask into addition-based and reparameterization-based PEFT methods, including LoRA, Adapters, and their variants. By leveraging Fisher information to select critical parameters within these methods, FISH-Tuning achieves superior performance without additional memory overhead or inference latency. Experimental results across various datasets and pre-trained models demonstrate that FISH-Tuning consistently outperforms the vanilla PEFT methods with the same proportion of trainable parameters.
Authors: Yuhao Wang, Heyang Liu, Ziyang Cheng, Ronghua Wu, Qunshan Gu, Yanfeng Wang, Yu Wang
Abstract: Speech large language models (LLMs) have emerged as a prominent research focus in speech processing. We propose VocalNet-1B and VocalNet-8B, a series of high-performance, low-latency speech LLMs enabled by a scalable and model-agnostic training framework for real-time voice interaction. Departing from the conventional next-token prediction (NTP), we introduce multi-token prediction (MTP), a novel approach optimized for speech LLMs that simultaneously improves generation speed and quality. Experiments show that VocalNet outperforms mainstream Omni LLMs despite using significantly less training data, while also surpassing existing open-source speech LLMs by a substantial margin. To support reproducibility and community advancement, we will open-source all model weights, inference code, training data, and framework implementations upon publication.
Authors: Bing Wang, Bingrui Zhao, Ximing Li, Changchun Li, Wanfu Gao, Shengsheng Wang
Abstract: Over the past decade, social media platforms have been key in spreading rumors, leading to significant negative impacts. To counter this, the community has developed various Rumor Detection (RD) algorithms to automatically identify them using user comments as evidence. However, these RD methods often fail in the early stages of rumor propagation when only limited user comments are available, leading the community to focus on a more challenging topic named Rumor Early Detection (RED). Typically, existing RED methods learn from limited semantics in early comments. However, our preliminary experiment reveals that the RED models always perform best when the number of training and test comments is consistent and extensive. This inspires us to address the RED issue by generating more human-like comments to support this hypothesis. To implement this idea, we tune a comment generator by simulating expert collaboration and controversy and propose a new RED framework named CAMERED. Specifically, we integrate a mixture-of-expert structure into a generative language model and present a novel routing network for expert collaboration. Additionally, we synthesize a knowledgeable dataset and design an adversarial learning strategy to align the style of generated comments with real-world comments. We further integrate generated and original comments with a mutual controversy fusion module. Experimental results show that CAMERED outperforms state-of-the-art RED baseline models and generation methods, demonstrating its effectiveness.
Authors: Aviv Brokman, Xuguang Ai, Yuhang Jiang, Shashank Gupta, Ramakanth Kavuluru
Abstract: Objective: Zero-shot methodology promises to cut down on costs of dataset annotation and domain expertise needed to make use of NLP. Generative large language models trained to align with human goals have achieved high zero-shot performance across a wide variety of tasks. As of yet, it is unclear how well these models perform on biomedical relation extraction (RE). To address this knowledge gap, we explore patterns in the performance of OpenAI LLMs across a diverse sampling of RE tasks. Methods: We use OpenAI GPT-4-turbo and their reasoning model o1 to conduct end-to-end RE experiments on seven datasets. We use the JSON generation capabilities of GPT models to generate structured output in two ways: (1) by defining an explicit schema describing the structure of relations, and (2) using a setting that infers the structure from the prompt language. Results: Our work is the first to study and compare the performance of the GPT-4 and o1 for the end-to-end zero-shot biomedical RE task across a broad array of datasets. We found the zero-shot performances to be proximal to that of fine-tuned methods. The limitations of this approach are that it performs poorly on instances containing many relations and errs on the boundaries of textual mentions. Conclusion: Recent large language models exhibit promising zero-shot capabilities in complex biomedical RE tasks, offering competitive performance with reduced dataset curation and NLP modeling needs at the cost of increased computing, potentially increasing medical community accessibility. Addressing the limitations we identify could further boost reliability. The code, data, and prompts for all our experiments are publicly available: https://github.com/bionlproc/ZeroShotRE
Authors: Michael J Bommarito, Daniel Martin Katz, Jillian Bommarito
Abstract: We present NUPunkt and CharBoundary, two sentence boundary detection libraries optimized for high-precision, high-throughput processing of legal text in large-scale applications such as due diligence, e-discovery, and legal research. These libraries address the critical challenges posed by legal documents containing specialized citations, abbreviations, and complex sentence structures that confound general-purpose sentence boundary detectors. Our experimental evaluation on five diverse legal datasets comprising over 25,000 documents and 197,000 annotated sentence boundaries demonstrates that NUPunkt achieves 91.1% precision while processing 10 million characters per second with modest memory requirements (432 MB). CharBoundary models offer balanced and adjustable precision-recall tradeoffs, with the large model achieving the highest F1 score (0.782) among all tested methods. Notably, NUPunkt provides a 29-32% precision improvement over general-purpose tools while maintaining exceptional throughput, processing multi-million document collections in minutes rather than hours. Both libraries run efficiently on standard CPU hardware without requiring specialized accelerators. NUPunkt is implemented in pure Python with zero external dependencies, while CharBoundary relies only on scikit-learn and optional ONNX runtime integration for optimized performance. Both libraries are available under the MIT license, can be installed via PyPI, and can be interactively tested at https://sentences.aleainstitute.ai/. These libraries address critical precision issues in retrieval-augmented generation systems by preserving coherent legal concepts across sentences, where each percentage improvement in precision yields exponentially greater reductions in context fragmentation, creating cascading benefits throughout retrieval pipelines and significantly enhancing downstream reasoning quality.
Authors: Yougang Lyu, Shijie Ren, Yue Feng, Zihan Wang, Zhumin Chen, Zhaochun Ren, Maarten de Rijke
Abstract: Large language models (LLMs) have shown potential in supporting decision-making applications, particularly as personal conversational assistants in the financial, healthcare, and legal domains. While prompt engineering strategies have enhanced the capabilities of LLMs in decision-making, cognitive biases inherent to LLMs present significant challenges. Cognitive biases are systematic patterns of deviation from norms or rationality in decision-making that can lead to the production of inaccurate outputs. Existing cognitive bias mitigation strategies assume that input prompts contain (exactly) one type of cognitive bias and therefore fail to perform well in realistic settings where there maybe any number of biases. To fill this gap, we propose a cognitive debiasing approach, called self-debiasing, that enhances the reliability of LLMs by iteratively refining prompts. Our method follows three sequential steps -- bias determination, bias analysis, and cognitive debiasing -- to iteratively mitigate potential cognitive biases in prompts. Experimental results on finance, healthcare, and legal decision-making tasks, using both closed-source and open-source LLMs, demonstrate that the proposed self-debiasing method outperforms both advanced prompt engineering methods and existing cognitive debiasing techniques in average accuracy under no-bias, single-bias, and multi-bias settings.
Authors: Kees van Deemter
Abstract: In this very personal workography, I relate my 40-year experiences as a researcher and educator in and around Artificial Intelligence (AI), more specifically Natural Language Processing. I describe how curiosity, and the circumstances of the day, led me to work in both industry and academia, and in various countries, including The Netherlands (Amsterdam, Eindhoven, and Utrecht), the USA (Stanford), England (Brighton), Scotland (Aberdeen), and China (Beijing and Harbin). People and anecdotes play a large role in my story; the history of AI forms its backdrop. I focus on things that might be of interest to (even) younger colleagues, given the choices they face in their own work and life at a time when AI is finally emerging from the shadows.
Authors: Yidong Wang
Abstract: The Needle In A Haystack (NIAH) task has been widely used to evaluate the long-context question-answering capabilities of Large Language Models (LLMs). However, its reliance on simple retrieval limits its effectiveness. To address this limitation, recent studies have introduced the Multiple Needles In A Haystack Reasoning (MNIAH-R) task, which incorporates supporting documents (Multiple needles) of multi-hop reasoning tasks into a distracting context (Haystack}). Despite this advancement, existing approaches still fail to address the issue of models providing direct answers from internal knowledge, and they do not explain or mitigate the decline in accuracy as context length increases. In this paper, we tackle the memory-based answering problem by filtering out direct-answer questions, and we reveal that performance degradation is primarily driven by the reduction in the length of the thinking process as the input length increases. Building on this insight, we decompose the thinking process into retrieval and reasoning stages and introduce a reflection mechanism for multi-round extension. We also train a model using the generated iterative thinking process, which helps mitigate the performance degradation. Furthermore, we demonstrate the application of this retrieval-reflection capability in mathematical reasoning scenarios, improving GPT-4o's performance on AIME2024.
Authors: Kazuki Yano, Takumi Ito, Jun Suzuki
Abstract: Pre-training large language models (LLMs) faces significant memory challenges due to the large size of model parameters. We introduce STaged parameter-Efficient Pre-training (STEP), which integrates parameter-efficient tuning techniques with model growth. We conduct experiments on pre-training LLMs of various sizes and demonstrate that STEP achieves up to a 53.9% reduction in maximum memory requirements compared to vanilla pre-training while maintaining equivalent performance. Furthermore, we show that the model by STEP performs comparably to vanilla pre-trained models on downstream tasks after instruction tuning.
Authors: Zihao Li, Shaoxiong Ji, Hengyu Luo, J\"org Tiedemann
Abstract: Large Language Models (LLMs) exhibit significant disparities in performance across languages, primarily benefiting high-resource languages while marginalizing underrepresented ones. Continual Pretraining (CPT) has emerged as a promising approach to address this imbalance, although the relative effectiveness of monolingual, bilingual, and code-augmented data strategies remains unclear. This study systematically evaluates 36 CPT configurations involving three multilingual base models, across 30+ languages categorized as altruistic, selfish, and stagnant, spanning various resource levels. Our findings reveal three major insights: (1) Bilingual CPT improves multilingual classification but often causes language mixing issues during generation. (2) Including programming code data during CPT consistently enhances multilingual classification accuracy, particularly benefiting low-resource languages, but introduces a trade-off by slightly degrading generation quality. (3) Contrary to prior work, we observe substantial deviations from language classifications according to their impact on cross-lingual transfer: Languages classified as altruistic often negatively affect related languages, selfish languages show conditional and configuration-dependent behavior, and stagnant languages demonstrate surprising adaptability under certain CPT conditions. These nuanced interactions emphasize the complexity of multilingual representation learning, underscoring the importance of systematic studies on generalizable language classification to inform future multilingual CPT strategies.
Authors: Hengyu Luo, Zihao Li, Joseph Attieh, Sawal Devkota, Ona de Gibert, Shaoxiong Ji, Peiqin Lin, Bhavani Sai Praneeth Varma Mantina, Ananda Sreenidhi, Ra\'ul V\'azquez, Mengjie Wang, Samea Yusofi, J\"org Tiedemann
Abstract: Large language models (LLMs) are advancing at an unprecedented pace globally, with regions increasingly adopting these models for applications in their primary language. Evaluation of these models in diverse linguistic environments, especially in low-resource languages, has become a major challenge for academia and industry. Existing evaluation frameworks are disproportionately focused on English and a handful of high-resource languages, thereby overlooking the realistic performance of LLMs in multilingual and lower-resource scenarios. To address this gap, we introduce GlotEval, a lightweight framework designed for massively multilingual evaluation. Supporting seven key tasks (machine translation, text classification, summarization, open-ended generation, reading comprehension, sequence labeling, and intrinsic evaluation), spanning over dozens to hundreds of languages, GlotEval highlights consistent multilingual benchmarking, language-specific prompt templates, and non-English-centric machine translation. This enables a precise diagnosis of model strengths and weaknesses in diverse linguistic contexts. A multilingual translation case study demonstrates GlotEval's applicability for multilingual and language-specific evaluations.
Authors: Jimmy Wang, Thomas Zollo, Richard Zemel, Hongseok Namkoong
Abstract: Eliciting information to reduce uncertainty about a latent entity is a critical task in many application domains, e.g., assessing individual student learning outcomes, diagnosing underlying diseases, or learning user preferences. Though natural language is a powerful medium for this purpose, large language models (LLMs) and existing fine-tuning algorithms lack mechanisms for strategically gathering information to refine their own understanding of the latent entity. To harness the generalization power and world knowledge of LLMs in developing effective information-gathering strategies, we propose an adaptive elicitation framework that actively reduces uncertainty on the latent entity. Since probabilistic modeling of an abstract latent entity is difficult, our framework adopts a predictive view of uncertainty, using a meta-learned language model to simulate future observations and enable scalable uncertainty quantification over complex natural language. Through autoregressive forward simulation, our model quantifies how new questions reduce epistemic uncertainty, enabling the development of sophisticated information-gathering strategies to choose the most informative next queries. In experiments on the 20 questions game, dynamic opinion polling, and adaptive student assessment, our method consistently outperforms baselines in identifying critical unknowns and improving downstream predictions, illustrating the promise of strategic information gathering in natural language settings.
Authors: Vishnu Kabir Chhabra, Mohammad Mahdi Khalili
Abstract: The rapid growth of large language models has spurred significant interest in model compression as a means to enhance their accessibility and practicality. While extensive research has explored model compression through the lens of safety, findings suggest that safety-aligned models often lose elements of trustworthiness post-compression. Simultaneously, the field of mechanistic interpretability has gained traction, with notable discoveries, such as the identification of a single direction in the residual stream mediating refusal behaviors across diverse model architectures. In this work, we investigate the safety of compressed models by examining the mechanisms of refusal, adopting a novel interpretability-driven perspective to evaluate model safety. Furthermore, leveraging insights from our interpretability analysis, we propose a lightweight, computationally efficient method to enhance the safety of compressed models without compromising their performance or utility.
Authors: Yuantao Zhang, Zhankui Yang
Abstract: The rise of Large Language Models (LLMs) has brought about concerns regarding copyright infringement and unethical practices in data and model usage. For instance, slight modifications to existing LLMs may be used to falsely claim the development of new models, leading to issues of model copying and violations of ownership rights. This paper addresses these challenges by introducing a novel metric for quantifying LLM similarity, which leverages perplexity curves and differences in Menger curvature. Comprehensive experiments validate the performance of our methodology, demonstrating its superiority over baseline methods and its ability to generalize across diverse models and domains. Furthermore, we highlight the capability of our approach in detecting model replication through simulations, emphasizing its potential to preserve the originality and integrity of LLMs. Code is available at https://github.com/zyttt-coder/LLM_similarity.
Authors: Yuheng Wu, Wentao Guo, Zirui Liu, Heng Ji, Zhaozhuo Xu, Denghui Zhang
Abstract: This paper investigates the emergence of Theory-of-Mind (ToM) capabilities in large language models (LLMs) from a mechanistic perspective, focusing on the role of extremely sparse parameter patterns. We introduce a novel method to identify ToM-sensitive parameters and reveal that perturbing as little as 0.001% of these parameters significantly degrades ToM performance while also impairing contextual localization and language understanding. To understand this effect, we analyze their interaction with core architectural components of LLMs. Our findings demonstrate that these sensitive parameters are closely linked to the positional encoding module, particularly in models using Rotary Position Embedding (RoPE), where perturbations disrupt dominant-frequency activations critical for contextual processing. Furthermore, we show that perturbing ToM-sensitive parameters affects LLM's attention mechanism by modulating the angle between queries and keys under positional encoding. These insights provide a deeper understanding of how LLMs acquire social reasoning abilities, bridging AI interpretability with cognitive science. Our results have implications for enhancing model alignment, mitigating biases, and improving AI systems designed for human interaction.
Authors: Mingyang Wang, Heike Adel, Lukas Lange, Yihong Liu, Ercong Nie, Jannik Str\"otgen, Hinrich Sch\"utze
Abstract: Multilingual language models (MLMs) store factual knowledge across languages but often struggle to provide consistent responses to semantically equivalent prompts in different languages. While previous studies point out this cross-lingual inconsistency issue, the underlying causes remain unexplored. In this work, we use mechanistic interpretability methods to investigate cross-lingual inconsistencies in MLMs. We find that MLMs encode knowledge in a language-independent concept space through most layers, and only transition to language-specific spaces in the final layers. Failures during the language transition often result in incorrect predictions in the target language, even when the answers are correct in other languages. To mitigate this inconsistency issue, we propose a linear shortcut method that bypasses computations in the final layers, enhancing both prediction accuracy and cross-lingual consistency. Our findings shed light on the internal mechanisms of MLMs and provide a lightweight, effective strategy for producing more consistent factual outputs.
Authors: T\'ulio Sousa de Gois, Paloma Batista Cardoso
Abstract: Negation is a universal feature of natural languages. In Brazilian Portuguese, the most commonly used negation particle is n\~ao, which can scope over nouns or verbs. When it scopes over a verb, n\~ao can occur in three positions: pre-verbal (NEG1), double negation (NEG2), or post-verbal (NEG3), e.g., n\~ao gosto, n\~ao gosto n\~ao, gosto n\~ao ("I do not like it"). From a variationist perspective, these structures are different forms of expressing negation. Pragmatically, they serve distinct communicative functions, such as politeness and modal evaluation. Despite their grammatical acceptability, these forms differ in frequency. NEG1 dominates across Brazilian regions, while NEG2 and NEG3 appear more rarely, suggesting its use is contextually restricted. This low-frequency challenges research, often resulting in subjective, non-generalizable interpretations of verbal negation with n\~ao. To address this, we developed negativas, a tool for automatically identifying NEG1, NEG2, and NEG3 in transcribed data. The tool's development involved four stages: i) analyzing a dataset of 22 interviews from the Falares Sergipanos database, annotated by three linguists, ii) creating a code using natural language processing (NLP) techniques, iii) running the tool, iv) evaluating accuracy. Inter-annotator consistency, measured using Fleiss' Kappa, was moderate (0.57). The tool identified 3,338 instances of n\~ao, classifying 2,085 as NEG1, NEG2, or NEG3, achieving a 93% success rate. However, negativas has limitations. NEG1 accounted for 91.5% of identified structures, while NEG2 and NEG3 represented 7.2% and 1.2%, respectively. The tool struggled with NEG2, sometimes misclassifying instances as overlapping structures (NEG1/NEG2/NEG3).
Authors: Hongchao Fang, Can Qin, Ran Xu, Feng Liu, Yixin Liu, Lichao Sun, Dongwon Lee, Lifu Huang, Wenpeng Yin
Abstract: AI-generated content is becoming increasingly prevalent in the real world, leading to serious ethical and societal concerns. For instance, adversaries might exploit large multimodal models (LMMs) to create images that violate ethical or legal standards, while paper reviewers may misuse large language models (LLMs) to generate reviews without genuine intellectual effort. While prior work has explored detecting AI-generated images and texts, and occasionally tracing their source models, there is a lack of a systematic and fine-grained comparative study. Important dimensions--such as AI-generated images vs. text, fully vs. partially AI-generated images, and general vs. malicious use cases--remain underexplored. Furthermore, whether AI systems like GPT-4o can explain why certain forged content is attributed to specific generative models is still an open question, with no existing benchmark addressing this. To fill this gap, we introduce AI-FAKER, a comprehensive multimodal dataset with over 280,000 samples spanning multiple LLMs and LMMs, covering both general and malicious use cases for AI-generated images and texts. Our experiments reveal two key findings: (i) AI authorship detection depends not only on the generated output but also on the model's original training intent; and (ii) GPT-4o provides highly consistent but less specific explanations when analyzing content produced by OpenAI's own models, such as DALL-E and GPT-4o itself.
Authors: Jie Yang, Yiqiu Tang, Yongjie Li, Lihua Zhang, Haoran Zhang
Abstract: Large language models (LLMs) have emerged as powerful tools in the field of finance, particularly for risk management across different asset classes. In this work, we introduce a Cross-Asset Risk Management framework that utilizes LLMs to facilitate real-time monitoring of equity, fixed income, and currency markets. This innovative approach enables dynamic risk assessment by aggregating diverse data sources, ultimately enhancing decision-making processes. Our model effectively synthesizes and analyzes market signals to identify potential risks and opportunities while providing a holistic view of asset classes. By employing advanced analytics, we leverage LLMs to interpret financial texts, news articles, and market reports, ensuring that risks are contextualized within broader market narratives. Extensive backtesting and real-time simulations validate the framework, showing increased accuracy in predicting market shifts compared to conventional methods. The focus on real-time data integration enhances responsiveness, allowing financial institutions to manage risks adeptly under varying market conditions and promoting financial stability through the advanced application of LLMs in risk analysis.
Authors: Jie Yang, Yiqiu Tang, Yongjie Li, Lihua Zhang, Haoran Zhang
Abstract: Dynamic hedging strategies are essential for effective risk management in derivatives markets, where volatility and market sentiment can greatly impact performance. This paper introduces a novel framework that leverages large language models (LLMs) for sentiment analysis and news analytics to inform hedging decisions. By analyzing textual data from diverse sources like news articles, social media, and financial reports, our approach captures critical sentiment indicators that reflect current market conditions. The framework allows for real-time adjustments to hedging strategies, adapting positions based on continuous sentiment signals. Backtesting results on historical derivatives data reveal that our dynamic hedging strategies achieve superior risk-adjusted returns compared to conventional static approaches. The incorporation of LLM-driven sentiment analysis into hedging practices presents a significant advancement in decision-making processes within derivatives trading. This research showcases how sentiment-informed dynamic hedging can enhance portfolio management and effectively mitigate associated risks.
Authors: Weiwei Sun, Shengyu Feng, Shanda Li, Yiming Yang
Abstract: Although LLM-based agents have attracted significant attention in domains such as software engineering and machine learning research, their role in advancing combinatorial optimization (CO) remains relatively underexplored. This gap underscores the need for a deeper understanding of their potential in tackling structured, constraint-intensive problems-a pursuit currently limited by the absence of comprehensive benchmarks for systematic investigation. To address this, we introduce CO-Bench, a benchmark suite featuring 36 real-world CO problems drawn from a broad range of domains and complexity levels. CO-Bench includes structured problem formulations and curated data to support rigorous investigation of LLM agents. We evaluate multiple agent frameworks against established human-designed algorithms, revealing key strengths and limitations of current approaches and identifying promising directions for future research. CO-Bench is publicly available at https://github.com/sunnweiwei/CO-Bench.
Authors: Justin Miller, Tristram Alexander
Abstract: The challenge of clustering short text data lies in balancing informativeness with interpretability. Traditional evaluation metrics often overlook this trade-off. Inspired by linguistic principles of communicative efficiency, this paper investigates the optimal number of clusters by quantifying the trade-off between informativeness and cognitive simplicity. We use large language models (LLMs) to generate cluster names and evaluate their effectiveness through semantic density, information theory, and clustering accuracy. Our results show that Gaussian Mixture Model (GMM) clustering on embeddings generated by a LLM, increases semantic density compared to random assignment, effectively grouping similar bios. However, as clusters increase, interpretability declines, as measured by a generative LLM's ability to correctly assign bios based on cluster names. A logistic regression analysis confirms that classification accuracy depends on the semantic similarity between bios and their assigned cluster names, as well as their distinction from alternatives. These findings reveal a "Goldilocks zone" where clusters remain distinct yet interpretable. We identify an optimal range of 16-22 clusters, paralleling linguistic efficiency in lexical categorization. These insights inform both theoretical models and practical applications, guiding future research toward optimising cluster interpretability and usefulness.
Authors: Juan Sosa, Alejandro Urrego, Cesar Prieto, Emma J. Camargo-D\'iaz
Abstract: Case 03 of the Special Jurisdiction for Peace (JEP), focused on the so-called false positives in Colombia, represents one of the most harrowing episodes of the Colombian armed conflict. This article proposes an innovative methodology based on natural language analysis and semantic co-occurrence models to explore, systematize, and visualize narrative patterns present in the public hearings of victims and appearing parties. By constructing skipgram networks and analyzing their modularity, the study identifies thematic clusters that reveal regional and procedural status differences, providing empirical evidence on dynamics of victimization, responsibility, and acknowledgment in this case. This computational approach contributes to the collective construction of both judicial and extrajudicial truth, offering replicable tools for other transitional justice cases. The work is grounded in the pillars of truth, justice, reparation, and non-repetition, proposing a critical and in-depth reading of contested memories.
Authors: Quan Shi, Carlos Jimenez, Stephen Dong, Brian Seo, Caden Yao, Adam Kelch, Karthik Narasimhan
Abstract: As language models achieve increasingly human-like capabilities in conversational text generation, a critical question emerges: to what extent can these systems simulate the characteristics of specific individuals? To evaluate this, we introduce IMPersona, a framework for evaluating LMs at impersonating specific individuals' writing style and personal knowledge. Using supervised fine-tuning and a hierarchical memory-inspired retrieval system, we demonstrate that even modestly sized open-source models, such as Llama-3.1-8B-Instruct, can achieve impersonation abilities at concerning levels. In blind conversation experiments, participants (mis)identified our fine-tuned models with memory integration as human in 44.44% of interactions, compared to just 25.00% for the best prompting-based approach. We analyze these results to propose detection methods and defense strategies against such impersonation attempts. Our findings raise important questions about both the potential applications and risks of personalized language models, particularly regarding privacy, security, and the ethical deployment of such technologies in real-world contexts.
Authors: Yuya Ogasa, Yuki Arase
Abstract: This study tackles token-level hallucination detection in outputs of large language models. Previous studies revealed that attention exhibits irregular patterns when hallucination occurs. Inspired by this, we extract features from the attention matrix that provide complementary views of (a) the average attention each token receives, which helps identify whether certain tokens are overly influential or ignored, (b) the diversity of attention each token receives, which reveals whether attention is biased toward specific subsets, and (c) the diversity of tokens a token attends to during generation, which indicates whether the model references a narrow or broad range of information. These features are input to a Transformer-based classifier to conduct token-level classification to identify hallucinated spans. Experimental results indicate that the proposed method outperforms strong baselines on hallucination detection with longer input contexts, i.e., data-to-text and summarization tasks.
Authors: Cong Sun, Kurt Teichman, Yiliang Zhou, Brian Critelli, David Nauheim, Graham Keir, Xindi Wang, Judy Zhong, Adam E Flanders, George Shih, Yifan Peng
Abstract: In this retrospective study, a dataset was constructed with two parts. The first part included 1,656 synthetic chest radiology reports generated by GPT-4 using specified prompts, with 828 being error-free synthetic reports and 828 containing errors. The second part included 614 reports: 307 error-free reports between 2011 and 2016 from the MIMIC-CXR database and 307 corresponding synthetic reports with errors generated by GPT-4 on the basis of these MIMIC-CXR reports and specified prompts. All errors were categorized into four types: negation, left/right, interval change, and transcription errors. Then, several models, including Llama-3, GPT-4, and BiomedBERT, were refined using zero-shot prompting, few-shot prompting, or fine-tuning strategies. Finally, the performance of these models was evaluated using the F1 score, 95\% confidence interval (CI) and paired-sample t-tests on our constructed dataset, with the prediction results further assessed by radiologists. Using zero-shot prompting, the fine-tuned Llama-3-70B-Instruct model achieved the best performance with the following F1 scores: 0.769 for negation errors, 0.772 for left/right errors, 0.750 for interval change errors, 0.828 for transcription errors, and 0.780 overall. In the real-world evaluation phase, two radiologists reviewed 200 randomly selected reports output by the model. Of these, 99 were confirmed to contain errors detected by the models by both radiologists, and 163 were confirmed to contain model-detected errors by at least one radiologist. Generative LLMs, fine-tuned on synthetic and MIMIC-CXR radiology reports, greatly enhanced error detection in radiology reports.
Authors: Ayan Sengupta, Siddhant Chaudhary, Tanmoy Chakraborty
Abstract: We introduce compression laws for language language models (LLMs). While recent scaling laws have sought to understand how LLMs scale with respect to model size, pre-training data, and computational resources, we focus on understanding how model compression affects the performance of a pre-trained LLM on downstream tasks. We empirically examine the effects of structured model compression on LLMs through over $1000$ experiments across eight models with sizes ranging from $0.5B$ to $14B$ parameters. Our findings indicate that the test cross-entropy loss increases quadratically with the compression ratio, whereas performance on downstream tasks declines only linearly. Our study emphasizes the importance of recovery fine-tuning in enhancing generation loss, showing that the test loss of compressed LLMs can improve by up to 55% with recovery fine-tuning. At higher compression ratios (up to 90%), compressed LLMs demonstrate a speed increase of 60% during inference compared to their uncompressed counterparts, compensating for the performance degradation at this level. However, for smaller models ($\le 7B$), the computational gains are limited, peaking at just 35%. We conclude that model compression can be highly beneficial for larger models, especially when a smaller model within the same computational budget is not available. These insights provide the practical guidelines for utilizing model compression techniques for adopting LLMs in real-life applications in resource-constrained settings.
Authors: Shenyang Liu, Yang Gao, Shaoyan Zhai, Liqiang Wang
Abstract: Prompt Recovery, reconstructing prompts from the outputs of large language models (LLMs), has grown in importance as LLMs become ubiquitous. Most users access LLMs through APIs without internal model weights, relying only on outputs and logits, which complicates recovery. This paper explores a unique prompt recovery task focused on reconstructing prompts for style transfer and rephrasing, rather than typical question-answering. We introduce a dataset created with LLM assistance, ensuring quality through multiple techniques, and test methods like zero-shot, few-shot, jailbreak, chain-of-thought, fine-tuning, and a novel canonical-prompt fallback for poor-performing cases. Our results show that one-shot and fine-tuning yield the best outcomes but highlight flaws in traditional sentence similarity metrics for evaluating prompt recovery. Contributions include (1) a benchmark dataset, (2) comprehensive experiments on prompt recovery strategies, and (3) identification of limitations in current evaluation metrics, all of which advance general prompt recovery research, where the structure of the input prompt is unrestricted.
Authors: Priyanshu Kumar, Devansh Jain, Akhila Yerukola, Liwei Jiang, Himanshu Beniwal, Thomas Hartvigsen, Maarten Sap
Abstract: Truly multilingual safety moderation efforts for Large Language Models (LLMs) have been hindered by a narrow focus on a small set of languages (e.g., English, Chinese) as well as a limited scope of safety definition, resulting in significant gaps in moderation capabilities. To bridge these gaps, we release POLYGUARD, a new state-of-the-art multilingual safety model for safeguarding LLM generations, and the corresponding training and evaluation datasets. POLYGUARD is trained on POLYGUARDMIX, the largest multilingual safety training corpus to date containing 1.91M samples across 17 languages (e.g., Chinese, Czech, English, Hindi). We also introduce POLYGUARDPROMPTS, a high quality multilingual benchmark with 29K samples for the evaluation of safety guardrails. Created by combining naturally occurring multilingual human-LLM interactions and human-verified machine translations of an English-only safety dataset (WildGuardMix; Han et al., 2024), our datasets contain prompt-output pairs with labels of prompt harmfulness, response harmfulness, and response refusal. Through extensive evaluations across multiple safety and toxicity benchmarks, we demonstrate that POLYGUARD outperforms existing state-of-the-art open-weight and commercial safety classifiers by 5.5%. Our contributions advance efforts toward safer multilingual LLMs for all global users.
Authors: Xiaokai Wang, Guiran Liu, Binrong Zhu, Jacky He, Hongye Zheng, Hanlu Zhang
Abstract: This study proposes a medical entity extraction method based on Transformer to enhance the information extraction capability of medical literature. Considering the professionalism and complexity of medical texts, we compare the performance of different pre-trained language models (BERT, BioBERT, PubMedBERT, ClinicalBERT) in medical entity extraction tasks. Experimental results show that PubMedBERT achieves the best performance (F1-score = 88.8%), indicating that a language model pre-trained on biomedical literature is more effective in the medical domain. In addition, we analyze the impact of different entity extraction methods (CRF, Span-based, Seq2Seq) and find that the Span-based approach performs best in medical entity extraction tasks (F1-score = 88.6%). It demonstrates superior accuracy in identifying entity boundaries. In low-resource scenarios, we further explore the application of Few-shot Learning in medical entity extraction. Experimental results show that even with only 10-shot training samples, the model achieves an F1-score of 79.1%, verifying the effectiveness of Few-shot Learning under limited data conditions. This study confirms that the combination of pre-trained language models and Few-shot Learning can enhance the accuracy of medical entity extraction. Future research can integrate knowledge graphs and active learning strategies to improve the model's generalization and stability, providing a more effective solution for medical NLP research. Keywords- Natural Language Processing, medical named entity recognition, pre-trained language model, Few-shot Learning, information extraction, deep learning
Authors: Daniel Bershatsky, Ivan Oseledets
Abstract: A common assumption is that MoE routers primarily leverage semantic features for expert selection. However, our study challenges this notion by demonstrating that positional token information also plays a crucial role in routing decisions. Through extensive empirical analysis, we provide evidence supporting this hypothesis, develop a phenomenological explanation of the observed behavior, and discuss practical implications for MoE-based architectures.
Authors: David Herrera-Poyatos, Carlos Pel\'aez-Gonz\'alez, Cristina Zuheros, Andr\'es Herrera-Poyatos, Virilo Tejedor, Francisco Herrera, Rosana Montes
Abstract: Large Language Models (LLMs) have significantly advanced sentiment analysis, yet their inherent uncertainty and variability pose critical challenges to achieving reliable and consistent outcomes. This paper systematically explores the Model Variability Problem (MVP) in LLM-based sentiment analysis, characterized by inconsistent sentiment classification, polarization, and uncertainty arising from stochastic inference mechanisms, prompt sensitivity, and biases in training data. We analyze the core causes of MVP, presenting illustrative examples and a case study to highlight its impact. In addition, we investigate key challenges and mitigation strategies, paying particular attention to the role of temperature as a driver of output randomness and emphasizing the crucial role of explainability in improving transparency and user trust. By providing a structured perspective on stability, reproducibility, and trustworthiness, this study helps develop more reliable, explainable, and robust sentiment analysis models, facilitating their deployment in high-stakes domains such as finance, healthcare, and policymaking, among others.
Authors: Archana Sahu, Plaban Kumar Bhowmick
Abstract: In this paper, we have presented a method for identifying missing items known as gaps in the student answers by comparing them against the corresponding model answer/reference answers, automatically. The gaps can be identified at word, phrase or sentence level. The identified gaps are useful in providing feedback to the students for formative assessment. The problem of gap identification has been modelled as an alignment of a pair of directed graphs representing a student answer and the corresponding model answer for a given question. To validate the proposed approach, the gap annotated student answers considering answers from three widely known datasets in the short answer grading domain, namely, University of North Texas (UNT), SciEntsBank, and Beetle have been developed and this gap annotated student answers' dataset is available at: https://github.com/sahuarchana7/gaps-answers-dataset. Evaluation metrics used in the traditional machine learning tasks have been adopted to evaluate the task of gap identification. Though performance of the proposed approach varies across the datasets and the types of the answers, overall the performance is observed to be promising.
Authors: Yao Tao, Yehui Tang, Yun Wang, Mingjian Zhu, Hailin Hu, Yunhe Wang
Abstract: Despite the recent success of large language models (LLMs), LLMs are particularly challenging in long-sequence inference scenarios due to the quadratic computational complexity of the attention mechanism. Inspired by the interpretability theory of feature attribution in neural network models, we observe that not all tokens have the same contribution. Based on this observation, we propose a novel token pruning framework, namely Saliency-driven Dynamic Token Pruning (SDTP), to gradually and dynamically prune redundant tokens based on the input context. Specifically, a lightweight saliency-driven prediction module is designed to estimate the importance score of each token with its hidden state, which is added to different layers of the LLM to hierarchically prune redundant tokens. Furthermore, a ranking-based optimization strategy is proposed to minimize the ranking divergence of the saliency score and the predicted importance score. Extensive experiments have shown that our framework is generalizable to various models and datasets. By hierarchically pruning 65\% of the input tokens, our method greatly reduces 33\% $\sim$ 47\% FLOPs and achieves speedup up to 1.75$\times$ during inference, while maintaining comparable performance. We further demonstrate that SDTP can be combined with KV cache compression method for further compression.
Authors: Anantharaman Janakiraman, Behnaz Ghoraani
Abstract: Text summarization is crucial for mitigating information overload across domains like journalism, medicine, and business. This research evaluates summarization performance across 17 large language models (OpenAI, Google, Anthropic, open-source) using a novel multi-dimensional framework. We assessed models on seven diverse datasets (BigPatent, BillSum, CNN/DailyMail, PubMed, SAMSum, WikiHow, XSum) at three output lengths (50, 100, 150 tokens) using metrics for factual consistency, semantic similarity, lexical overlap, and human-like quality, while also considering efficiency factors. Our findings reveal significant performance differences, with specific models excelling in factual accuracy (deepseek-v3), human-like quality (claude-3-5-sonnet), and processing efficiency/cost-effectiveness (gemini-1.5-flash, gemini-2.0-flash). Performance varies dramatically by dataset, with models struggling on technical domains but performing well on conversational content. We identified a critical tension between factual consistency (best at 50 tokens) and perceived quality (best at 150 tokens). Our analysis provides evidence-based recommendations for different use cases, from high-stakes applications requiring factual accuracy to resource-constrained environments needing efficient processing. This comprehensive approach enhances evaluation methodology by integrating quality metrics with operational considerations, incorporating trade-offs between accuracy, efficiency, and cost-effectiveness to guide model selection for specific applications.
Authors: Chitranshu Harbola, Anupam Purwar
Abstract: In the evolving landscape of conversational AI, generating concise, context-aware, and human-like dialogue using small and medium-sized language models (LLMs) remains a complex challenge. This study investigates the influence of LoRA rank, dataset scale, and prompt prefix design on both knowledge retention and stylistic alignment. While fine-tuning improves fluency and enables stylistic customization, its ability to integrate unseen knowledge is constrained -- particularly with smaller datasets. Conversely, RAG-augmented models, equipped to incorporate external documents at inference, demonstrated superior factual accuracy on out-of-distribution prompts, though they lacked the stylistic consistency achieved by fine-tuning. Evaluations by LLM-based judges across knowledge accuracy, conversational quality, and conciseness suggest that fine-tuning is best suited for tone adaptation, whereas RAG excels at real-time knowledge augmentation.
Authors: Qi Zhang, Huitong Pan, Zhijia Chen, Longin Jan Latecki, Cornelia Caragea, Eduard Dragut
Abstract: Distantly Supervised Named Entity Recognition (DS-NER) has attracted attention due to its scalability and ability to automatically generate labeled data. However, distant annotation introduces many mislabeled instances, limiting its performance. Most of the existing work attempt to solve this problem by developing intricate models to learn from the noisy labels. An alternative approach is to attempt to clean the labeled data, thus increasing the quality of distant labels. This approach has received little attention for NER. In this paper, we propose a training dynamics-based label cleaning approach, which leverages the behavior of a model as training progresses to characterize the distantly annotated samples. We also introduce an automatic threshold estimation strategy to locate the errors in distant labels. Extensive experimental results demonstrate that: (1) models trained on our cleaned DS-NER datasets, which were refined by directly removing identified erroneous annotations, achieve significant improvements in F1-score, ranging from 3.18% to 8.95%; and (2) our method outperforms numerous advanced DS-NER approaches across four datasets.
Authors: Patrick Queiroz Da Silva, Hari Sethuraman, Dheeraj Rajagopal, Hannaneh Hajishirzi, Sachin Kumar
Abstract: Steering methods for language models (LMs) have gained traction as lightweight alternatives to fine-tuning, enabling targeted modifications to model activations. However, prior studies primarily report results on a few models, leaving critical gaps in understanding the robustness of these methods. In this work, we systematically examine three prominent steering methods -- DoLa, function vectors, and task vectors. In contrast to the original studies, which evaluated a handful of models, we test up to 36 models belonging to 14 families with sizes ranging from 1.5B to 70B parameters. Our experiments reveal substantial variability in the effectiveness of the steering approaches, with a large number of models showing no improvement and at times degradation in steering performance. Our analysis demonstrate fundamental flaws in the assumptions underlying these methods, challenging their reliability as scalable steering solutions.
Authors: Eylon Caplan, Tania Chakraborty, Dan Goldwasser
Abstract: Understanding how people of various demographics think, feel, and express themselves (collectively called group expression) is essential for social science and underlies the assessment of bias in Large Language Models (LLMs). While LLMs can effectively summarize group expression when provided with empirical examples, coming up with generalizable theories of how a group's expression manifests in real-world text is challenging. In this paper, we define a new task called Group Theorization, in which a system must write theories that differentiate expression across demographic groups. We make available a large dataset on this task, Splits!, constructed by splitting Reddit posts by neutral topics (e.g. sports, cooking, and movies) and by demographics (e.g. occupation, religion, and race). Finally, we suggest a simple evaluation framework for assessing how effectively a method can generate 'better' theories about group expression, backed by human validation. We publicly release the raw corpora and evaluation scripts for Splits! to help researchers assess how methods infer--and potentially misrepresent--group differences in expression. We make Splits! and our evaluation module available at https://github.com/eyloncaplan/splits.
Authors: Yuren Mao, Yu Mi, Peigen Liu, Mengfei Zhang, Hanqing Liu, Yunjun Gao
Abstract: Cell type annotation is critical for understanding cellular heterogeneity. Based on single-cell RNA-seq data and deep learning models, good progress has been made in annotating a fixed number of cell types within a specific tissue. However, universal cell annotation, which can generalize across tissues, discover novel cell types, and extend to novel cell types, remains less explored. To fill this gap, this paper proposes scAgent, a universal cell annotation framework based on Large Language Models (LLMs). scAgent can identify cell types and discover novel cell types in diverse tissues; furthermore, it is data efficient to learn novel cell types. Experimental studies in 160 cell types and 35 tissues demonstrate the superior performance of scAgent in general cell-type annotation, novel cell discovery, and extensibility to novel cell type.
Authors: Hyunseo Shin, Wonseok Hwang
Abstract: Recent advancements in large language models (LLMs) have significantly enhanced the performance of conversational AI systems. To extend their capabilities to knowledge-intensive domains such as biomedical and legal fields, where the accuracy is critical, LLMs are often combined with information retrieval (IR) systems to generate responses based on retrieved documents. However, for IR systems to effectively support such applications, they must go beyond simple semantic matching and accurately capture diverse query intents, including causal relationships. Existing IR models primarily focus on retrieving documents based on surface-level semantic similarity, overlooking deeper relational structures such as causality. To address this, we propose CAWAI, a retrieval model that is trained with dual objectives: semantic and causal relations. Our extensive experiments demonstrate that CAWAI outperforms various models on diverse causal retrieval tasks especially under large-scale retrieval settings. We also show that CAWAI exhibits strong zero-shot generalization across scientific domain QA tasks.
Authors: Yifei Yu, Qian-Wen Zhang, Lingfeng Qiao, Di Yin, Fang Li, Jie Wang, Zengxi Chen, Suncong Zheng, Xiaolong Liang, Xing Sun
Abstract: Evaluating the ability of large language models (LLMs) to handle extended contexts is critical, particularly for retrieving information relevant to specific queries embedded within lengthy inputs. We introduce Sequential-NIAH, a benchmark specifically designed to evaluate the capability of LLMs to extract sequential information items (known as needles) from long contexts. The benchmark comprises three types of needle generation pipelines: synthetic, real, and open-domain QA. It includes contexts ranging from 8K to 128K tokens in length, with a dataset of 14,000 samples (2,000 reserved for testing). To facilitate evaluation on this benchmark, we trained a synthetic data-driven evaluation model capable of evaluating answer correctness based on chronological or logical order, achieving an accuracy of 99.49% on synthetic test data. We conducted experiments on six well-known LLMs, revealing that even the best-performing model achieved a maximum accuracy of only 63.15%. Further analysis highlights the growing challenges posed by increasing context lengths and the number of needles, underscoring substantial room for improvement. Additionally, noise robustness experiments validate the reliability of the benchmark, making Sequential-NIAH an important reference for advancing research on long text extraction capabilities of LLMs.
Authors: Will Cai, Tianneng Shi, Xuandong Zhao, Dawn Song
Abstract: The proliferation of Large Language Models (LLMs) accessed via black-box APIs introduces a significant trust challenge: users pay for services based on advertised model capabilities (e.g., size, performance), but providers may covertly substitute the specified model with a cheaper, lower-quality alternative to reduce operational costs. This lack of transparency undermines fairness, erodes trust, and complicates reliable benchmarking. Detecting such substitutions is difficult due to the black-box nature, typically limiting interaction to input-output queries. This paper formalizes the problem of model substitution detection in LLM APIs. We systematically evaluate existing verification techniques, including output-based statistical tests, benchmark evaluations, and log probability analysis, under various realistic attack scenarios like model quantization, randomized substitution, and benchmark evasion. Our findings reveal the limitations of methods relying solely on text outputs, especially against subtle or adaptive attacks. While log probability analysis offers stronger guarantees when available, its accessibility is often limited. We conclude by discussing the potential of hardware-based solutions like Trusted Execution Environments (TEEs) as a pathway towards provable model integrity, highlighting the trade-offs between security, performance, and provider adoption. Code is available at https://github.com/sunblaze-ucb/llm-api-audit
Authors: Yubo Li, Xiaobin Shen, Xinyu Yao, Xueying Ding, Yidi Miao, Ramayya Krishnan, Rema Padman
Abstract: Recent advancements in large language models (LLMs) have revolutionized their ability to handle single-turn tasks, yet real-world applications demand sophisticated multi-turn interactions. This survey provides a comprehensive review of recent advancements in evaluating and enhancing multi-turn interactions in LLMs. Focusing on task-specific scenarios, from instruction following in diverse domains such as math and coding to complex conversational engagements in roleplay, healthcare, education, and even adversarial jailbreak settings, we systematically examine the challenges of maintaining context, coherence, fairness, and responsiveness over prolonged dialogues. The paper organizes current benchmarks and datasets into coherent categories that reflect the evolving landscape of multi-turn dialogue evaluation. In addition, we review a range of enhancement methodologies under multi-turn settings, including model-centric strategies (contextual learning, supervised fine-tuning, reinforcement learning, and new architectures), external integration approaches (memory-augmented, retrieval-based methods, and knowledge graph), and agent-based techniques for collaborative interactions. Finally, we discuss open challenges and propose future directions for research to further advance the robustness and effectiveness of multi-turn interactions in LLMs. Related resources and papers are available at https://github.com/yubol-cmu/Awesome-Multi-Turn-LLMs.
Authors: Minki Kang, Jongwon Jeong, Jaewoong Cho
Abstract: Recent studies have demonstrated that test-time compute scaling effectively improves the performance of small language models (sLMs). However, prior research has mainly examined test-time compute scaling with an additional larger model as a verifier, leaving self-verification by sLMs underexplored. In this work, we investigate whether sLMs can reliably self-verify their outputs under test-time scaling. We find that even with knowledge distillation from larger verifiers, sLMs struggle with verification tasks requiring memorization, such as numerical calculations and fact-checking. To address this limitation, we propose Tool-integrated self-verification (T1), which delegates memorization-heavy verification steps to external tools, such as a code interpreter. Our theoretical analysis shows that tool integration reduces memorization demands and improves test-time scaling performance. Experiments on the MATH benchmark demonstrate that, with T1, a Llama-3.2 1B model under test-time scaling outperforms the significantly larger Llama-3.1 8B model. Moreover, T1 generalizes effectively to both mathematical (MATH500) and multi-domain knowledge-intensive tasks (MMLU-Pro). Our findings highlight the potential of tool integration to substantially improve the self-verification abilities of sLMs.
Authors: Shubham Kumar Nigam, Balaramamahanthi Deepak Patnaik, Shivam Mishra, Noel Shallum, Kripabandhu Ghosh, Arnab Bhattacharya
Abstract: In the landscape of Fact-based Judgment Prediction and Explanation (FJPE), reliance on factual data is essential for developing robust and realistic AI-driven decision-making tools. This paper introduces TathyaNyaya, the largest annotated dataset for FJPE tailored to the Indian legal context, encompassing judgments from the Supreme Court of India and various High Courts. Derived from the Hindi terms "Tathya" (fact) and "Nyaya" (justice), the TathyaNyaya dataset is uniquely designed to focus on factual statements rather than complete legal texts, reflecting real-world judicial processes where factual data drives outcomes. Complementing this dataset, we present FactLegalLlama, an instruction-tuned variant of the LLaMa-3-8B Large Language Model (LLM), optimized for generating high-quality explanations in FJPE tasks. Finetuned on the factual data in TathyaNyaya, FactLegalLlama integrates predictive accuracy with coherent, contextually relevant explanations, addressing the critical need for transparency and interpretability in AI-assisted legal systems. Our methodology combines transformers for binary judgment prediction with FactLegalLlama for explanation generation, creating a robust framework for advancing FJPE in the Indian legal domain. TathyaNyaya not only surpasses existing datasets in scale and diversity but also establishes a benchmark for building explainable AI systems in legal analysis. The findings underscore the importance of factual precision and domain-specific tuning in enhancing predictive performance and interpretability, positioning TathyaNyaya and FactLegalLlama as foundational resources for AI-assisted legal decision-making.
Authors: Ankush Raut, Xiaofeng Zhu, Maria Leonor Pacheco
Abstract: This paper evaluates the ability of Large Language Models (LLMs) to leverage contextual information in the form of structured linguistic representations. Specifically, we examine the impact of encoding both short and long contexts using Abstract Meaning Representation (AMR) structures across a diverse set of language tasks. We perform our analysis using 8-bit quantized and instruction-tuned versions of Llama 3.1 (8B), Phi-3, and Mistral 7B. Our results indicate that, for tasks involving short contexts, augmenting the prompt with the AMR of the original language context often degrades the performance of the underlying LLM. However, for tasks that involve long contexts, such as dialogue summarization in the SAMSum dataset, this enhancement improves LLM performance, for example, by increasing the zero-shot cosine similarity score of Llama 3.1 from 66.2% to 76%. This improvement is more evident in the newer and larger LLMs, but does not extend to the older or smaller ones. In addition, we observe that LLMs can effectively reconstruct the original text from a linearized AMR, achieving a cosine similarity of 81.3% in the best-case scenario.
Authors: Leonardo Ranaldi, Federico Ranaldi, Fabio Massimo Zanzotto, Barry Haddow, Alexandra Birch
Abstract: Retrieval-augmented generation (RAG) is key to enhancing large language models (LLMs) to systematically access richer factual knowledge. Yet, using RAG brings intrinsic challenges, as LLMs must deal with potentially conflicting knowledge, especially in multilingual retrieval, where the heterogeneity of knowledge retrieved may deliver different outlooks. To make RAG more analytical, critical and grounded, we introduce Dialectic-RAG (DRAG), a modular approach guided by Argumentative Explanations, i.e., structured reasoning process that systematically evaluates retrieved information by comparing, contrasting, and resolving conflicting perspectives. Given a query and a set of multilingual related documents, DRAG selects and exemplifies relevant knowledge for delivering dialectic explanations that, by critically weighing opposing arguments and filtering extraneous content, clearly determine the final response. Through a series of in-depth experiments, we show the impact of our framework both as an in-context learning strategy and for constructing demonstrations to instruct smaller models. The final results demonstrate that DRAG significantly improves RAG approaches, requiring low-impact computational effort and providing robustness to knowledge perturbations.
Authors: Mia Jacobsen, Ross Deans Kristensen-McLachlan
Abstract: We consider the textual profiles of different fanfiction maturity ratings, how they vary across fan groups, and how this relates to reader engagement metrics. Previous studies have shown that fanfiction writing is motivated by a combination of admiration for and frustration with the fan object. These findings emerge when looking at fanfiction as a whole, as well as when it is divided into subgroups, also called fandoms. However, maturity ratings are used to indicate the intended audience of the fanfiction, as well as whether the story includes mature themes and explicit scenes. Since these ratings can be used to filter readers and writers, they can also be seen as a proxy for different reader/writer motivations and desires. We find that explicit fanfiction in particular has a distinct textual profile when compared to other maturity ratings. These findings thus nuance our understanding of reader/writer motivations in fanfiction communities, and also highlights the influence of the community norms and fan behavior more generally on these cultural products.
Authors: Ruikang Liu, Yuxuan Sun, Manyi Zhang, Haoli Bai, Xianzhi Yu, Tiezheng Yu, Chun Yuan, Lu Hou
Abstract: Recent advancements in reasoning language models have demonstrated remarkable performance in complex tasks, but their extended chain-of-thought reasoning process increases inference overhead. While quantization has been widely adopted to reduce the inference cost of large language models, its impact on reasoning models remains understudied. In this study, we conduct the first systematic study on quantized reasoning models, evaluating the open-sourced DeepSeek-R1-Distilled Qwen and LLaMA families ranging from 1.5B to 70B parameters, and QwQ-32B. Our investigation covers weight, KV cache, and activation quantization using state-of-the-art algorithms at varying bit-widths, with extensive evaluation across mathematical (AIME, MATH-500), scientific (GPQA), and programming (LiveCodeBench) reasoning benchmarks. Our findings reveal that while lossless quantization can be achieved with W8A8 or W4A16 quantization, lower bit-widths introduce significant accuracy risks. We further identify model size, model origin, and task difficulty as critical determinants of performance. Contrary to expectations, quantized models do not exhibit increased output lengths. In addition, strategically scaling the model sizes or reasoning steps can effectively enhance the performance. All quantized models and codes will be open-sourced in https://github.com/ruikangliu/Quantized-Reasoning-Models.
URLs: https://github.com/ruikangliu/Quantized-Reasoning-Models.
Authors: Sam Kirkham
Abstract: A fundamental challenge in the cognitive sciences is discovering the dynamics that govern behaviour. Take the example of spoken language, which is characterised by a highly variable and complex set of physical movements that map onto the small set of cognitive units that comprise language. What are the fundamental dynamical principles behind the movements that structure speech production? In this study, we discover models in the form of symbolic equations that govern articulatory gestures during speech. A sparse symbolic regression algorithm is used to discover models from kinematic data on the tongue and lips. We explore these candidate models using analytical techniques and numerical simulations, and find that a second-order linear model achieves high levels of accuracy, but a nonlinear force is required to properly model articulatory dynamics in approximately one third of cases. This supports the proposal that an autonomous, nonlinear, second-order differential equation is a viable dynamical law for articulatory gestures in speech. We conclude by identifying future opportunities and obstacles in data-driven model discovery and outline prospects for discovering the dynamical principles that govern language, brain and behaviour.
Authors: Hongtao Wang, Renchi Yang, Hewen Wang, Haoran Zheng, Jianliang Xu
Abstract: Textual interaction networks (TINs) are an omnipresent data structure used to model the interplay between users and items on e-commerce websites, social networks, etc., where each interaction is associated with a text description. Classifying such textual interactions (TIC) finds extensive use in detecting spam reviews in e-commerce, fraudulent transactions in finance, and so on. Existing TIC solutions either (i) fail to capture the rich text semantics due to the use of context-free text embeddings, and/or (ii) disregard the bipartite structure and node heterogeneity of TINs, leading to compromised TIC performance. In this work, we propose SAFT, a new architecture that integrates language- and graph-based modules for the effective fusion of textual and structural semantics in the representation learning of interactions. In particular, line graph attention (LGA)/gated attention units (GAUs) and pretrained language models (PLMs) are capitalized on to model the interaction-level and token-level signals, which are further coupled via the proxy token in an iterative and contextualized fashion. Additionally, an efficient and theoretically-grounded approach is developed to encode the local and global topology information pertaining to interactions into structural embeddings. The resulting embeddings not only inject the structural features underlying TINs into the textual interaction encoding but also facilitate the design of graph sampling strategies. Extensive empirical evaluations on multiple real TIN datasets demonstrate the superiority of SAFT over the state-of-the-art baselines in TIC accuracy.
Authors: Longdi Xian, Jianzhang Ni, Mingzhu Wang
Abstract: Depression is a prevalent mental health disorder that is difficult to detect early due to subjective symptom assessments. Recent advancements in large language models have offered efficient and cost-effective approaches for this objective. In this study, we evaluated the performance of four LLMs in depression detection using clinical interview data. We selected the best performing model and further tested it in the severity evaluation scenario and knowledge enhanced scenario. The robustness was evaluated in complex diagnostic scenarios using a dataset comprising 51074 statements from six different mental disorders. We found that DeepSeek V3 is the most reliable and cost-effective model for depression detection, performing well in both zero-shot and few-shot scenarios, with zero-shot being the most efficient choice. The evaluation of severity showed low agreement with the human evaluator, particularly for mild depression. The model maintains stably high AUCs for detecting depression in complex diagnostic scenarios. These findings highlight DeepSeek V3s strong potential for text-based depression detection in real-world clinical applications. However, they also underscore the need for further refinement in severity assessment and the mitigation of potential biases to enhance clinical reliability.
Authors: Ran Xu, Wenqi Shi, Yuchen Zhuang, Yue Yu, Joyce C. Ho, Haoyu Wang, Carl Yang
Abstract: Retrieval-Augmented Generation (RAG) systems often struggle to handle multi-hop question-answering tasks accurately due to irrelevant context retrieval and limited complex reasoning capabilities. We introduce Collab-RAG, a collaborative training framework that leverages mutual enhancement between a white-box small language model (SLM) and a blackbox large language model (LLM) for RAG. Specifically, the SLM decomposes complex queries into simpler sub-questions, thus enhancing the accuracy of the retrieval and facilitating more effective reasoning by the black-box LLM. Concurrently, the black-box LLM provides feedback signals to improve the SLM's decomposition capability. We observe that Collab-RAG relies solely on supervision from an affordable black-box LLM without additional distillation from frontier LLMs, yet demonstrates strong generalization across multiple black-box LLMs. Experimental evaluations across five multi-hop QA datasets demonstrate that Collab-RAG substantially outperforms existing black-box-only and SLM fine-tuning baselines by 1.8%-14.2% on average. In particular, our fine-tuned 3B SLM surpasses a frozen 32B LLM in question decomposition, highlighting the efficiency of Collab-RAG in improving reasoning and retrieval for complex questions. The code of Collab-RAG is available on https://github.com/ritaranx/Collab-RAG/.
Authors: Jos\'e Pombal, Dongkeun Yoon, Patrick Fernandes, Ian Wu, Seungone Kim, Ricardo Rei, Graham Neubig, Andr\'e F. T. Martins
Abstract: The use of language models for automatically evaluating long-form text (LLM-as-a-judge) is becoming increasingly common, yet most LLM judges are optimized exclusively for English, with strategies for enhancing their multilingual evaluation capabilities remaining largely unexplored in the current literature. This has created a disparity in the quality of automatic evaluation methods for non-English languages, ultimately hindering the development of models with better multilingual capabilities. To bridge this gap, we introduce M-Prometheus, a suite of open-weight LLM judges ranging from 3B to 14B parameters that can provide both direct assessment and pairwise comparison feedback on multilingual outputs. M-Prometheus models outperform state-of-the-art open LLM judges on multilingual reward benchmarks spanning more than 20 languages, as well as on literary machine translation (MT) evaluation covering 4 language pairs. Furthermore, M-Prometheus models can be leveraged at decoding time to significantly improve generated outputs across all 3 tested languages, showcasing their utility for the development of better multilingual models. Lastly, through extensive ablations, we identify the key factors for obtaining an effective multilingual judge, including backbone model selection and training on natively multilingual feedback data instead of translated data. We release our models, training dataset, and code.
Authors: Yuzhe Zhang, Min Cen, Hong Zhang
Abstract: Distantly supervised named entity recognition (DS-NER) has been proposed to exploit the automatically labeled training data by external knowledge bases instead of human annotations. However, it tends to suffer from a high false negative rate due to the inherent incompleteness. To address this issue, we present a novel approach called \textbf{C}onstraint \textbf{M}ulti-class \textbf{P}ositive and \textbf{U}nlabeled Learning (CMPU), which introduces a constraint factor on the risk estimator of multiple positive classes. It suggests that the constraint non-negative risk estimator is more robust against overfitting than previous PU learning methods with limited positive data. Solid theoretical analysis on CMPU is provided to prove the validity of our approach. Extensive experiments on two benchmark datasets that were labeled using diverse external knowledge sources serve to demonstrate the superior performance of CMPU in comparison to existing DS-NER methods.
Authors: Shion Fukuhata, Yoshinobu Kano
Abstract: When fine-tuning BERT models for specific tasks, it is common to select part of the final layer's output and input it into a newly created fully connected layer. However, it remains unclear which part of the final layer should be selected and what information each dimension of the layers holds. In this study, we comprehensively investigated the effectiveness and redundancy of token vectors, layers, and dimensions through BERT fine-tuning on GLUE tasks. The results showed that outputs other than the CLS vector in the final layer contain equivalent information, most tasks require only 2-3 dimensions, and while the contribution of lower layers decreases, there is little difference among higher layers. We also evaluated the impact of freezing pre-trained layers and conducted cross-fine-tuning, where fine-tuning is applied sequentially to different tasks. The findings suggest that hidden layers may change significantly during fine-tuning, BERT has considerable redundancy, enabling it to handle multiple tasks simultaneously, and its number of dimensions may be excessive.
Authors: Carlos Pel\'aez-Gonz\'alez, Andr\'es Herrera-Poyatos, Cristina Zuheros, David Herrera-Poyatos, Virilo Tejedor, Francisco Herrera
Abstract: The study of large language models (LLMs) is a key area in open-world machine learning. Although LLMs demonstrate remarkable natural language processing capabilities, they also face several challenges, including consistency issues, hallucinations, and jailbreak vulnerabilities. Jailbreaking refers to the crafting of prompts that bypass alignment safeguards, leading to unsafe outputs that compromise the integrity of LLMs. This work specifically focuses on the challenge of jailbreak vulnerabilities and introduces a novel taxonomy of jailbreak attacks grounded in the training domains of LLMs. It characterizes alignment failures through generalization, objectives, and robustness gaps. Our primary contribution is a perspective on jailbreak, framed through the different linguistic domains that emerge during LLM training and alignment. This viewpoint highlights the limitations of existing approaches and enables us to classify jailbreak attacks on the basis of the underlying model deficiencies they exploit. Unlike conventional classifications that categorize attacks based on prompt construction methods (e.g., prompt templating), our approach provides a deeper understanding of LLM behavior. We introduce a taxonomy with four categories -- mismatched generalization, competing objectives, adversarial robustness, and mixed attacks -- offering insights into the fundamental nature of jailbreak vulnerabilities. Finally, we present key lessons derived from this taxonomic study.
Authors: Ling Hu, Yuemei Xu, Xiaoyang Gu, Letao Han
Abstract: Despite the impressive performance of large language models (LLMs), they can present unintended biases and harmful behaviors driven by encoded values, emphasizing the urgent need to understand the value mechanisms behind them. However, current research primarily evaluates these values through external responses with a focus on AI safety, lacking interpretability and failing to assess social values in real-world contexts. In this paper, we propose a novel framework called ValueExploration, which aims to explore the behavior-driven mechanisms of National Social Values within LLMs at the neuron level. As a case study, we focus on Chinese Social Values and first construct C-voice, a large-scale bilingual benchmark for identifying and evaluating Chinese Social Values in LLMs. By leveraging C-voice, we then identify and locate the neurons responsible for encoding these values according to activation difference. Finally, by deactivating these neurons, we analyze shifts in model behavior, uncovering the internal mechanism by which values influence LLM decision-making. Extensive experiments on four representative LLMs validate the efficacy of our framework. The benchmark and code will be available.
Authors: Anastasiia Ivanova, Natalia Fedorova, Sergey Tilga, Ekaterina Artemova
Abstract: The rapid development of AI-driven tools, particularly large language models (LLMs), is reshaping professional writing. Still, key aspects of their adoption such as languages support, ethics, and long-term impact on writers voice and creativity remain underexplored. In this work, we conducted a questionnaire (N = 301) and an interactive survey (N = 36) targeting professional writers regularly using AI. We examined LLM-assisted writing practices across 25+ languages, ethical concerns, and user expectations. The findings of the survey demonstrate important insights, reflecting upon the importance of: LLMs adoption for non-English speakers; the degree of misinformation, domain and style adaptation; usability and key features of LLMs. These insights can guide further development, benefiting both writers and a broader user base.
Authors: Charco Hui, Yalu Wen
Abstract: Natural language processing models often face challenges due to limited labeled data, especially in domain specific areas, e.g., clinical trials. To overcome this, text augmentation techniques are commonly used to increases sample size by transforming the original input data into artificial ones with the label preserved. However, traditional text classification methods ignores the relationship between augmented texts and treats them as independent samples which may introduce classification error. Therefore, we propose a novel approach called 'Batch Aggregation' (BAGG) which explicitly models the dependence of text inputs generated through augmentation by incorporating an additional layer that aggregates results from correlated texts. Through studying multiple benchmark data sets across different domains, we found that BAGG can improve classification accuracy. We also found that the increase of performance with BAGG is more obvious in domain specific data sets, with accuracy improvements of up to 10-29%. Through the analysis of benchmark data, the proposed method addresses limitations of traditional techniques and improves robustness in text classification tasks. Our result demonstrates that BAGG offers more robust results and outperforms traditional approaches when training data is limited.
Authors: Jiawei Lian, Jianhong Pan, Lefan Wang, Yi Wang, Shaohui Mei, Lap-Pui Chau
Abstract: Large language models (LLMs) are foundational explorations to artificial general intelligence, yet their alignment with human values via instruction tuning and preference learning achieves only superficial compliance. Here, we demonstrate that harmful knowledge embedded during pretraining persists as indelible "dark patterns" in LLMs' parametric memory, evading alignment safeguards and resurfacing under adversarial inducement at distributional shifts. In this study, we first theoretically analyze the intrinsic ethical vulnerability of aligned LLMs by proving that current alignment methods yield only local "safety regions" in the knowledge manifold. In contrast, pretrained knowledge remains globally connected to harmful concepts via high-likelihood adversarial trajectories. Building on this theoretical insight, we empirically validate our findings by employing semantic coherence inducement under distributional shifts--a method that systematically bypasses alignment constraints through optimized adversarial prompts. This combined theoretical and empirical approach achieves a 100% attack success rate across 19 out of 23 state-of-the-art aligned LLMs, including DeepSeek-R1 and LLaMA-3, revealing their universal vulnerabilities.
Authors: Aravind Krishnan, Siva Reddy, Marius Mosbach
Abstract: Machine unlearning is concerned with the task of removing knowledge learned from particular data points from a trained model. In the context of large language models (LLMs), unlearning has recently received increased attention, particularly for removing knowledge about named entities from models for privacy purposes. While various approaches have been proposed to address the unlearning problem, most existing approaches treat all data points to be unlearned equally, i.e., unlearning that Montreal is a city in Canada is treated exactly the same as unlearning the phone number of the first author of this paper. In this work, we show that this all data is equal assumption does not hold for LLM unlearning. We study how the success of unlearning depends on the frequency of the knowledge we want to unlearn in the pre-training data of a model and find that frequency strongly affects unlearning, i.e., more frequent knowledge is harder to unlearn. Additionally, we uncover a misalignment between probability and generation-based evaluations of unlearning and show that this problem worsens as models become larger. Overall, our experiments highlight the need for better evaluation practices and novel methods for LLM unlearning that take the training data of models into account.
Authors: Venkat Srinivasan, Vishaal Jatav, Anushka Chandrababu, Geetika Sharma
Abstract: Large language models (LLMs) have shown significant abilities in retrieving medical knowledge, reasoning over it and answering medical questions comparably to physicians. However, these models are not interpretable, hallucinate, are difficult to maintain and require enormous compute resources for training and inference. In this paper, we report results from Gyan, an explainable language model based on an alternative architecture, on the PubmedQA data set. The Gyan LLM is a compositional language model and the model is decoupled from knowledge. Gyan is trustable, transparent, does not hallucinate and does not require significant training or compute resources. Gyan is easily transferable across domains. Gyan-4.3 achieves SOTA results on PubmedQA with 87.1% accuracy compared to 82% by MedPrompt based on GPT-4 and 81.8% by Med-PaLM 2 (Google and DeepMind). We will be reporting results for other medical data sets - MedQA, MedMCQA, MMLU - Medicine in the future.
Authors: Tianshi Zheng, Yixiang Chen, Chengxi Li, Chunyang Li, Qing Zong, Haochen Shi, Baixuan Xu, Yangqiu Song, Ginny Y. Wong, Simon See
Abstract: Chain-of-Thought (CoT) prompting has been widely recognized for its ability to enhance reasoning capabilities in large language models (LLMs) through the generation of explicit explanatory rationales. However, our study reveals a surprising contradiction to this prevailing perspective. Through extensive experiments involving 16 state-of-the-art LLMs and nine diverse pattern-based in-context learning (ICL) datasets, we demonstrate that CoT and its reasoning variants consistently underperform direct answering across varying model scales and benchmark complexities. To systematically investigate this unexpected phenomenon, we designed extensive experiments to validate several hypothetical explanations. Our analysis uncovers a fundamental explicit-implicit duality driving CoT's performance in pattern-based ICL: while explicit reasoning falters due to LLMs' struggles to infer underlying patterns from demonstrations, implicit reasoning-disrupted by the increased contextual distance of CoT rationales-often compensates, delivering correct answers despite flawed rationales. This duality explains CoT's relative underperformance, as noise from weak explicit inference undermines the process, even as implicit mechanisms partially salvage outcomes. Notably, even long-CoT reasoning models, which excel in abstract and symbolic reasoning, fail to fully overcome these limitations despite higher computational costs. Our findings challenge existing assumptions regarding the universal efficacy of CoT, yielding novel insights into its limitations and guiding future research toward more nuanced and effective reasoning methodologies for LLMs.
Authors: Liu Xiao, Li Zhiyuan, Lin Yueyu
Abstract: Test-time scaling has emerged as a prominent research direction in machine learning, enabling models to enhance their expressive capabilities during inference.Transformers, renowned for striking a delicate balance between efficiency and expressiveness, have benefited from test-time scaling techniques that leverage an expanding key-value (KV) cache to significantly improve performance.In this paper, we introduce a novel state-based approach to test-time scaling, which we term state tuning, tailored to the RNN-based RWKV-7 model.By exploiting the unique strengths of RWKV-7, our method achieves state-of-the-art performance on the target task without altering the model's pre-trained weights. Our approach centers on three key innovations. First, we develop an observer framework that allows a smaller model to replicate and learn the state dynamics of the RWKV-7 model. Second, we employ a kernel method to dynamically upscale the state size, enhancing the model's capacity to capture intricate patterns. Third, we integrate Decorrelated Backpropagation (DBP) to optimize the upscaled state matrix, thereby improving convergence and expressivity. By tuning only the state matrix, we demonstrate that a smaller model can outperform larger models on the given task. This method preserves the efficiency of the original RWKV-7 architecture while harnessing the power of test-time scaling to deliver superior results. Our findings underscore the potential of state tuning as an effective strategy for advancing model performance in resource-constrained settings. Our code is https://github.com/TorchRWKV/flash-linear-attention.
Authors: Saeid Ario Vaghefi, Aymane Hachcham, Veronica Grasso, Jiska Manicus, Nakiete Msemo, Chiara Colesanti Senni, Markus Leippold
Abstract: Tracking financial investments in climate adaptation is a complex and expertise-intensive task, particularly for Early Warning Systems (EWS), which lack standardized financial reporting across multilateral development banks (MDBs) and funds. To address this challenge, we introduce an LLM-based agentic AI system that integrates contextual retrieval, fine-tuning, and multi-step reasoning to extract relevant financial data, classify investments, and ensure compliance with funding guidelines. Our study focuses on a real-world application: tracking EWS investments in the Climate Risk and Early Warning Systems (CREWS) Fund. We analyze 25 MDB project documents and evaluate multiple AI-driven classification methods, including zero-shot and few-shot learning, fine-tuned transformer-based classifiers, chain-of-thought (CoT) prompting, and an agent-based retrieval-augmented generation (RAG) approach. Our results show that the agent-based RAG approach significantly outperforms other methods, achieving 87\% accuracy, 89\% precision, and 83\% recall. Additionally, we contribute a benchmark dataset and expert-annotated corpus, providing a valuable resource for future research in AI-driven financial tracking and climate finance transparency.
Authors: Xinglin Lyu, Wei Tang, Yuang Li, Xiaofeng Zhao, Ming Zhu, Junhui Li, Yunfei Lu, Min Zhang, Daimeng Wei, Hao Yang, Min Zhang
Abstract: Document-level context is crucial for handling discourse challenges in text-to-text document-level machine translation (MT). Despite the increased discourse challenges introduced by noise from automatic speech recognition (ASR), the integration of document-level context in speech translation (ST) remains insufficiently explored. In this paper, we develop DoCIA, an online framework that enhances ST performance by incorporating document-level context. DoCIA decomposes the ST pipeline into four stages. Document-level context is integrated into the ASR refinement, MT, and MT refinement stages through auxiliary LLM (large language model)-based modules. Furthermore, DoCIA leverages document-level information in a multi-level manner while minimizing computational overhead. Additionally, a simple yet effective determination mechanism is introduced to prevent hallucinations from excessive refinement, ensuring the reliability of the final results. Experimental results show that DoCIA significantly outperforms traditional ST baselines in both sentence and discourse metrics across four LLMs, demonstrating its effectiveness in improving ST performance.
Authors: Geyang Guo, Tarek Naous, Hiromi Wakaki, Yukiko Nishimura, Yuki Mitsufuji, Alan Ritter, Wei Xu
Abstract: Existing language models (LMs) often exhibit a Western-centric bias and struggle to represent diverse cultural knowledge. Previous attempts to address this rely on synthetic data and express cultural knowledge only in English. In this work, we study whether a small amount of human-written, multilingual cultural preference data can improve LMs across various model families and sizes. We first introduce CARE, a multilingual resource of 24.1k responses with human preferences on 2,580 questions about Chinese and Arab cultures, all carefully annotated by native speakers and offering more balanced coverage. Using CARE, we demonstrate that cultural alignment improves existing LMs beyond generic resources without compromising general capabilities. Moreover, we evaluate the cultural awareness of LMs, native speakers, and retrieved web content when queried in different languages. Our experiment reveals regional disparities among LMs, which may also be reflected in the documentation gap: native speakers often take everyday cultural commonsense and social norms for granted, while non-natives are more likely to actively seek out and document them. CARE is publicly available at https://github.com/Guochry/CARE (we plan to add Japanese data in the near future).
Authors: Mehdi Fatemi, Banafsheh Rafiee, Mingjie Tang, Kartik Talamadupula
Abstract: Despite significant advancements in large language models (LLMs), a major drawback of reasoning models is their enormous token usage, which increases computational cost, resource requirements, and response time. In this work, we revisit the core principles of reinforcement learning (RL) and, through mathematical analysis, demonstrate that the tendency to generate lengthy responses arises inherently from RL-based optimization during training. This finding questions the prevailing assumption that longer responses inherently improve reasoning accuracy. Instead, we uncover a natural correlation between conciseness and accuracy that has been largely overlooked. Moreover, we show that introducing a secondary phase of RL post-training, using a small set of problems and limited resources, can significantly reduce a model's chain of thought while maintaining or even enhancing accuracy. Finally, we validate our conclusions through extensive experimental results.
Authors: Richard A. Blythe, Casimir Fisch
Abstract: Establishing a communication system is hard because the intended meaning of a signal is unknown to its receiver when first produced, and the signaller also has no idea how that signal will be interpreted. Most theoretical accounts of the emergence of communication systems rely on feedback to reinforce behaviours that have led to successful communication in the past. However, providing such feedback requires already being able to communicate the meaning that was intended or interpreted. Therefore these accounts cannot explain how communication can be bootstrapped from non-communicative behaviours. Here we present a model that shows how a communication system, capable of expressing an unbounded number of meanings, can emerge as a result of individual behavioural differences in a large population without any pre-existing means to determine communicative success. The two key cognitive capabilities responsible for this outcome are behaving predictably in a given situation, and an alignment of psychological states ahead of signal production that derives from shared intentionality. Since both capabilities can exist independently of communication, our results are compatible with theories in which large flexible socially-learned communication systems like language are the product of a general but well-developed capacity for social cognition.
Authors: Sefika Efeoglu, Adrian Paschke, Sonja Schimmler
Abstract: Real-world data, such as news articles, social media posts, and chatbot conversations, is inherently dynamic and non-stationary, presenting significant challenges for constructing real-time structured representations through knowledge graphs (KGs). Relation Extraction (RE), a fundamental component of KG creation, often struggles to adapt to evolving data when traditional models rely on static, outdated datasets. Continual Relation Extraction (CRE) methods tackle this issue by incrementally learning new relations while preserving previously acquired knowledge. This study investigates the application of pre-trained language models (PLMs), specifically large language models (LLMs), to CRE, with a focus on leveraging memory replay to address catastrophic forgetting. We evaluate decoder-only models (eg, Mistral-7B and Llama2-7B) and encoder-decoder models (eg, Flan-T5 Base) on the TACRED and FewRel datasets. Task-incremental fine-tuning of LLMs demonstrates superior performance over earlier approaches using encoder-only models like BERT on TACRED, excelling in seen-task accuracy and overall performance (measured by whole and average accuracy), particularly with the Mistral and Flan-T5 models. Results on FewRel are similarly promising, achieving second place in whole and average accuracy metrics. This work underscores critical factors in knowledge transfer, language model architecture, and KG completeness, advancing CRE with LLMs and memory replay for dynamic, real-time relation extraction.
Authors: Jungyeul Park
Abstract: The development of lexicalized grammars, particularly Tree-Adjoining Grammar (TAG), has significantly advanced our understanding of syntax and semantics in natural language processing (NLP). While existing syntactic resources like the Penn Treebank and Universal Dependencies offer extensive annotations for phrase-structure and dependency parsing, there is a lack of large-scale corpora grounded in lexicalized grammar formalisms. To address this gap, we introduce TAGbank, a corpus of TAG derivations automatically extracted from existing syntactic treebanks. This paper outlines a methodology for mapping phrase-structure annotations to TAG derivations, leveraging the generative power of TAG to support parsing, grammar induction, and semantic analysis. Our approach builds on the work of CCGbank, extending it to incorporate the unique structural properties of TAG, including its transparent derivation trees and its ability to capture long-distance dependencies. We also discuss the challenges involved in the extraction process, including ensuring consistency across treebank schemes and dealing with language-specific syntactic idiosyncrasies. Finally, we propose the future extension of TAGbank to include multilingual corpora, focusing on the Penn Korean and Penn Chinese Treebanks, to explore the cross-linguistic application of TAG's formalism. By providing a robust, derivation-based resource, TAGbank aims to support a wide range of computational tasks and contribute to the theoretical understanding of TAG's generative capacity.
Authors: Yiming Zhang, Harshita Diddee, Susan Holm, Hanchen Liu, Xinyue Liu, Vinay Samuel, Barry Wang, Daphne Ippolito
Abstract: Language models have demonstrated remarkable capabilities on standard benchmarks, yet they struggle increasingly from mode collapse, the inability to generate diverse and novel outputs. Our work introduces NoveltyBench, a benchmark specifically designed to evaluate the ability of language models to produce multiple distinct and high-quality outputs. NoveltyBench utilizes prompts curated to elicit diverse answers and filtered real-world user queries. Evaluating 20 leading language models, we find that current state-of-the-art systems generate significantly less diversity than human writers. Notably, larger models within a family often exhibit less diversity than their smaller counterparts, challenging the notion that capability on standard benchmarks translates directly to generative utility. While prompting strategies like in-context regeneration can elicit diversity, our findings highlight a fundamental lack of distributional diversity in current models, reducing their utility for users seeking varied responses and suggesting the need for new training and evaluation paradigms that prioritize creativity alongside quality.
Authors: Hang Li, Yucheng Chu, Kaiqi Yang, Yasemin Copur-Gencturk, Jiliang Tang
Abstract: The rise of artificial intelligence (AI) technologies, particularly large language models (LLMs), has brought significant advancements to the field of education. Among various applications, automatic short answer grading (ASAG), which focuses on evaluating open-ended textual responses, has seen remarkable progress with the introduction of LLMs. These models not only enhance grading performance compared to traditional ASAG approaches but also move beyond simple comparisons with predefined "golden" answers, enabling more sophisticated grading scenarios, such as rubric-based evaluation. However, existing LLM-powered methods still face challenges in achieving human-level grading performance in rubric-based assessments due to their reliance on fully automated approaches. In this work, we explore the potential of LLMs in ASAG tasks by leveraging their interactive capabilities through a human-in-the-loop (HITL) approach. Our proposed framework, GradeHITL, utilizes the generative properties of LLMs to pose questions to human experts, incorporating their insights to refine grading rubrics dynamically. This adaptive process significantly improves grading accuracy, outperforming existing methods and bringing ASAG closer to human-level evaluation.
Authors: Yang Yan, Yu Lu, Renjun Xu, Zhenzhong Lan
Abstract: Despite high benchmark scores, Large Language Models (LLMs) often fail simple problem, raising a critical question: Do LLMs learn mathematical principles or merely memorize patterns? Rather than designing increasingly complex benchmarks like recent works, we investigate this using elementary two-integer addition ($0$ to $2^{64}$), probing two core properties: commutativity ($A+B=B+A$) and compositional generalization (via isomorphic symbolic mappings, e.g., $7 \rightarrow y$). While state-of-the-art LLMs achieve 73.8-99.8\% accuracy on numerical addition, performance collapses to $\leq$7.5\% under symbolic mapping, indicating failure to generalize learned rules. Non-monotonic performance scaling with digit count and frequent commutativity violations (over 1,700 cases of $A+B \neq B+A$) further support this. Explicitly providing addition rules degrades performance by 81.2\% on average, while self-explanation maintains baseline accuracy, suggesting LLM arithmetic processing is misaligned with human-defined principles. Our findings indicate current LLMs rely on memory pattern over genuine rule learning, highlighting architectural limitations and the need for new approaches to achieve true mathematical reasoning.
Authors: Yucheng Chu, Peng He, Hang Li, Haoyu Han, Kaiqi Yang, Yu Xue, Tingting Li, Joseph Krajcik, Jiliang Tang
Abstract: Short answer assessment is a vital component of science education, allowing evaluation of students' complex three-dimensional understanding. Large language models (LLMs) that possess human-like ability in linguistic tasks are increasingly popular in assisting human graders to reduce their workload. However, LLMs' limitations in domain knowledge restrict their understanding in task-specific requirements and hinder their ability to achieve satisfactory performance. Retrieval-augmented generation (RAG) emerges as a promising solution by enabling LLMs to access relevant domain-specific knowledge during assessment. In this work, we propose an adaptive RAG framework for automated grading that dynamically retrieves and incorporates domain-specific knowledge based on the question and student answer context. Our approach combines semantic search and curated educational sources to retrieve valuable reference materials. Experimental results in a science education dataset demonstrate that our system achieves an improvement in grading accuracy compared to baseline LLM approaches. The findings suggest that RAG-enhanced grading systems can serve as reliable support with efficient performance gains.
Authors: Pedro Ferreira, Wilker Aziz, Ivan Titov
Abstract: Chain-of-thought explanations are widely used to inspect the decision process of large language models (LLMs) and to evaluate the trustworthiness of model outputs, making them important for effective collaboration between LLMs and humans. We demonstrate that preference optimization - a key step in the alignment phase - can inadvertently reduce the faithfulness of these explanations. This occurs because the reward model (RM), which guides alignment, is tasked with optimizing both the expected quality of the response and the appropriateness of the explanations (e.g., minimizing bias or adhering to safety standards), creating potential conflicts. The RM lacks a mechanism to assess the consistency between the model's internal decision process and the generated explanation. Consequently, the LLM may engage in "reward hacking" by producing a final response that scores highly while giving an explanation tailored to maximize reward rather than accurately reflecting its reasoning. To address this issue, we propose enriching the RM's input with a causal attribution of the prediction, allowing the RM to detect discrepancies between the generated self-explanation and the model's decision process. In controlled settings, we show that this approach reduces the tendency of the LLM to generate misleading explanations.
Authors: Runpeng Dai, Run Yang, Fan Zhou, Hongtu Zhu
Abstract: Large Language Models (LLMs) and Vision-Language Models (VLMs) have become essential to general artificial intelligence, exhibiting remarkable capabilities in task understanding and problem-solving. However, the real-world reliability of these models critically depends on their stability, which remains an underexplored area. Despite their widespread use, rigorous studies examining the stability of LLMs under various perturbations are still lacking. In this paper, we address this gap by proposing a novel stability measure for LLMs, inspired by statistical methods rooted in information geometry. Our measure possesses desirable invariance properties, making it well-suited for analyzing model sensitivity to both parameter and input perturbations. To assess the effectiveness of our approach, we conduct extensive experiments on models ranging in size from 1.5B to 13B parameters. Our results demonstrate the utility of our measure in identifying salient parameters and detecting vulnerable regions in input images or critical dimensions in token embeddings. Furthermore, leveraging our stability framework, we enhance model robustness during model merging, leading to improved performance.
Authors: Zhiqiang Wang, Pengbin Feng, Yanbin Lin, Shuzhang Cai, Zongao Bian, Jinghua Yan, Xingquan Zhu
Abstract: We propose Fuzzy Group Relative Policy Reward (FGRPR), a novel framework that integrates Group Relative Policy Optimization (GRPO) with a fuzzy reward function to enhance learning efficiency. Unlike the conventional binary 0/1 accuracy reward, our fuzzy reward model provides nuanced incentives, encouraging more precise outputs. Experimental results demonstrate that GRPO with a standard 0/1 accuracy reward underperforms compared to supervised fine-tuning (SFT). In contrast, FGRPR, applied to Qwen2.5-VL(3B and 7B), surpasses all baseline models, including GPT4o, LLaMA2(90B), and SFT, across five in-domain datasets. On an out-of-domain dataset, FGRPR achieves performance comparable to SFT but excels when target values are larger, as its fuzzy reward function assigns higher rewards to closer approximations. This approach is broadly applicable to tasks where the precision of the answer is critical. Code and data: https://github.com/yeyimilk/CrowdVLM-R1
Authors: Erfan Shayegani, G M Shahariar, Sara Abdali, Lei Yu, Nael Abu-Ghazaleh, Yue Dong
Abstract: Multimodal Language Models (MMLMs) typically undergo post-training alignment to prevent harmful content generation. However, these alignment stages focus primarily on the assistant role, leaving the user role unaligned, and stick to a fixed input prompt structure of special tokens, leaving the model vulnerable when inputs deviate from these expectations. We introduce Role-Modality Attacks (RMA), a novel class of adversarial attacks that exploit role confusion between the user and assistant and alter the position of the image token to elicit harmful outputs. Unlike existing attacks that modify query content, RMAs manipulate the input structure without altering the query itself. We systematically evaluate these attacks across multiple Vision Language Models (VLMs) on eight distinct settings, showing that they can be composed to create stronger adversarial prompts, as also evidenced by their increased projection in the negative refusal direction in the residual stream, a property observed in prior successful attacks. Finally, for mitigation, we propose an adversarial training approach that makes the model robust against input prompt perturbations. By training the model on a range of harmful and benign prompts all perturbed with different RMA settings, it loses its sensitivity to Role Confusion and Modality Manipulation attacks and is trained to only pay attention to the content of the query in the input prompt structure, effectively reducing Attack Success Rate (ASR) while preserving the model's general utility.
Authors: Kaiyuan Hou, Minghui Zhao, Lilin Xu, Yuang Fan, Xiaofan Jiang
Abstract: The rapid emergence of Vision-Language Models (VLMs) has significantly advanced multimodal understanding, enabling applications in scene comprehension and visual reasoning. While these models have been primarily evaluated and developed for front-view image understanding, their capabilities in interpreting top-down images have received limited attention, partly due to the scarcity of diverse top-down datasets and the challenges in collecting such data. In contrast, top-down vision provides explicit spatial overviews and improved contextual understanding of scenes, making it particularly valuable for tasks like autonomous navigation, aerial imaging, and spatial planning. In this work, we address this gap by introducing TDBench, a comprehensive benchmark for VLMs in top-down image understanding. TDBench is constructed from public top-down view datasets and high-quality simulated images, including diverse real-world and synthetic scenarios. TDBench consists of visual question-answer pairs across ten evaluation dimensions of image understanding. Moreover, we conduct four case studies that commonly happen in real-world scenarios but are less explored. By revealing the strengths and limitations of existing VLM through evaluation results, we hope TDBench to provide insights for motivating future research. Project homepage: https://github.com/Columbia-ICSL/TDBench
Authors: Weiqing Li, Guochao Jiang, Xiangyong Ding, Zhangcheng Tao, Chuzhan Hao, Chenfeng Xu, Yuewei Zhang, Hao Wang
Abstract: Disaggregated inference has become an essential framework that separates the prefill (P) and decode (D) stages in large language model inference to improve throughput. However, the KV cache transfer faces significant delays between prefill and decode nodes. The block-wise calling method and discontinuous KV cache memory allocation increase the number of calls to the transmission kernel. Additionally, existing frameworks often fix the roles of P and D nodes, leading to computational imbalances. In this paper, we propose FlowKV, a novel disaggregated inference framework, which reduces the average transmission latency of KV cache by 96%, from 0.944s to 0.053s, almost eliminating the transfer time relative to the total request latency by optimizing the KV cache transfer. FlowKV introduces the Load-Aware Scheduler for balanced request scheduling and flexible PD node allocation. This design maximizes hardware resource utilization, achieving peak system throughput across various scenarios, including normal, computational imbalance, and extreme overload conditions. Experimental results demonstrate that FlowKV significantly accelerates inference by 15.2%-48.9% on LongBench dataset compared to the baseline and supports applications with heterogeneous GPUs.
Authors: Grgur Kova\v{c}, J\'er\'emy Perez, R\'emy Portelas, Peter Ford Dominey, Pierre-Yves Oudeyer
Abstract: Large language models (LLMs) are increasingly contributing to the creation of content on the Internet. This creates a feedback loop as subsequent generations of models will be trained on this generated, synthetic data. This phenomenon is receiving increasing interest, in particular because previous studies have shown that it may lead to distribution shift - models misrepresent and forget the true underlying distributions of human data they are expected to approximate (e.g. resulting in a drastic loss of quality). In this study, we study the impact of human data properties on distribution shift dynamics in iterated training loops. We first confirm that the distribution shift dynamics greatly vary depending on the human data by comparing four datasets (two based on Twitter and two on Reddit). We then test whether data quality may influence the rate of this shift. We find that it does on the twitter, but not on the Reddit datasets. We then focus on a Reddit dataset and conduct a more exhaustive evaluation of a large set of dataset properties. This experiment associated lexical diversity with larger, and semantic diversity with smaller detrimental shifts, suggesting that incorporating text with high lexical (but limited semantic) diversity could exacerbate the degradation of generated text. We then focus on the evolution of political bias, and find that the type of shift observed (bias reduction, amplification or inversion) depends on the political lean of the human (true) distribution. Overall, our work extends the existing literature on the consequences of recursive fine-tuning by showing that this phenomenon is highly dependent on features of the human data on which training occurs. This suggests that different parts of internet (e.g. GitHub, Reddit) may undergo different types of shift depending on their properties.
Authors: Chris Samarinas, Hamed Zamani
Abstract: We present a novel approach for training small language models for reasoning-intensive document ranking that combines knowledge distillation with reinforcement learning optimization. While existing methods often rely on expensive human annotations or large black-box language models, our methodology leverages web data and a teacher LLM to automatically generate high-quality training examples with relevance explanations. By framing document ranking as a reinforcement learning problem and incentivizing explicit reasoning capabilities, we train a compact 3B parameter language model that achieves state-of-the-art performance on the BRIGHT benchmark. Our model ranks third on the leaderboard while using substantially fewer parameters than other approaches, outperforming models that are over 20 times larger. Through extensive experiments, we demonstrate that generating explanations during inference, rather than directly predicting relevance scores, enables more effective reasoning with smaller language models. The self-supervised nature of our method offers a scalable and interpretable solution for modern information retrieval systems.
Authors: Dahun Kim, AJ Piergiovanni, Ganesh Mallya, Anelia Angelova
Abstract: We introduce VideoComp, a benchmark and learning framework for advancing video-text compositionality understanding, aimed at improving vision-language models (VLMs) in fine-grained temporal alignment. Unlike existing benchmarks focused on static image-text compositionality or isolated single-event videos, our benchmark targets alignment in continuous multi-event videos. Leveraging video-text datasets with temporally localized event captions (e.g. ActivityNet-Captions, YouCook2), we construct two compositional benchmarks, ActivityNet-Comp and YouCook2-Comp. We create challenging negative samples with subtle temporal disruptions such as reordering, action word replacement, partial captioning, and combined disruptions. These benchmarks comprehensively test models' compositional sensitivity across extended, cohesive video-text sequences. To improve model performance, we propose a hierarchical pairwise preference loss that strengthens alignment with temporally accurate pairs and gradually penalizes increasingly disrupted ones, encouraging fine-grained compositional learning. To mitigate the limited availability of densely annotated video data, we introduce a pretraining strategy that concatenates short video-caption pairs to simulate multi-event sequences. We evaluate video-text foundational models and large multimodal models (LMMs) on our benchmark, identifying both strengths and areas for improvement in compositionality. Overall, our work provides a comprehensive framework for evaluating and enhancing model capabilities in achieving fine-grained, temporally coherent video-text alignment.
Authors: Wasi Uddin Ahmad, Aleksander Ficek, Mehrzad Samadi, Jocelyn Huang, Vahid Noroozi, Somshubra Majumdar, Boris Ginsburg
Abstract: Large Language Models (LLMs) have transformed software development by enabling code generation, automated debugging, and complex reasoning. However, their continued advancement is constrained by the scarcity of high-quality, publicly available supervised fine-tuning (SFT) datasets tailored for coding tasks. To bridge this gap, we introduce OpenCodeInstruct, the largest open-access instruction tuning dataset, comprising 5 million diverse samples. Each sample includes a programming question, solution, test cases, execution feedback, and LLM-generated quality assessments. We fine-tune various base models, including LLaMA and Qwen, across multiple scales (1B+, 3B+, and 7B+) using our dataset. Comprehensive evaluations on popular benchmarks (HumanEval, MBPP, LiveCodeBench, and BigCodeBench) demonstrate substantial performance improvements achieved by SFT with OpenCodeInstruct. We also present a detailed methodology encompassing seed data curation, synthetic instruction and solution generation, and filtering.
Authors: Xin Quan, Marco Valentino, Danilo S. Carvalho, Dhairya Dalal, Andr\'e Freitas
Abstract: A persistent challenge in AI is the effective integration of material and formal inference - the former concerning the plausibility and contextual relevance of arguments, while the latter focusing on their logical and structural validity. Large Language Models (LLMs), by virtue of their extensive pre-training on large textual corpora, exhibit strong capabilities in material inference. However, their reasoning often lacks formal rigour and verifiability. At the same time, LLMs' linguistic competence positions them as a promising bridge between natural and formal languages, opening up new opportunities for combining these two modes of reasoning. In this paper, we introduce PEIRCE, a neuro-symbolic framework designed to unify material and formal inference through an iterative conjecture-criticism process. Within this framework, LLMs play the central role of generating candidate solutions in natural and formal languages, which are then evaluated and refined via interaction with external critique models. These critiques include symbolic provers, which assess formal validity, as well as soft evaluators that measure the quality of the generated arguments along linguistic and epistemic dimensions such as plausibility, coherence, and parsimony. While PEIRCE is a general-purpose framework, we demonstrate its capabilities in the domain of natural language explanation generation - a setting that inherently demands both material adequacy and formal correctness.
Authors: Marios Kokkodis, Richard Demsyn-Jones, Vijay Raghavan
Abstract: Are traditional classification approaches irrelevant in this era of AI hype? We show that there are multiclass classification problems where predictive models holistically outperform LLM prompt-based frameworks. Given text and images from home-service project descriptions provided by Thumbtack customers, we build embeddings-based softmax models that predict the professional category (e.g., handyman, bathroom remodeling) associated with each problem description. We then compare against prompts that ask state-of-the-art LLM models to solve the same problem. We find that the embeddings approach outperforms the best LLM prompts in terms of accuracy, calibration, latency, and financial cost. In particular, the embeddings approach has 49.5% higher accuracy than the prompting approach, and its superiority is consistent across text-only, image-only, and text-image problem descriptions. Furthermore, it yields well-calibrated probabilities, which we later use as confidence signals to provide contextualized user experience during deployment. On the contrary, prompting scores are overly uninformative. Finally, the embeddings approach is 14 and 81 times faster than prompting in processing images and text respectively, while under realistic deployment assumptions, it can be up to 10 times cheaper. Based on these results, we deployed a variation of the embeddings approach, and through A/B testing we observed performance consistent with our offline analysis. Our study shows that for multiclass classification problems that can leverage proprietary datasets, an embeddings-based approach may yield unequivocally better results. Hence, scientists, practitioners, engineers, and business leaders can use our study to go beyond the hype and consider appropriate predictive models for their classification use cases.
Authors: Yingcong Li, Davoud Ataee Tarzanagh, Ankit Singh Rawat, Maryam Fazel, Samet Oymak
Abstract: Linear attention methods offer a compelling alternative to softmax attention due to their efficiency in recurrent decoding. Recent research has focused on enhancing standard linear attention by incorporating gating while retaining its computational benefits. Such Gated Linear Attention (GLA) architectures include competitive models such as Mamba and RWKV. In this work, we investigate the in-context learning capabilities of the GLA model and make the following contributions. We show that a multilayer GLA can implement a general class of Weighted Preconditioned Gradient Descent (WPGD) algorithms with data-dependent weights. These weights are induced by the gating mechanism and the input, enabling the model to control the contribution of individual tokens to prediction. To further understand the mechanics of this weighting, we introduce a novel data model with multitask prompts and characterize the optimization landscape of learning a WPGD algorithm. Under mild conditions, we establish the existence and uniqueness (up to scaling) of a global minimum, corresponding to a unique WPGD solution. Finally, we translate these findings to explore the optimization landscape of GLA and shed light on how gating facilitates context-aware learning and when it is provably better than vanilla linear attention.
Authors: Jinyang Li, Sangwon Hyun, M. Ali Babar
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities in code generation. However, the quality of the generated code is heavily dependent on the structure and composition of the prompts used. Crafting high-quality prompts is a challenging task that requires significant knowledge and skills of prompt engineering. To advance the automation support for the prompt engineering for LLM-based code generation, we propose a novel solution Diffusion-Driven Prompt Tuning (DDPT) that learns how to generate optimal prompt embedding from Gaussian Noise to automate the prompt engineering for code generation. We evaluate the feasibility of diffusion-based optimization and abstract the optimal prompt embedding as a directional vector toward the optimal embedding. We use the code generation loss given by the LLMs to help the diffusion model capture the distribution of optimal prompt embedding during training. The trained diffusion model can build a path from the noise distribution to the optimal distribution at the sampling phrase, the evaluation result demonstrates that DDPT helps improve the prompt optimization for code generation.
Authors: Ximing Lu, Seungju Han, David Acuna, Hyunwoo Kim, Jaehun Jung, Shrimai Prabhumoye, Niklas Muennighoff, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro, Yejin Choi
Abstract: Large reasoning models exhibit remarkable reasoning capabilities via long, elaborate reasoning trajectories. Supervised fine-tuning on such reasoning traces, also known as distillation, can be a cost-effective way to boost reasoning capabilities of student models. However, empirical observations reveal that these reasoning trajectories are often suboptimal, switching excessively between different lines of thought, resulting in under-thinking, over-thinking, and even degenerate responses. We introduce Retro-Search, an MCTS-inspired search algorithm, for distilling higher quality reasoning paths from large reasoning models. Retro-Search retrospectively revises reasoning paths to discover better, yet shorter traces, which can then lead to student models with enhanced reasoning capabilities with shorter, thus faster inference. Our approach can enable two use cases: self-improvement, where models are fine-tuned on their own Retro-Search-ed thought traces, and weak-to-strong improvement, where a weaker model revises stronger model's thought traces via Retro-Search. For self-improving, R1-distill-7B, fine-tuned on its own Retro-Search-ed traces, reduces the average reasoning length by 31.2% while improving performance by 7.7% across seven math benchmarks. For weak-to-strong improvement, we retrospectively revise R1-671B's traces from the OpenThoughts dataset using R1-distill-32B as the Retro-Search-er, a model 20x smaller. Qwen2.5-32B, fine-tuned on this refined data, achieves performance comparable to R1-distill-32B, yielding an 11.3% reduction in reasoning length and a 2.4% performance improvement compared to fine-tuning on the original OpenThoughts data. Our work counters recently emergent viewpoints that question the relevance of search algorithms in the era of large reasoning models, by demonstrating that there are still opportunities for algorithmic advancements, even for frontier models.
Authors: Mohammad Amaan Sayeed, Engin Tekin, Maryam Nadeem, Nancy A. ElNaker, Aahan Singh, Natalia Vassilieva, Boulbaba Ben Amor
Abstract: Unlocking the next generation of biotechnology and therapeutic innovation demands overcoming the inherent complexity and resource-intensity of conventional protein engineering methods. Recent GenAI-powered computational techniques often rely on the availability of the target protein's 3D structures and specific binding sites to generate high-affinity binders, constraints exhibited by models such as AlphaProteo and RFdiffusion. In this work, we explore the use of Protein Language Models (pLMs) for high-affinity binder generation. We introduce Prot42, a novel family of Protein Language Models (pLMs) pretrained on vast amounts of unlabeled protein sequences. By capturing deep evolutionary, structural, and functional insights through an advanced auto-regressive, decoder-only architecture inspired by breakthroughs in natural language processing, Prot42 dramatically expands the capabilities of computational protein design based on language only. Remarkably, our models handle sequences up to 8,192 amino acids, significantly surpassing standard limitations and enabling precise modeling of large proteins and complex multi-domain sequences. Demonstrating powerful practical applications, Prot42 excels in generating high-affinity protein binders and sequence-specific DNA-binding proteins. Our innovative models are publicly available, offering the scientific community an efficient and precise computational toolkit for rapid protein engineering.
Authors: Ivan Ilin
Abstract: Computing the full Hessian matrix -- the matrix of second-order derivatives for an entire Large Language Model (LLM) is infeasible due to its sheer size. In this technical report, we aim to provide a comprehensive guide on how to accurately compute at least a small portion of the Hessian for LLMs using PyTorch autograd library. We also demonstrate how to compute the full diagonal of the Hessian matrix using multiple samples of vector-Hessian Products (HVPs). We hope that both this guide and the accompanying GitHub code will be valuable resources for practitioners and researchers interested in better understanding the behavior and structure of the Hessian in LLMs.
Authors: Noga Ben Yoash, Meni Brief, Oded Ovadia, Gil Shenderovitz, Moshik Mishaeli, Rachel Lemberg, Eitam Sheetrit
Abstract: We introduce SECQUE, a comprehensive benchmark for evaluating large language models (LLMs) in financial analysis tasks. SECQUE comprises 565 expert-written questions covering SEC filings analysis across four key categories: comparison analysis, ratio calculation, risk assessment, and financial insight generation. To assess model performance, we develop SECQUE-Judge, an evaluation mechanism leveraging multiple LLM-based judges, which demonstrates strong alignment with human evaluations. Additionally, we provide an extensive analysis of various models' performance on our benchmark. By making SECQUE publicly available, we aim to facilitate further research and advancements in financial AI.
Authors: Alberto Larrauri
Abstract: It is an open question whether the search and decision versions of promise CSPs are equivalent. Most known algorithms for PCSPs solve only their \emph{decision} variant, and it is unknown whether they can be adapted to solve \emph{search} as well. The main approaches, called BLP, AIP and BLP+AIP, handle a PCSP by finding a solution to a relaxation of some integer program. We prove that rounding those solutions to a proper search certificate can be as hard as any problem in the class TFNP. In other words, these algorithms are ineffective for search. Building on the algebraic approach to PCSPs, we find sufficient conditions that imply ineffectiveness for search. Our tools are tailored to algorithms that are characterized by minions in a suitable way, and can also be used to prove undecidability results for meta-problems. This way, we show that the families of templates solvable via BLP, AIP, and BLP+AIP are undecidable. Using the same techniques we also analyze several algebraic conditions that are known to guarantee the tractability of finite-template CSPs. We prove that several meta-problems related to cyclic polymorphims and WNUs are undecidable for PCSPs. In particular, there is no algorithm deciding whether a finite PCSP template (1) admits cyclic a polymorphism, (2) admits a WNU.
Authors: Yimu Wang, Mozhgan Nasr Azadani, Sean Sedwards, Krzysztof Czarnecki
Abstract: Redundancy of visual tokens in multi-modal large language models (MLLMs) significantly reduces their computational efficiency. Recent approaches, such as resamplers and summarizers, have sought to reduce the number of visual tokens, but at the cost of visual reasoning ability. To address this, we propose LEO-MINI, a novel MLLM that significantly reduces the number of visual tokens and simultaneously boosts visual reasoning capabilities. For efficiency, LEO-MINI incorporates CoTR, a novel token reduction module to consolidate a large number of visual tokens into a smaller set of tokens, using the similarity between visual tokens, text tokens, and a compact learnable query. For effectiveness, to scale up the model's ability with minimal computational overhead, LEO-MINI employs MMoE, a novel mixture of multi-modal experts module. MMOE employs a set of LoRA experts with a novel router to switch between them based on the input text and visual tokens instead of only using the input hidden state. MMoE also includes a general LoRA expert that is always activated to learn general knowledge for LLM reasoning. For extracting richer visual features, MMOE employs a set of vision experts trained on diverse domain-specific data. To demonstrate LEO-MINI's improved efficiency and performance, we evaluate it against existing efficient MLLMs on various benchmark vision-language tasks.
Authors: Martin Weyssow, Chengran Yang, Junkai Chen, Yikun Li, Huihui Huang, Ratnadira Widyasari, Han Wei Ang, Frank Liauw, Eng Lieh Ouh, Lwin Khin Shar, David Lo
Abstract: Large language models (LLMs) have shown promising performance in software vulnerability detection (SVD), yet their reasoning capabilities remain unreliable. Existing approaches relying on chain-of-thought (CoT) struggle to provide relevant and actionable security assessments. Additionally, effective SVD requires not only generating coherent reasoning but also differentiating between well-founded and misleading yet plausible security assessments, an aspect overlooked in prior work. To this end, we introduce R2Vul, a novel approach that distills structured reasoning into small LLMs using reinforcement learning from AI feedback (RLAIF). Through RLAIF, R2Vul enables LLMs to produce structured, security-aware reasoning that is actionable and reliable while explicitly learning to distinguish valid assessments from misleading ones. We evaluate R2Vul across five languages against SAST tools, CoT, instruction tuning, and classification-based baselines. Our results show that R2Vul with structured reasoning distillation enables a 1.5B student LLM to rival larger models while improving generalization to out-of-distribution vulnerabilities. Beyond model improvements, we contribute a large-scale, multilingual preference dataset featuring structured reasoning to support future research in SVD.
Authors: Manlai Liang, JiaMing Zhang, Xiong Li, Jinlong Li
Abstract: The increasing size of the Key-Value (KV) cache during the Large Language Models long-context inference is the main obstacle for its balance between the deployment cost and task accuracy. To reduce the KV cache size in such scenarios, most previous efforts leveraged on the attention weight to evict non-critical cache tokens. But there is a trade-off in those methods, they usually require major modifiation of the inference infrastructure and significant computation overhead. Base on the fact that the Large Lanuage models are autoregresssive models, we propose {\it LagKV}, a KV allocation strategy only relying on straight forward comparison among KV themself. It is a totally attention free method which offers easy integration to the main stream inference platform and comparable performance comparing to other complicated KV compression methods. Results on LongBench and PasskeyRetrieval show that, our approach achieves nearly zero loss when the ratio is $2\times$ and $\approx 90\%$ of the original model performance for $8\times$. Especially in the 64-digit passkey retrieval task, our mehod outperforms the attention weight based method $H_2O$ over $60\%$ with same compression ratios. Our code is available at \url{https://github.com/AI-Lab-China-Merchants-Bank/LagKV}.
Authors: Anna Goldie, Azalia Mirhoseini, Hao Zhou, Irene Cai, Christopher D. Manning
Abstract: Reinforcement learning has been shown to improve the performance of large language models. However, traditional approaches like RLHF or RLAIF treat the problem as single-step. As focus shifts toward more complex reasoning and agentic tasks, language models must take multiple steps of text generation, reasoning and environment interaction before generating a solution. We propose a synthetic data generation and RL methodology targeting multi-step optimization scenarios. This approach, called Step-Wise Reinforcement Learning (SWiRL), iteratively generates multi-step reasoning and tool use data, and then learns from that data. It employs a simple step-wise decomposition that breaks each multi-step trajectory into multiple sub-trajectories corresponding to each action by the original model. It then applies synthetic data filtering and RL optimization on these sub-trajectories. We evaluated SWiRL on a number of multi-step tool use, question answering, and mathematical reasoning tasks. Our experiments show that SWiRL outperforms baseline approaches by 21.5%, 12.3%, 14.8%, 11.1%, and 15.3% in relative accuracy on GSM8K, HotPotQA, CofCA, MuSiQue, and BeerQA, respectively. Excitingly, the approach exhibits generalization across tasks: for example, training only on HotPotQA (text question-answering) improves zero-shot performance on GSM8K (a math dataset) by a relative 16.9%.
Authors: Danial Amin, Joni Salminen, Farhan Ahmed, Sonja M. H. Tervola, Sankalp Sethi, Bernard J. Jansen
Abstract: Although Generative AI (GenAI) has the potential for persona development, many challenges must be addressed. This research systematically reviews 52 articles from 2022-2024, with important findings. First, closed commercial models are frequently used in persona development, creating a monoculture Second, GenAI is used in various stages of persona development (data collection, segmentation, enrichment, and evaluation). Third, similar to other quantitative persona development techniques, there are major gaps in persona evaluation for AI generated personas. Fourth, human-AI collaboration models are underdeveloped, despite human oversight being crucial for maintaining ethical standards. These findings imply that realizing the full potential of AI-generated personas will require substantial efforts across academia and industry. To that end, we provide a list of research avenues to inspire future work.
Authors: Rean Fernandes, Andr\'e Biedenkapp, Frank Hutter, Noor Awad
Abstract: Legal reasoning tasks present unique challenges for large language models (LLMs) due to the complexity of domain-specific knowledge and reasoning processes. This paper investigates how effectively smaller language models (Llama 2 7B and Llama 3 8B) can be fine-tuned with a limited dataset of 1,514 Multi-state Bar Examination (MBE) questions to improve legal question answering accuracy. We evaluate these models on the 2022 MBE questions licensed from JD Advising, the same dataset used in the 'GPT-4 passes the Bar exam' study. Our methodology involves collecting approximately 200 questions per legal domain across 7 domains. We distill the dataset using Llama 3 (70B) to transform explanations into a structured IRAC (Issue, Rule, Application, Conclusion) format as a guided reasoning process to see if it results in better performance over the non-distilled dataset. We compare the non-fine-tuned models against their supervised fine-tuned (SFT) counterparts, trained for different sample sizes per domain, to study the effect on accuracy and prompt adherence. We also analyse option selection biases and their mitigation following SFT. In addition, we consolidate the performance across multiple variables: prompt type (few-shot vs zero-shot), answer ordering (chosen-option first vs generated-explanation first), response format (Numbered list vs Markdown vs JSON), and different decoding temperatures. Our findings show that domain-specific SFT helps some model configurations achieve close to human baseline performance, despite limited computational resources and a relatively small dataset. We release both the gathered SFT dataset and the family of Supervised Fine-tuned (SFT) adapters optimised for MBE performance. This establishes a practical lower bound on resources needed towards achieving effective legal question answering in smaller LLMs.
Authors: Ming Li, Ruiyi Zhang, Jian Chen, Jiuxiang Gu, Yufan Zhou, Franck Dernoncourt, Wanrong Zhu, Tianyi Zhou, Tong Sun
Abstract: Despite the existing evolution of Multimodal Large Language Models (MLLMs), a non-neglectable limitation remains in their struggle with visual text grounding, especially in text-rich images of documents. Document images, such as scanned forms and infographics, highlight critical challenges due to their complex layouts and textual content. However, current benchmarks do not fully address these challenges, as they mostly focus on visual grounding on natural images, rather than text-rich document images. Thus, to bridge this gap, we introduce TRIG, a novel task with a newly designed instruction dataset for benchmarking and improving the Text-Rich Image Grounding capabilities of MLLMs in document question-answering. Specifically, we propose an OCR-LLM-human interaction pipeline to create 800 manually annotated question-answer pairs as a benchmark and a large-scale training set of 90$ synthetic data based on four diverse datasets. A comprehensive evaluation of various MLLMs on our proposed benchmark exposes substantial limitations in their grounding capability on text-rich images. In addition, we propose two simple and effective TRIG methods based on general instruction tuning and plug-and-play efficient embedding, respectively. By finetuning MLLMs on our synthetic dataset, they promisingly improve spatial reasoning and grounding capabilities.
Authors: Ngoc Bui, Hieu Trung Nguyen, Shantanu Kumar, Julian Theodore, Weikang Qiu, Viet Anh Nguyen, Rex Ying
Abstract: Advances in Large Language Models (LLMs) paved the way for their emerging applications in various domains, such as human behavior simulations, where LLMs could augment human-generated data in social science research and machine learning model training. However, pretrained LLMs often fail to capture the behavioral diversity of target populations due to the inherent variability across individuals and groups. To address this, we propose \textit{Mixture of Personas} (MoP), a \textit{probabilistic} prompting method that aligns the LLM responses with the target population. MoP is a contextual mixture model, where each component is an LM agent characterized by a persona and an exemplar representing subpopulation behaviors. The persona and exemplar are randomly chosen according to the learned mixing weights to elicit diverse LLM responses during simulation. MoP is flexible, requires no model finetuning, and is transferable across base models. Experiments for synthetic data generation show that MoP outperforms competing methods in alignment and diversity metrics.
Authors: Hengran Zhang, Keping Bi, Jiafeng Guo, Xiaojie Sun, Shihao Liu, Daiting Shi, Dawei Yin, Xueqi Cheng
Abstract: Dense retrieval is a crucial task in Information Retrieval (IR) and is the foundation for downstream tasks such as re-ranking. Recently, large language models (LLMs) have shown compelling semantic understanding capabilities and are appealing to researchers studying dense retrieval. LLMs, as decoder-style generative models, are competent at language generation while falling short on modeling global information due to the lack of attention to tokens afterward. Inspired by the classical word-based language modeling approach for IR, i.e., the query likelihood (QL) model, we seek to sufficiently utilize LLMs' generative ability by QL maximization. However, instead of ranking documents with QL estimation, we introduce an auxiliary task of QL maximization to yield a better backbone for contrastively learning a discriminative retriever. We name our model as LLM-QL. To condense global document semantics to a single vector during QL modeling, LLM-QL has two major components, Attention Stop (AS) and Input Corruption (IC). AS stops the attention of predictive tokens to previous tokens until the ending token of the document. IC masks a portion of tokens in the input documents during prediction. Experiments on MSMARCO show that LLM-QL can achieve significantly better performance than other LLM-based retrievers and using QL estimated by LLM-QL for ranking outperforms word-based QL by a large margin.
Authors: Hengran Zhang, Minghao Tang, Keping Bi, Jiafeng Guo, Shihao Liu, Daiting Shi, Dawei Yin, Xueqi Cheng
Abstract: Retrieval models typically rely on costly human-labeled query-document relevance annotations for training and evaluation. To reduce this cost and leverage the potential of Large Language Models (LLMs) in relevance judgments, we aim to explore whether LLM-generated annotations can effectively replace human annotations in training retrieval models. Retrieval usually emphasizes relevance, which indicates "topic-relatedness" of a document to a query, while in RAG, the value of a document (or utility) depends on how it contributes to answer generation. Recognizing this mismatch, some researchers use LLM performance on downstream tasks with documents as labels, but this approach requires manual answers for specific tasks, leading to high costs and limited generalization. In another line of work, prompting LLMs to select useful documents as RAG references eliminates the need for human annotation and is not task-specific. If we leverage LLMs' utility judgments to annotate retrieval data, we may retain cross-task generalization without human annotation in large-scale corpora. Therefore, we investigate utility-focused annotation via LLMs for large-scale retriever training data across both in-domain and out-of-domain settings on the retrieval and RAG tasks. To reduce the impact of low-quality positives labeled by LLMs, we design a novel loss function, i.e., Disj-InfoNCE. Our experiments reveal that: (1) Retrievers trained on utility-focused annotations significantly outperform those trained on human annotations in the out-of-domain setting on both tasks, demonstrating superior generalization capabilities. (2) LLM annotation does not replace human annotation in the in-domain setting. However, incorporating just 20% human-annotated data enables retrievers trained with utility-focused annotations to match the performance of models trained entirely with human annotations.
Authors: Adri\'an Bazaga, Rexhina Blloshmi, Bill Byrne, Adri\`a de Gispert
Abstract: Large Language Models (LLMs) have emerged as powerful tools for generating coherent text, understanding context, and performing reasoning tasks. However, they struggle with temporal reasoning, which requires processing time-related information such as event sequencing, durations, and inter-temporal relationships. These capabilities are critical for applications including question answering, scheduling, and historical analysis. In this paper, we introduce TISER, a novel framework that enhances the temporal reasoning abilities of LLMs through a multi-stage process that combines timeline construction with iterative self-reflection. Our approach leverages test-time scaling to extend the length of reasoning traces, enabling models to capture complex temporal dependencies more effectively. This strategy not only boosts reasoning accuracy but also improves the traceability of the inference process. Experimental results demonstrate state-of-the-art performance across multiple benchmarks, including out-of-distribution test sets, and reveal that TISER enables smaller open-source models to surpass larger closed-weight models on challenging temporal reasoning tasks.
Authors: Mingyang Fu, Yuyang Peng, Benlin Liu, Yao Wan, Dongping Chen
Abstract: We introduce LiveVQA, an automatically collected dataset of latest visual knowledge from the Internet with synthesized VQA problems. LiveVQA consists of 3,602 single- and multi-hop visual questions from 6 news websites across 14 news categories, featuring high-quality image-text coherence and authentic information. Our evaluation across 15 MLLMs (e.g., GPT-4o, Gemma-3, and Qwen-2.5-VL family) demonstrates that stronger models perform better overall, with advanced visual reasoning capabilities proving crucial for complex multi-hop questions. Despite excellent performance on textual problems, models with tools like search engines still show significant gaps when addressing visual questions requiring latest visual knowledge, highlighting important areas for future research.
Authors: Yuwei Zhang, Haode Zhang, Li-Ming Zhan, Albert Y. S. Lam, Xiao-Ming Wu
Abstract: New intent discovery aims to uncover novel intent categories from user utterances to expand the set of supported intent classes. It is a critical task for the development and service expansion of a practical dialogue system. Despite its importance, this problem remains under-explored in the literature. Existing approaches typically rely on a large amount of labeled utterances and employ pseudo-labeling methods for representation learning and clustering, which are label-intensive, inefficient, and inaccurate. In this paper, we provide new solutions to two important research questions for new intent discovery: (1) how to learn semantic utterance representations and (2) how to better cluster utterances. Particularly, we first propose a multi-task pre-training strategy to leverage rich unlabeled data along with external labeled data for representation learning. Then, we design a new contrastive loss to exploit self-supervisory signals in unlabeled data for clustering. Extensive experiments on three intent recognition benchmarks demonstrate the high effectiveness of our proposed method, which outperforms state-of-the-art methods by a large margin in both unsupervised and semi-supervised scenarios. The source code will be available at https://github.com/zhang-yu-wei/MTP-CLNN.
Authors: Aiyuan Yang, Bin Xiao, Bingning Wang, Borong Zhang, Ce Bian, Chao Yin, Chenxu Lv, Da Pan, Dian Wang, Dong Yan, Fan Yang, Fei Deng, Feng Wang, Feng Liu, Guangwei Ai, Guosheng Dong, Haizhou Zhao, Hang Xu, Haoze Sun, Hongda Zhang, Hui Liu, Jiaming Ji, Jian Xie, JunTao Dai, Kun Fang, Lei Su, Liang Song, Lifeng Liu, Liyun Ru, Luyao Ma, Mang Wang, Mickel Liu, MingAn Lin, Nuolan Nie, Peidong Guo, Ruiyang Sun, Tao Zhang, Tianpeng Li, Tianyu Li, Wei Cheng, Weipeng Chen, Xiangrong Zeng, Xiaochuan Wang, Xiaoxi Chen, Xin Men, Xin Yu, Xuehai Pan, Yanjun Shen, Yiding Wang, Yiyu Li, Youxin Jiang, Yuchen Gao, Yupeng Zhang, Zenan Zhou, Zhiying Wu
Abstract: Large language models (LLMs) have demonstrated remarkable performance on a variety of natural language tasks based on just a few examples of natural language instructions, reducing the need for extensive feature engineering. However, most powerful LLMs are closed-source or limited in their capability for languages other than English. In this technical report, we present Baichuan 2, a series of large-scale multilingual language models containing 7 billion and 13 billion parameters, trained from scratch, on 2.6 trillion tokens. Baichuan 2 matches or outperforms other open-source models of similar size on public benchmarks like MMLU, CMMLU, GSM8K, and HumanEval. Furthermore, Baichuan 2 excels in vertical domains such as medicine and law. We will release all pre-training model checkpoints to benefit the research community in better understanding the training dynamics of Baichuan 2.
Authors: Haotian Li, Bin Yu, Yuliang Wei, Kai Wang, Richard Yi Da Xu, Bailing Wang
Abstract: Knowledge graph completion (KGC) revolves around populating missing triples in a knowledge graph using available information. Text-based methods, which depend on textual descriptions of triples, often encounter difficulties when these descriptions lack sufficient information for accurate prediction-an issue inherent to the datasets and not easily resolved through modeling alone. To address this and ensure data consistency, we first use large language models (LLMs) to generate coherent descriptions, bridging the semantic gap between queries and answers. Secondly, we utilize inverse relations to create a symmetric graph, thereby providing augmented training samples for KGC. Additionally, we employ the label information inherent in knowledge graphs (KGs) to enhance the existing contrastive framework, making it fully supervised. These efforts have led to significant performance improvements on the WN18RR and FB15k-237 datasets. According to standard evaluation metrics, our approach achieves a 4.2% improvement in Hit@1 on WN18RR and a 3.4% improvement in Hit@3 on FB15k-237, demonstrating superior performance.
Authors: Christopher Davis, Andrew Caines, {\O}istein Andersen, Shiva Taslimipoor, Helen Yannakoudakis, Zheng Yuan, Christopher Bryant, Marek Rei, Paula Buttery
Abstract: Thanks to recent advances in generative AI, we are able to prompt large language models (LLMs) to produce texts which are fluent and grammatical. In addition, it has been shown that we can elicit attempts at grammatical error correction (GEC) from LLMs when prompted with ungrammatical input sentences. We evaluate how well LLMs can perform at GEC by measuring their performance on established benchmark datasets. We go beyond previous studies, which only examined GPT* models on a selection of English GEC datasets, by evaluating seven open-source and three commercial LLMs on four established GEC benchmarks. We investigate model performance and report results against individual error types. Our results indicate that LLMs do not always outperform supervised English GEC models except in specific contexts -- namely commercial LLMs on benchmarks annotated with fluency corrections as opposed to minimal edits. We find that several open-source models outperform commercial ones on minimal edit benchmarks, and that in some settings zero-shot prompting is just as competitive as few-shot prompting.
Authors: Jiacheng Liu, Sewon Min, Luke Zettlemoyer, Yejin Choi, Hannaneh Hajishirzi
Abstract: Are $n$-gram language models still relevant in this era of neural large language models (LLMs)? Our answer is yes, and we showcase their values in both text analysis and improving neural LLMs. This was done by modernizing $n$-gram LMs in two aspects. First, we train them at the same data scale as neural LLMs -- 5 trillion tokens. This is the largest $n$-gram LM ever built. Second, existing $n$-gram LMs use small $n$ which hinders their performance; we instead allow $n$ to be arbitrarily large, by introducing a new $\infty$-gram LM with backoff. Instead of pre-computing $n$-gram count tables (which would be very expensive), we develop an engine named infini-gram -- powered by suffix arrays -- that can compute $\infty$-gram (as well as $n$-gram with arbitrary $n$) probabilities with millisecond-level latency. The $\infty$-gram framework and infini-gram engine enable us to conduct many novel and interesting analyses of human-written and machine-generated text: we find that the $\infty$-gram LM has fairly high accuracy for next-token prediction (47%), and can complement neural LLMs to greatly reduce their perplexity. When analyzing machine-generated text, we also observe irregularities in the machine--$\infty$-gram agreement level with respect to the suffix length, which indicates deficiencies in neural LLM pretraining and the positional embeddings of Transformers.
Authors: Yuhang Zhou, Paiheng Xu, Xiyao Wang, Xuan Lu, Ge Gao, Wei Ai
Abstract: Emojis, which encapsulate semantics beyond mere words or phrases, have become prevalent in social network communications. This has spurred increasing scholarly interest in exploring their attributes and functionalities. However, emoji-related research and application face two primary challenges. First, researchers typically rely on crowd-sourcing to annotate emojis in order to understand their sentiments, usage intentions, and semantic meanings. Second, subjective interpretations by users can often lead to misunderstandings of emojis and cause the communication barrier. Large Language Models (LLMs) have achieved significant success in various annotation tasks, with ChatGPT demonstrating expertise across multiple domains. In our study, we assess ChatGPT's effectiveness in handling previously annotated and downstream tasks. Our objective is to validate the hypothesis that ChatGPT can serve as a viable alternative to human annotators in emoji research and that its ability to explain emoji meanings can enhance clarity and transparency in online communications. Our findings indicate that ChatGPT has extensive knowledge of emojis. It is adept at elucidating the meaning of emojis across various application scenarios and demonstrates the potential to replace human annotators in a range of tasks.
Authors: Jiatong Li, Wei Liu, Zhihao Ding, Wenqi Fan, Yuqiang Li, Qing Li
Abstract: Large Language Models (LLMs) have demonstrated exceptional performance in biochemical tasks, especially the molecule caption translation task, which aims to bridge the gap between molecules and natural language texts. However, previous methods in adapting LLMs to the molecule-caption translation task required extra domain-specific pre-training stages, suffered weak alignment between molecular and textual spaces, or imposed stringent demands on the scale of LLMs. To resolve the challenges, we propose In-Context Molecule Adaptation (ICMA), as a new paradigm allowing LLMs to learn the molecule-text alignment from context examples via In-Context Molecule Tuning. Specifically, ICMA incorporates the following three stages: Hybrid Context Retrieval, Post-retrieval Re-ranking, and In-context Molecule Tuning. Initially, Hybrid Context Retrieval utilizes BM25 Caption Retrieval and Molecule Graph Retrieval to retrieve similar informative context examples. Additionally, Post-retrieval Re-ranking is composed of Sequence Reversal and Random Walk selection to further improve the quality of retrieval results. Finally, In-Context Molecule Tuning unlocks the in-context learning and reasoning capability of LLMs with the retrieved examples and adapts the parameters of LLMs for better alignment between molecules and texts. Experimental results demonstrate that ICMA can empower LLMs to achieve state-of-the-art or comparable performance without extra training corpora and intricate structures, showing that LLMs are inherently in-context molecule learners.
Authors: Andrea Nini, Oren Halvani, Lukas Graner, Valerio Gherardi, Shunichi Ishihara
Abstract: Authorship Verification (AV) is a key area of research in digital text forensics, which addresses the fundamental question of whether two texts were written by the same person. Numerous computational approaches have been proposed over the last two decades in an attempt to address this challenge. However, existing AV methods often suffer from high complexity, low explainability and especially from a lack of clear scientific justification. We propose a simpler method based on modeling the grammar of an author following Cognitive Linguistics principles. These models are used to calculate $\lambda_G$ (LambdaG): the ratio of the likelihoods of a document given the candidate's grammar versus given a reference population's grammar. Our empirical evaluation, conducted on twelve datasets and compared against seven baseline methods, demonstrates that LambdaG achieves superior performance, including against several neural network-based AV methods. LambdaG is also robust to small variations in the composition of the reference population and provides interpretable visualizations, enhancing its explainability. We argue that its effectiveness is due to the method's compatibility with Cognitive Linguistics theories predicting that a person's grammar is a behavioral biometric.
Authors: Yu-Ying Chuang, Melanie J. Bell, Yu-Hsiang Tseng, R. Harald Baayen
Abstract: The pitch contours of Mandarin two-character words are generally understood as being shaped by the underlying tones of the constituent single-character words, in interaction with articulatory constraints imposed by factors such as speech rate, co-articulation with adjacent tones, segmental make-up, and predictability. This study shows that tonal realization is also partially determined by words' meanings. We first show, on the basis of a corpus of Taiwan Mandarin spontaneous conversations, using a generalized additive regression model, and focusing on the rise-fall tone pattern, that after controlling for effects of speaker and context, word type is a stronger predictor of tonal realization than all the previously established word-form related predictors combined. Importantly, the addition of information about meaning in context improves prediction accuracy even further. We then proceed to show, using computational modeling with context-specific word embeddings, that token-specific pitch contours predict word type with 50% accuracy on held-out data, and that context-sensitive, token-specific embeddings can predict the shape of pitch contours with 40% accuracy. These accuracies, which are an order of magnitude above chance level, suggest that the relation between words' pitch contours and their meanings are sufficiently strong to be potentially functional for language users. The theoretical implications of these empirical findings are discussed.
Authors: Mubashara Akhtar, Chenxi Pang, Andreea Marzoca, Yasemin Altun, Julian Martin Eisenschlos
Abstract: Language models, potentially augmented with tool usage such as retrieval are becoming the go-to means of answering questions. Understanding and answering questions in real-world settings often requires retrieving information from different sources, processing and aggregating data to extract insights, and presenting complex findings in form of structured artifacts such as novel tables, charts, or infographics. In this paper, we introduce TANQ, the first open domain question answering dataset where the answers require building tables from information across multiple sources. We release the full source attribution for every cell in the resulting table and benchmark state-of-the-art language models in open, oracle, and closed book setups. Our best-performing baseline, Gemini Flash reaches an overall F1 score of 60.7, lagging behind human performance by 12.3 points. We analyse baselines' performance across different dataset attributes such as different skills required for this task, including multi-hop reasoning, math operations, and unit conversions. We further discuss common failures in model-generated answers, suggesting that TANQ is a complex task with many challenges ahead.
Authors: Bingxiang He, Ning Ding, Cheng Qian, Jia Deng, Ganqu Cui, Lifan Yuan, Haiwen Hong, Huan-ang Gao, Longtao Huang, Hui Xue, Huimin Chen, Zhiyuan Liu, Maosong Sun
Abstract: Understanding alignment techniques begins with comprehending zero-shot generalization brought by instruction tuning, but little of the mechanism has been understood. Existing work has largely been confined to the task level, without considering that tasks are artificially defined and, to LLMs, merely consist of tokens and representations. To bridge this gap, we investigate zero-shot generalization from the perspective of the data itself. We first demonstrate that zero-shot generalization happens very early during instruction tuning, with loss serving as a stable indicator. Next, we investigate training data arrangement through similarity and granularity perspectives, confirming that the timing of exposure to certain training examples may greatly facilitate generalization on unseen tasks. Finally, we propose a more grounded training data arrangement framework, Test-centric Multi-turn Arrangement, and show its effectiveness in promoting continual learning and further loss reduction. For the first time, we show that zero-shot generalization during instruction tuning is a form of similarity-based generalization between training and test data at the instance level. Our code is released at https://github.com/thunlp/Dynamics-of-Zero-Shot-Generalization.
URLs: https://github.com/thunlp/Dynamics-of-Zero-Shot-Generalization.
Authors: Yixing Li, Yuxian Gu, Li Dong, Dequan Wang, Yu Cheng, Furu Wei
Abstract: In the field of large language models (LLMs), Knowledge Distillation (KD) is a critical technique for transferring capabilities from teacher models to student models. However, existing KD methods face limitations and challenges in distillation of LLMs, including efficiency and insufficient measurement capabilities of traditional KL divergence. It is shown that LLMs can serve as an implicit reward function, which we define as a supplement to KL divergence. In this work, we propose Direct Preference Knowledge Distillation (DPKD) for LLMs. DPKD utilizes distribution divergence to represent the preference loss and implicit reward function. We re-formulate KD of LLMs into two stages: first optimizing and objective consisting of implicit reward and reverse KL divergence and then improving the preference probability of teacher outputs over student outputs. We conducted experiments and analysis on various datasets with LLM parameters ranging from 120M to 13B and demonstrate the broad applicability and effectiveness of our DPKD approach. Meanwhile, we prove the value and effectiveness of the introduced implicit reward and output preference in KD through experiments and theoretical analysis. The DPKD method outperforms the baseline method in both output response precision and exact match percentage. Code and data are available at https://aka.ms/dpkd.
URLs: https://aka.ms/dpkd.
Authors: Oleg Vasilyev, Randy Sawaya, John Bohannon
Abstract: A query encoder of a dual passage retrieval system can be tuned for specific types of queries or domains, while the precomputed and stored documents representations are kept intact. Switching from one query encoder to another when needed is easily feasible, unlike overhauling the embeddings of a whole knowledge base. In this work we raise a question: Can the generic, original qualities of the encoder be preserved or at least left not too degraded when it is tuned on a narrow domain? We conducted experiments on a high quality multilingual embedding model: Tuning it on a single English-only dataset, we observe that the tuning not only preserves the multilingual qualities, but even improves them. The embedding qualities on distinctly different data are also improved or at least preserved. Drawing on our observations, we suggest a more general hypothesis: Tuning with intentionally low learning rate can preserve or improve a system's properties acquired in training, but not specifically targeted by tuning. We call this adiabatic tuning and provide tentative explanations.
Authors: Bruce W. Lee, Yeongheon Lee, Hyunsoo Cho
Abstract: Large Language Models (LLMs) exhibit non-deterministic behavior, and prompting has emerged as a primary method for steering their outputs toward desired directions. One popular strategy involves assigning a specific "persona" to the model to induce more varied and context-sensitive responses, akin to the diversity found in human perspectives. However, contrary to the expectation that persona-based prompting would yield a wide range of opinions, our experiments demonstrate that LLMs maintain consistent value orientations. In particular, we observe a persistent inertia in their responses, where certain moral and value dimensions, especially harm avoidance and fairness, remain distinctly skewed in one direction despite varied persona settings. To investigate this phenomenon systematically, use role-play at scale, which combines randomized, diverse persona prompts with a macroscopic trend analysis of model outputs. Our findings highlight the strong internal biases and value preferences in LLMs, underscoring the need for careful scrutiny and potential adjustment of these models to ensure balanced and equitable applications.
Authors: Taiwei Shi, Zhuoer Wang, Longqi Yang, Ying-Chun Lin, Zexue He, Mengting Wan, Pei Zhou, Sujay Jauhar, Sihao Chen, Shan Xia, Hongfei Zhang, Jieyu Zhao, Xiaofeng Xu, Xia Song, Jennifer Neville
Abstract: As large language models (LLMs) continue to advance, aligning these models with human preferences has emerged as a critical challenge. Traditional alignment methods, relying on human or LLM annotated datasets, are limited by their resource-intensive nature, inherent subjectivity, misalignment with real-world user preferences, and the risk of feedback loops that amplify model biases. To overcome these limitations, we introduce WildFeedback, a novel framework that leverages in-situ user feedback during conversations with LLMs to create preference datasets automatically. Given a corpus of multi-turn user-LLM conversation, WildFeedback identifies and classifies user feedback to LLM responses between conversation turns. The user feedback is then used to create examples of preferred and dispreferred responses according to users' preference. Our experiments demonstrate that LLMs fine-tuned on WildFeedback dataset exhibit significantly improved alignment with user preferences, as evidenced by both traditional benchmarks and our proposed checklist-guided evaluation. By incorporating in-situ feedback from actual users, WildFeedback addresses the scalability, subjectivity, and bias challenges that plague existing approaches, marking a significant step toward developing LLMs that are more responsive to the diverse and evolving needs of their users.
Authors: Menglin Liu, Ge Shi
Abstract: Large language models (LLMs) offer substantial promise for text classification in political science, yet their effectiveness often depends on high-quality prompts and exemplars. To address this, we introduce a three-stage framework that enhances LLM performance through automatic prompt optimization, dynamic exemplar selection, and a consensus mechanism. Our approach automates prompt refinement using task-specific exemplars, eliminating speculative trial-and-error adjustments and producing structured prompts aligned with human-defined criteria. In the second stage, we dynamically select the most relevant exemplars, ensuring contextually appropriate guidance for each query. Finally, our consensus mechanism mimics the role of multiple human coders for a single task, combining outputs from LLMs to achieve high reliability and consistency at a reduced cost. Evaluated across tasks including sentiment analysis, stance detection, and campaign ad tone classification, our method enhances classification accuracy without requiring task-specific model retraining or extensive manual adjustments to prompts. This framework not only boosts accuracy, interpretability and transparency but also provides a cost-effective, scalable solution tailored to political science applications. An open-source Python package (PoliPrompt) is available on GitHub.
Authors: Oghenefejiro Isaacs Anigboro, Charlie M. Crawford, Grace Proebsting, Dana\"e Metaxa, Sorelle A. Friedler
Abstract: Automated content moderation has long been used to help identify and filter undesired user-generated content online. Generative AI systems now use such filters to keep undesired generated content from being created by or shown to users. From classrooms to Hollywood, as generative AI is increasingly used for creative or expressive text generation, whose stories will these technologies allow to be told, and whose will they suppress? In this paper, we define and introduce measures of speech suppression, focusing on speech related to different identity groups incorrectly filtered by a range of content moderation APIs. Using both short-form, user-generated datasets traditional in content moderation and longer generative AI-focused data, including two datasets we introduce in this work, we create a benchmark for measurement of speech suppression for nine identity groups. Across one traditional and four generative AI-focused automated content moderation services tested, we find that identity-related speech is more likely to be incorrectly suppressed than other speech. We find differences in identity-related speech suppression for traditional versus generative AI data, with APIs performing better on generative AI data but worse on longer text instances, and by identity, with identity-specific reasons for incorrect flagging behavior. Overall, we find that on traditional short-form data incorrectly suppressed speech is likely to be political, while for generative AI creative data it is likely to be television violence.
Authors: Daehwan Nam, Gary Geunbae Lee
Abstract: Semantic parsers convert natural language to logical forms, which can be evaluated on knowledge bases (KBs) to produce denotations. Recent semantic parsers have been developed with sequence-to-sequence (seq2seq) pre-trained language models (PLMs) or large language models, where the models treat logical forms as sequences of tokens. For syntactic and semantic validity, the semantic parsers use grammars that enable constrained decoding. However, the grammars lack the ability to utilize large information of KBs, although logical forms contain representations of KB elements, such as entities or relations. In this work, we propose a grammar augmented with candidate expressions for semantic parsing on a large KB with a seq2seq PLM. The grammar defines actions as production rules, and our semantic parser predicts actions during inference under the constraints by types and candidate expressions. We apply the grammar to knowledge base question answering, where the constraints by candidate expressions assist a semantic parser to generate valid KB elements. We also introduce two special rules, sub-type inference and union types, and a mask caching algorithm. In particular, sub-type inference and the mask caching algorithm greatly increase the decoding speed of our semantic parser. We experimented on two benchmarks, KQA Pro and Overnight, where the constraints by candidate expressions increased the accuracy of our semantic parser, whether it was trained with strong supervision or weak supervision. In addition, our semantic parser had a fast decoding speed in the experiments.
Authors: Qiyuan Zhang, Yufei Wang, Tiezheng YU, Yuxin Jiang, Chuhan Wu, Liangyou Li, Yasheng Wang, Xin Jiang, Lifeng Shang, Ruiming Tang, Fuyuan Lyu, Chen Ma
Abstract: With significant efforts in recent studies, LLM-as-a-Judge has become a cost-effective alternative to human evaluation for assessing text generation quality in a wide range of tasks. However, there still remains a reliability gap between LLM-as-a-Judge and human evaluation. One important reason is the lack of guided oracles in the evaluation process. Motivated by the role of reference pervasively used in classic text evaluation, we introduce RevisEval, a novel text generation evaluation paradigm via the response-adapted references. RevisEval is driven by the key observation that an ideal reference should maintain the necessary relevance to the response to be evaluated. Specifically, RevisEval leverages the text revision capabilities of large language models (LLMs) to adaptively revise the response, then treat the revised text as the reference (response-adapted reference) for the subsequent evaluation. Extensive experiments demonstrate that RevisEval outperforms traditional reference-free and reference-based evaluation paradigms that use LLM-as-a-Judge across NLG tasks and open-ended instruction-following tasks. More importantly, our response-adapted references can further boost the classical text metrics, e.g., BLEU and BERTScore, compared to traditional references and even rival the LLM-as-a-Judge. A detailed analysis is also conducted to confirm RevisEval's effectiveness in bias reduction, the impact of inference cost, and reference relevance.
Authors: Tianzhu Ye, Li Dong, Yuqing Xia, Yutao Sun, Yi Zhu, Gao Huang, Furu Wei
Abstract: Transformer tends to overallocate attention to irrelevant context. In this work, we introduce Diff Transformer, which amplifies attention to the relevant context while canceling noise. Specifically, the differential attention mechanism calculates attention scores as the difference between two separate softmax attention maps. The subtraction cancels noise, promoting the emergence of sparse attention patterns. Experimental results on language modeling show that Diff Transformer outperforms Transformer in various settings of scaling up model size and training tokens. More intriguingly, it offers notable advantages in practical applications, such as long-context modeling, key information retrieval, hallucination mitigation, in-context learning, and reduction of activation outliers. By being less distracted by irrelevant context, Diff Transformer can mitigate hallucination in question answering and text summarization. For in-context learning, Diff Transformer not only enhances accuracy but is also more robust to order permutation, which was considered as a chronic robustness issue. The results position Diff Transformer as a highly effective and promising architecture to advance large language models.
Authors: Michael C. Stern, Jason A. Shaw
Abstract: We investigate the dynamics of labial constriction trajectories during the production of /b/ and /m/ in English and Mandarin. We find that, across languages and contexts, the ratio of instantaneous displacement to instantaneous velocity generally follows an exponential decay curve from movement onset to movement offset. We formalize this empirical discovery in a differential equation and, in combination with an assumption of point attractor dynamics, derive a nonlinear second-order dynamical system describing labial constriction trajectories. The equation has only two parameters, T and r. T corresponds to the target state and r corresponds to movement rapidity. Thus, each of the parameters corresponds to a phonetically relevant dimension of control. Nonlinear regression demonstrates that the model provides excellent fits to individual movement trajectories. Moreover, trajectories simulated from the model qualitatively match empirical trajectories, and capture key kinematic variables like duration, peak velocity, and time to achieve peak velocity. The model constitutes a proposal for the dynamics of individual articulatory movements, and thus offers a novel foundation from which to understand additional influences on articulatory kinematics like prosody, inter-movement coordination, and stochastic noise.
Authors: Ruobing Wang, Qingfei Zhao, Yukun Yan, Daren Zha, Yuxuan Chen, Shi Yu, Zhenghao Liu, Yixuan Wang, Shuo Wang, Xu Han, Zhiyuan Liu, Maosong Sun
Abstract: Retrieval-Augmented Generation (RAG) mitigates factual errors and hallucinations in Large Language Models (LLMs) for question-answering (QA) by incorporating external knowledge. However, existing adaptive RAG methods rely on LLMs to predict retrieval timing and directly use retrieved information for generation, often failing to reflect real information needs and fully leverage retrieved knowledge. We develop DeepNote, an adaptive RAG framework that achieves in-depth and robust exploration of knowledge sources through note-centric adaptive retrieval. DeepNote employs notes as carriers for refining and accumulating knowledge. During in-depth exploration, it uses these notes to determine retrieval timing, formulate retrieval queries, and iteratively assess knowledge growth, ultimately leveraging the best note for answer generation. Extensive experiments and analyses demonstrate that DeepNote significantly outperforms all baselines (+10.2% to +20.1%) and exhibits the ability to gather knowledge with both high density and quality. Additionally, DPO further improves the performance of DeepNote. The code and data are available at https://github.com/thunlp/DeepNote.
Authors: Shwai He, Tao Ge, Guoheng Sun, Bowei Tian, Xiaoyang Wang, Dong Yu
Abstract: Traditional transformer models often allocate a fixed amount of computational resources to every input token, leading to inefficient and unnecessary computation. To address this, the Mixture of Depths (MoD) was introduced to dynamically adjust the computational depth by skipping less important layers. Despite its promise, current MoD approaches remain under-explored and face two main challenges: (1) high training costs due to the need to train the entire model along with the routers that determine which layers to skip, and (2) the risk of performance degradation when important layers are bypassed. In response to the first issue, we propose Router-Tuning, a method that fine-tunes only the router on a small dataset, drastically reducing the computational overhead associated with full model training. For the second challenge, we propose MindSkip, which deploys Attention with Dynamic Depths. This method preserves the model's performance while significantly enhancing computational and memory efficiency. Extensive experiments demonstrate that our approach delivers competitive results while dramatically improving the computation efficiency, e.g., 21\% speedup and only a 0.2\% performance drop. The code is released at https://github.com/CASE-Lab-UMD/Router-Tuning.
Authors: Xiaocheng Zhang, Xi Wang, Yifei Lu, Zhuangzhuang Ye, Jianing Wang, Mengjiao Bao, Peng Yan, Xiaohong Su
Abstract: While fact verification remains fundamental, explanation generation serves as a critical enabler for trustworthy fact-checking systems by producing interpretable rationales and facilitating comprehensive verification processes. However, current benchmarks exhibit critical limitations in three dimensions: (1) absence of explanatory annotations, (2) English-centric language bias, and (3) inadequate temporal relevance. To bridge these gaps, we present TrendFact, the first Chinese fact-checking benchmark incorporating structured natural language explanations. TrendFact comprises 7,643 carefully curated samples from trending social media content and professional fact-checking repositories, covering domains such as public health, political discourse, and economic claims. It supports various forms of reasoning, including numerical computation, logical reasoning, and common sense verification. The rigorous multistage construction process ensures high data quality and provides significant challenges. Furthermore, we propose the ECS to complement existing evaluation metrics. To establish effective baselines for TrendFact, we propose FactISR, a dual-component method integrating evidence triangulation and iterative self-reflection mechanism. Experimental results demonstrate that current leading reasoning models (e.g., DeepSeek-R1, o1) have significant limitations on TrendFact, underscoring the real-world challenges it presents. FactISR significantly enhances reasoning model performance, offering new insights for explainable and complex fact-checking.
Authors: A M Muntasir Rahman, Junyi Ye, Wei Yao, Sierra S. Liu, Jesse Yu, Jonathan Yu, Wenpeng Yin, Guiling Wang
Abstract: Consider the math problem: "Lily received 3 cookies from her best friend yesterday and ate 5 for breakfast. Today, her friend gave her 3 more cookies. How many cookies does Lily have now?" Many large language models (LLMs) in previous research approach this problem by calculating the answer "1" using the equation "3 - 5 + 3." However, from a human perspective, we recognize the inherent flaw in this problem: Lily cannot eat 5 cookies if she initially only had 3. This discrepancy prompts a key question: Are current LLMs merely Blind Solver that apply mathematical operations without deeper reasoning, or can they function as Logical Thinker capable of identifying logical inconsistencies? To explore this question, we propose a benchmark dataset, FaultyMath, which includes faulty math problems of rich diversity: i) multiple mathematical categories, e.g., algebra, geometry, number theory, etc., ii) varying levels of difficulty, and iii) different origins of faultiness -- ranging from violations of common sense and ambiguous statements to mathematical contradictions and more. We evaluate a broad spectrum of LLMs, including open-source, closed-source, and math-specialized models, using FaultyMath across three dimensions: (i) How accurately can the models detect faulty math problems without being explicitly prompted to do so? (ii) When provided with hints -- either correct or misleading -- about the validity of the problems, to what extent do LLMs adapt to become reliable Logical Thinker? (iii) How trustworthy are the explanations generated by LLMs when they recognize a math problem as flawed? Through extensive experimentation and detailed analysis, our results demonstrate that existing LLMs largely function as Blind Solver and fall short of the reasoning capabilities required to perform as Logical Thinker.
Authors: Zhichao Wang, Bin Bi, Zixu Zhu, Xiangbo Mao, Jun Wang, Shiyu Wang
Abstract: By pretraining on trillions of tokens, an LLM gains the capability of text generation. However, to enhance its utility and reduce potential harm, SFT and alignment are applied sequentially to the pretrained model. Due to the differing nature and objective functions of SFT and alignment, catastrophic forgetting has become a significant issue. To address this, we introduce Unified Fine-Tuning (UFT), which integrates SFT and alignment into a single training stage using the same objective and loss functions through an implicit reward function. Our experimental results demonstrate that UFT outperforms SFT on instruction-tuning data alone. Moreover, when combining instruction-tuning data with alignment data, UFT effectively prevents catastrophic forgetting across these two stages and shows a clear advantage over sequentially applying SFT and alignment. This is evident in the significant improvements observed in the \textbf{ifeval} task for instruction-following and the \textbf{truthful-qa} task for factuality. The proposed general fine-tuning framework UFT establishes an effective and efficient pretraining-UFT paradigm for LLM training.
Authors: Lizhe Fang, Yifei Wang, Zhaoyang Liu, Chenheng Zhang, Stefanie Jegelka, Jinyang Gao, Bolin Ding, Yisen Wang
Abstract: Handling long-context inputs is crucial for large language models (LLMs) in tasks such as extended conversations, document summarization, and many-shot in-context learning. While recent approaches have extended the context windows of LLMs and employed perplexity (PPL) as a standard evaluation metric, PPL has proven unreliable for assessing long-context capabilities. The underlying cause of this limitation has remained unclear. In this work, we provide a comprehensive explanation for this issue. We find that PPL overlooks key tokens, which are essential for long-context understanding, by averaging across all tokens and thereby obscuring the true performance of models in long-context scenarios. To address this, we propose \textbf{LongPPL}, a novel metric that focuses on key tokens by employing a long-short context contrastive method to identify them. Our experiments demonstrate that LongPPL strongly correlates with performance on various long-context benchmarks (e.g., Pearson correlation of -0.96), significantly outperforming traditional PPL in predictive accuracy. Additionally, we introduce \textbf{LongCE} (Long-context Cross-Entropy) loss, a re-weighting strategy for fine-tuning that prioritizes key tokens, leading to consistent improvements across diverse benchmarks. In summary, these contributions offer deeper insights into the limitations of PPL and present effective solutions for accurately evaluating and enhancing the long-context capabilities of LLMs. Code is available at https://github.com/PKU-ML/LongPPL.
Authors: Qingfei Zhao, Ruobing Wang, Yukuo Cen, Daren Zha, Shicheng Tan, Jie Tang
Abstract: Retrieval-Augmented Generation (RAG) has emerged as a reliable external knowledge augmentation technique to mitigate hallucination issues and parameterized knowledge limitations in Large Language Models (LLMs). Existing adaptive RAG (ARAG) systems excel at in-depth exploration within a single source but struggle to effectively and controllably explore different retrieval sources, as they fail to foresee their internal knowledge features. We develop a novel multi-source ARAG system, PrefRAG, which enhances RAG by enabling in-depth and controllable exploration of diverse retrieval sources through preference-driven adaptive retrieval and self-reflection. PrefRAG first fully explores controllable local sources in adaptive retrieval and supplements with the web when appropriate, ultimately selecting the optimal source for knowledge observation. Subsequently, PrefRAG feeds answer quality feedback into the retrieval process, optimizing it from the generation perspective to produce higher-quality responses. Extensive experiments confirm its superiority, high retrieval efficiency, and knowledge controllability. PrefRAG outperforms Vanilla RAG and the leading MS-ARAG by up to 25.6% and 13.9% respectively. Additionally, PrefRAG trained with DPO achieves higher performance. The code and data are available at https://github.com/QingFei1/PrefRAG.git.
Authors: Weiyun Wang, Zhe Chen, Wenhai Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Jinguo Zhu, Xizhou Zhu, Lewei Lu, Yu Qiao, Jifeng Dai
Abstract: Existing open-source multimodal large language models (MLLMs) generally follow a training process involving pre-training and supervised fine-tuning. However, these models suffer from distribution shifts, which limit their multimodal reasoning, particularly in the Chain-of-Thought (CoT) performance. To address this, we introduce a preference optimization (PO) process to enhance the multimodal reasoning capabilities of MLLMs. Specifically, (1) on the data side, we design an automated preference data construction pipeline to create MMPR, a high-quality, large-scale multimodal reasoning preference dataset; and (2) on the model side, we explore integrating PO with MLLMs, developing a simple yet effective method, termed Mixed Preference Optimization (MPO), which boosts multimodal CoT performance. Our approach enhances the multimodal reasoning abilities of both InternVL2-8B and InternVL2-76B. Notably, our model, InternVL2-8B-MPO, achieves an accuracy of 67.0 on MathVista, outperforming InternVL2-8B by 8.7 points and achieving performance comparable to the 10$\times$ larger InternVL2-76B. We hope this study could inspire further advancements in MLLMs. Code, data, and model are released.
Authors: Sarkar Snigdha Sarathi Das, Ryo Kamoi, Bo Pang, Yusen Zhang, Caiming Xiong, Rui Zhang
Abstract: The effectiveness of large language models (LLMs) is closely tied to the design of prompts, making prompt optimization essential for enhancing their performance across a wide range of tasks. Many existing approaches to automating prompt engineering rely exclusively on textual feedback, refining prompts based solely on inference errors identified by large, computationally expensive LLMs. Unfortunately, smaller models struggle to generate high-quality feedback, resulting in complete dependence on large LLM judgment. Moreover, these methods fail to leverage more direct and finer-grained information, such as gradients, due to operating purely in text space. To this end, we introduce GReaTer, a novel prompt optimization technique that directly incorporates gradient information over task-specific reasoning. By utilizing task loss gradients, GReaTer enables self-optimization of prompts for open-source, lightweight language models without the need for costly closed-source LLMs. This allows high-performance prompt optimization without dependence on massive LLMs, closing the gap between smaller models and the sophisticated reasoning often needed for prompt refinement. Extensive evaluations across diverse reasoning tasks including BBH, GSM8k, and FOLIO demonstrate that GReaTer consistently outperforms previous state-of-the-art prompt optimization methods, even those reliant on powerful LLMs. Additionally, GReaTer-optimized prompts frequently exhibit better transferability and, in some cases, boost task performance to levels comparable to or surpassing those achieved by larger language models, highlighting the effectiveness of prompt optimization guided by gradients over reasoning. Code of GReaTer is available at https://github.com/psunlpgroup/GreaTer.
Authors: Miles Q. Li, Benjamin C. M. Fung, Shih-Chia Huang
Abstract: Pretraining large language models is a complex endeavor influenced by multiple factors, including model architecture, data quality, training continuity, and hardware constraints. In this paper, we share insights gained from the experience of training DMaS-LLaMa-Lite, a fully open source, 1.7-billion-parameter, LLaMa-based model, on approximately 20 billion tokens of carefully curated data. We chronicle the full training trajectory, documenting how evolving validation loss levels and downstream benchmarks reflect transitions from incoherent text to fluent, contextually grounded output. Beyond pretraining, we extend our analysis to include a post-training phase focused on instruction tuning, where the model was refined to produce more contextually appropriate, user-aligned responses. We highlight practical considerations such as the importance of restoring optimizer states when resuming from checkpoints, and the impact of hardware changes on training stability and throughput. While qualitative evaluation provides an intuitive understanding of model improvements, our analysis extends to various performance benchmarks, demonstrating how high-quality data and thoughtful scaling enable competitive results with significantly fewer training tokens. By detailing these experiences and offering training logs, checkpoints, and sample outputs, we aim to guide future researchers and practitioners in refining their pretraining strategies. The training script is available on Github at https://github.com/McGill-DMaS/DMaS-LLaMa-Lite-Training-Code. The model checkpoints are available on Huggingface at https://huggingface.co/collections/McGill-DMaS/dmas-llama-lite-6761d97ba903f82341954ceb.
URLs: https://github.com/McGill-DMaS/DMaS-LLaMa-Lite-Training-Code., https://huggingface.co/collections/McGill-DMaS/dmas-llama-lite-6761d97ba903f82341954ceb.
Authors: Ziming Guo, Chao Ma, Yinggang Sun, Tiancheng Zhao, Guangyao Wang, Hai Huang
Abstract: Recent advancements in large language models (LLMs) have significantly advanced text-to-SQL systems. However, most LLM-based methods often narrowly focus on SQL generation, neglecting the complexities of real-world conversational queries. This oversight can lead to unreliable responses, particularly for ambiguous questions that cannot be directly addressed with SQL. To bridge this gap, we propose MMSQL, a comprehensive test suite designed to evaluate the question classification and SQL generation capabilities of LLMs by simulating real-world scenarios with diverse question types and multi-turn Q&A interactions. Using MMSQL, we assessed the performance of popular LLMs, including both open-source and closed-source models, and identified key factors impacting their performance in such scenarios. Moreover, we introduce an LLM-based multi-agent framework that employs specialized agents to identify question types and determine appropriate answering strategies. Our experiments demonstrate that this approach significantly enhances the model's ability to navigate the complexities of conversational dynamics, effectively handling the diverse and complex nature of user queries. Our dataset and code are publicly available at https://mcxiaoxiao.github.io/MMSQL.
Authors: Yanlin Feng, Simone Papicchio, Sajjadur Rahman
Abstract: Retrieval from graph data is crucial for augmenting large language models (LLM) with both open-domain knowledge and private enterprise data, and it is also a key component in the recent GraphRAG system (edge et al., 2024). Despite decades of research on knowledge graphs and knowledge base question answering, leading LLM frameworks (e.g. Langchain and LlamaIndex) have only minimal support for retrieval from modern encyclopedic knowledge graphs like Wikidata. In this paper, we analyze the root cause and suggest that modern RDF knowledge graphs (e.g. Wikidata, Freebase) are less efficient for LLMs due to overly large schemas that far exceed the typical LLM context window, use of resource identifiers, overlapping relation types and lack of normalization. As a solution, we propose property graph views on top of the underlying RDF graph that can be efficiently queried by LLMs using Cypher. We instantiated this idea on Wikidata and introduced CypherBench, the first benchmark with 11 large-scale, multi-domain property graphs with 7.8 million entities and over 10,000 questions. To achieve this, we tackled several key challenges, including developing an RDF-to-property graph conversion engine, creating a systematic pipeline for text-to-Cypher task generation, and designing new evaluation metrics.
Authors: Kedi Chen, Qin Chen, Jie Zhou, Xinqi Tao, Bowen Ding, Jingwen Xie, Mingchen Xie, Peilong Li, Feng Zheng, Liang He
Abstract: Large Language Models (LLMs) are prone to hallucination with non-factual or unfaithful statements, which undermines the applications in real-world scenarios. Recent researches focus on uncertainty-based hallucination detection, which utilizes the output probability of LLMs for uncertainty calculation and does not rely on external knowledge or frequent sampling from LLMs. Whereas, most approaches merely consider the uncertainty of each independent token, while the intricate semantic relations among tokens and sentences are not well studied, which limits the detection of hallucination that spans over multiple tokens and sentences in the passage. In this paper, we propose a method to enhance uncertainty modeling with semantic graph for hallucination detection. Specifically, we first construct a semantic graph that well captures the relations among entity tokens and sentences. Then, we incorporate the relations between two entities for uncertainty propagation to enhance sentence-level hallucination detection. Given that hallucination occurs due to the conflict between sentences, we further present a graph-based uncertainty calibration method that integrates the contradiction probability of the sentence with its neighbors in the semantic graph for uncertainty calculation. Extensive experiments on two datasets show the great advantages of our proposed approach. In particular, we obtain substantial improvements with 19.78% in passage-level hallucination detection.
Authors: Jian Hu, Jason Klein Liu, Wei Shen
Abstract: Reinforcement Learning from Human Feedback (RLHF) plays a crucial role in aligning large language models (LLMs) with human values and preferences. While state-of-the-art applications like ChatGPT/GPT-4 commonly employ Proximal Policy Optimization (PPO), the inclusion of a critic network introduces significant computational overhead. REINFORCE-based methods, such as REINFORCE Leave One-Out (RLOO), ReMax, and Group Relative Policy Optimization (GRPO), address this limitation by eliminating the critic network. However, these approaches face challenges in accurate advantage estimation. Specifically, they estimate advantages independently for responses to each prompt, which can lead to overfitting on simpler prompts and vulnerability to reward hacking. To address these challenges, we introduce REINFORCE++, a novel approach that removes the critic model while using the normalized reward of a batch as the baseline. Our empirical evaluation demonstrates that REINFORCE++ exhibits robust performance across various reward models without requiring prompt set truncation. Furthermore, it achieves superior generalization in both RLHF and long chain-of-thought (CoT) settings compared to existing REINFORCE-based methods. The implementation is available at https://github.com/OpenRLHF/OpenRLHF.
Authors: Yue Dai, Soyeon Caren Han, Wei Liu
Abstract: Chart question answering (ChartQA) is challenged by the heterogeneous composition of chart elements and the subtle data patterns they encode. This work introduces a novel joint multimodal scene graph framework that explicitly models the relationships among chart components and their underlying structures. The framework integrates both visual and textual graphs to capture structural and semantic characteristics, while a graph contrastive learning strategy aligns node representations across modalities enabling their seamless incorporation into a transformer decoder as soft prompts. Moreover, a set of tailored Chain of Thought (CoT) prompts is proposed to enhance multimodal large language models (MLLMs) in zero-s ot scenarios by mitigating hallucinations. Extensive evaluations on benchmarks including ChartQA, OpenCQA, and ChartX demonstrate significant performance improvements and validate the efficacy of the proposed approach.
Authors: Wenhao Zheng, Yixiao Chen, Weitong Zhang, Souvik Kundu, Yun Li, Zhengzhong Liu, Eric P. Xing, Hongyi Wang, Huaxiu Yao
Abstract: Large language models have achieved remarkable success in various tasks but suffer from high computational costs during inference, limiting their deployment in resource-constrained applications. To address this issue, we propose a novel Collaborative Inference with Token-lEvel Routing (CITER) framework that enables efficient collaboration between small and large language models (SLMs \& LLMs) through a token-level routing strategy. Specifically, CITER routes non-critical tokens to an SLM for efficiency and routes critical tokens to an LLM for generalization quality. We formulate router training as a policy optimization, where the router receives rewards based on both the quality of predictions and the inference costs of generation. This allows the router to learn to predict token-level routing scores and make routing decisions based on both the current token and the future impact of its decisions. To further accelerate the reward evaluation process, we introduce a shortcut which significantly reduces the costs of the reward estimation and improving the practicality of our approach. Extensive experiments on five benchmark datasets demonstrate that CITER reduces the inference costs while preserving high-quality generation, offering a promising solution for real-time and resource-constrained applications. Our data and code are available at https://github.com/aiming-lab/CITER.
Authors: V\'ictor Gallego
Abstract: We propose a novel dynamic safety framework that optimizes language model (LM) safety reasoning at inference time without modifying model weights. Building on recent advances in self-critique methods, our approach leverages a meta-critique mechanism that iteratively updates safety prompts-termed specifications-to drive the critique and revision process adaptively. This test-time optimization not only improves performance against adversarial jailbreak requests but also in diverse general safety-related tasks, such as avoiding moral harm or pursuing honest responses. Our empirical evaluations across several language models demonstrate that dynamically optimized safety prompts yield significantly higher safety scores compared to fixed system prompts and static self-critique defenses. Code released at https://github.com/vicgalle/meta-self-critique.git .
Authors: Hyundong Cho, Karishma Sharma, Nicolaas Jedema, Leonardo F. R. Ribeiro, Alessandro Moschitti, Ravi Krishnan, Jonathan May
Abstract: Language models are aligned to the collective voice of many, resulting in generic outputs that do not align with specific users' styles. In this work, we present Trial-Error-Explain In-Context Learning (TICL), a tuning-free method that personalizes language models for text generation tasks with fewer than 10 examples per user. TICL iteratively expands an in-context learning prompt via a trial-error-explain process, adding model-generated negative samples and explanations that provide fine-grained guidance towards a specific user's style. TICL achieves favorable win rates on pairwise comparisons with LLM-as-a-judge up to 91.5% against the previous state-of-the-art and outperforms competitive tuning-free baselines for personalized alignment tasks of writing emails, essays and news articles. Both lexical and qualitative analyses show that the negative samples and explanations enable language models to learn stylistic context more effectively and overcome the bias towards structural and formal phrases observed in their zero-shot outputs. By front-loading inference compute to create a user-specific in-context learning prompt that does not require extra generation steps at test time, TICL presents a novel yet simple approach for personalized alignment.
Authors: Qiyuan Zhang, Yufei Wang, Yuxin Jiang, Liangyou Li, Chuhan Wu, Yasheng Wang, Xin Jiang, Lifeng Shang, Ruiming Tang, Fuyuan Lyu, Chen Ma
Abstract: LLM-as-a-Judge, which generates chain-of-thought (CoT) judgments, has become a widely adopted auto-evaluation method. However, its reliability is compromised by the CoT reasoning's inability to capture comprehensive and deeper details, often leading to incomplete outcomes. Existing methods mainly rely on majority voting or criteria expansion, which is insufficient to address the limitation in CoT. We propose Crowd-based Comparative Evaluation, which introduces additional crowd responses to compare with the candidate responses, thereby exposing deeper and more comprehensive details within the candidate responses. This process effectively guides LLM-as-a-Judge to provide a more detailed CoT judgment. Extensive experiments demonstrate that our approach enhances evaluation reliability, achieving an average accuracy gain of 6.7% across five benchmarks. Moreover, our method produces higher-quality CoTs that facilitate judge distillation and exhibit superior performance in rejection sampling for supervised fine-tuning (SFT), referred to as crowd rejection sampling, thereby enabling more efficient SFT. Our analysis confirms that CoTs generated by ours are more comprehensive and of higher quality, and evaluation accuracy improves as inference scales.
Authors: Sean Wang, Yicheng Jiang, Yuxin Tang, Lu Cheng, Hanjie Chen
Abstract: Uncertainty Quantification (UQ) for Natural Language Generation (NLG) is crucial for assessing the performance of Large Language Models (LLMs), as it reveals confidence in predictions, identifies failure modes, and gauges output reliability. Conformal Prediction (CP), a model-agnostic method that generates prediction sets with a specified error rate, has been adopted for UQ in classification tasks, where the size of the prediction set indicates the model's uncertainty. However, when adapting CP to NLG, the sampling-based method for generating candidate outputs cannot guarantee the inclusion of the ground truth, limiting its applicability across a wide range of error rates. To address this, we propose \ourmethod, a method that explicitly adds the ground truth to the candidate outputs and uses logit scores to measure nonconformity. Our experiments with six LLMs on four NLG tasks show that \ourmethod outperforms baseline methods in calibrating error rates and empirical cover rates, offering accurate UQ across a wide range of user-specified error rates.
Authors: Arefeh Kazemi, Sri Balaaji Natarajan Kalaivendan, Joachim Wagner, Hamza Qadeer, Brian Davis
Abstract: Cyberbullying (CB) presents a pressing threat, especially to children, underscoring the urgent need for robust detection systems to ensure online safety. However, progress in developing such systems is hindered by the scarcity of large, labeled datasets that are specifically tailored for specialized tasks and the target age groups. Creating these datasets relies heavily on human annotation, which not only strains resources but also raises significant ethical and legal concerns due to annotators' exposure to harmful content, notwithstanding the acquisition of this type of data from vulnerable populations such as children. In this paper, we address these challenges by leveraging Large Language Models (LLMs) to generate synthetic data and labels. Our experiments demonstrate that synthetic data enables BERT-based CB classifiers to achieve performance close to that of those trained on fully authentic datasets (75.8% vs. 81.5% accuracy). Additionally, LLMs can effectively label authentic yet unlabeled data, allowing BERT classifiers to attain a comparable performance level (79.1% vs. 81.5% accuracy). These results highlight the potential of LLMs as a scalable, ethical, and cost-effective solution for generating data for CB detection.
Authors: Bojana Ba\v{s}aragin, Darija Medvecki, Gorana Goji\'c, Milena Oparnica, Dragi\v{s}a Mi\v{s}kovi\'c
Abstract: This study introduces a novel natural language processing pipeline that enhances customer service efficiency at Telekom Srbija, a leading Serbian telecommunications company, through automated email topic detection and labeling. Central to the pipeline is BERTopic, a modular framework that allows unsupervised topic modeling. After a series of preprocessing and postprocessing steps, we assign one of 12 topics and several additional labels to incoming emails, allowing the customer service to filter and access them through a custom-made application. The model's performance was evaluated by assessing the speed and correctness of the automatically assigned topics, with a weighted average processing time of 0.041 seconds per email and a weighted average F1 score of 0.96. The pipeline shows broad applicability across languages, particularly to those that are low-resourced and morphologically rich. The system now operates in the company's production environment, streamlining customer service operations through automated email classification.
Authors: Feng Gu, Zongxia Li, Carlos Rafael Colon, Benjamin Evans, Ishani Mondal, Jordan Lee Boyd-Graber
Abstract: Event annotation is important for identifying market changes, monitoring breaking news, and understanding sociological trends. Although expert annotators set the gold standards, human coding is expensive and inefficient. Unlike information extraction experiments that focus on single contexts, we evaluate a holistic workflow that removes irrelevant documents, merges documents about the same event, and annotates the events. Although LLM-based automated annotations are better than traditional TF-IDF-based methods or Event Set Curation, they are still not reliable annotators compared to human experts. However, adding LLMs to assist experts for Event Set Curation can reduce the time and mental effort required for Variable Annotation. When using LLMs to extract event variables to assist expert annotators, they agree more with the extracted variables than fully automated LLMs for annotation.
Authors: Xin Liu, Pei Liu, Guoming Tang
Abstract: The linear growth of key-value (KV) cache memory and quadratic computational in attention mechanisms complexity pose significant bottlenecks for large language models (LLMs) in long-context processing. While existing KV cache optimization methods address these challenges through token pruning or feature merging, they often incur irreversible information loss or require costly parameter retraining. To this end, we propose ZSMerge, a dynamic KV cache compression framework designed for efficient cache management, featuring three key operations: (1) fine-grained memory allocation guided by multi-dimensional token importance metrics at head-level granularity, (2) a residual merging mechanism that preserves critical context through compensated attention scoring, and (3) a zero-shot adaptation mechanism compatible with diverse LLM architectures without requiring retraining. ZSMerge significantly enhances memory efficiency and inference speed with negligible performance degradation across LLMs. When applied to LLaMA2-7B, it demonstrates a 20:1 compression ratio for key-value cache retention (reducing memory footprint to 5\% of baseline) while sustaining comparable generation quality, coupled with triple throughput gains at extreme 54k-token contexts that eliminate out-of-memory failures. The code is available at https://github.com/SusCom-Lab/ZSMerge.
Authors: Hai Zhao, Hongqiu Wu, Dongjie Yang, Anni Zou, Jiale Hong
Abstract: This paper reports the first brain-inspired large language model (BriLLM). This is a non-Transformer, non-GPT, non-traditional machine learning input-output controlled generative language model. The model is based on the Signal Fully-connected flowing (SiFu) definition on the directed graph in terms of the neural network, and has the interpretability of all nodes on the graph of the whole model, instead of the traditional machine learning model that only has limited interpretability at the input and output ends. In the language model scenario, the token is defined as a node in the graph. A randomly shaped or user-defined signal flow flows between nodes on the principle of "least resistance" along paths. The next token or node to be predicted or generated is the target of the signal flow. As a language model, BriLLM theoretically supports infinitely long $n$-gram models when the model size is independent of the input and predicted length of the model. The model's working signal flow provides the possibility of recall activation and innate multi-modal support similar to the cognitive patterns of the human brain. At present, we released the first BriLLM version in Chinese, with 4000 tokens, 32-dimensional node width, 16-token long sequence prediction ability, and language model prediction performance comparable to GPT-1. More computing power will help us explore the infinite possibilities depicted above.
Authors: Pritam Kadasi, Sriman Reddy Kondam, Srivathsa Vamsi Chaturvedula, Rudranshu Sen, Agnish Saha, Soumavo Sikdar, Sayani Sarkar, Suhani Mittal, Rohit Jindal, Mayank Singh
Abstract: With the massive surge in ML models on platforms like Hugging Face, users often lose track and struggle to choose the best model for their downstream tasks, frequently relying on model popularity indicated by download counts, likes, or recency. We investigate whether this popularity aligns with actual model performance and how the comprehensiveness of model documentation correlates with both popularity and performance. In our study, we evaluated a comprehensive set of 500 Sentiment Analysis models on Hugging Face. This evaluation involved massive annotation efforts, with human annotators completing nearly 80,000 annotations, alongside extensive model training and evaluation. Our findings reveal that model popularity does not necessarily correlate with performance. Additionally, we identify critical inconsistencies in model card reporting: approximately 80% of the models analyzed lack detailed information about the model, training, and evaluation processes. Furthermore, about 88% of model authors overstate their models' performance in the model cards. Based on our findings, we provide a checklist of guidelines for users to choose good models for downstream tasks.
Authors: Sarah Pungitore, Shashank Yadav, Vignesh Subbian
Abstract: Computational phenotyping is essential for biomedical research but often requires significant time and resources, especially since traditional methods typically involve extensive manual data review. While machine learning and natural language processing advancements have helped, further improvements are needed. Few studies have explored using Large Language Models (LLMs) for these tasks despite known advantages of LLMs for text-based tasks. To facilitate further research in this area, we developed an evaluation framework, Evaluation of PHEnotyping for Observational Health Data (PHEONA), that outlines context-specific considerations. We applied and demonstrated PHEONA on concept classification, a specific task within a broader phenotyping process for Acute Respiratory Failure (ARF) respiratory support therapies. From the sample concepts tested, we achieved high classification accuracy, suggesting the potential for LLM-based methods to improve computational phenotyping processes.
Authors: Leonor Barreiros, Isabel Coutinho, Gon\c{c}alo M. Correia, Bruno Martins
Abstract: Clinical coding is a critical task in healthcare, although traditional methods for automating clinical coding may not provide sufficient explicit evidence for coders in production environments. This evidence is crucial, as medical coders have to make sure there exists at least one explicit passage in the input health record that justifies the attribution of a code. We therefore propose to reframe the task as an entity linking problem, in which each document is annotated with its set of codes and respective textual evidence, enabling better human-machine collaboration. By leveraging parameter-efficient fine-tuning of Large Language Models (LLMs), together with constrained decoding, we introduce three approaches to solve this problem that prove effective at disambiguating clinical mentions and that perform well in few-shot scenarios.
Authors: Yuxuan Lu, Jing Huang, Yan Han, Bennet Bei, Yaochen Xie, Dakuo Wang, Jessie Wang, Qi He
Abstract: Recent research shows that LLMs can simulate ``believable'' human behaviors to power LLM agents via prompt-only methods. In this work, we focus on evaluating and improving LLM's objective ``accuracy'' rather than the subjective ``believability'' in the web action generation task, leveraging a large-scale, real-world dataset collected from online shopping human actions. We present the first comprehensive quantitative evaluation of state-of-the-art LLMs (e.g., DeepSeek-R1, Llama, and Claude) on the task of web action generation. Our results show that fine-tuning LLMs on real-world behavioral data substantially improves their ability to generate actions compared to prompt-only methods. Furthermore, incorporating synthesized reasoning traces into model training leads to additional performance gains, demonstrating the value of explicit rationale in behavior modeling. This work establishes a new benchmark for evaluating LLMs in behavior simulation and offers actionable insights into how real-world action data and reasoning augmentation can enhance the fidelity of LLM agents.
Authors: Gus G. Xia
Abstract: This paper introduces function alignment, a novel theory of mind and intelligence that is both intuitively compelling and structurally grounded. It explicitly models how meaning, interpretation, and analogy emerge from interactions among layered representations, forming a coherent framework capable not only of modeling minds but also of serving as a blueprint for building them. One of the key theoretical insights derived from function alignment is bounded interpretability, which provides a unified explanation for previously fragmented ideas in cognitive science, such as bounded rationality, symbol grounding, and analogy-making. Beyond modeling, the function alignment framework bridges disciplines often kept apart, linking computational architecture, psychological theory, and even contemplative traditions such as Zen. Rather than building on any philosophical systems, it offers a structural foundation upon which multiple ways of understanding the mind may be reconstructed.
Authors: Ivo Petrov, Jasper Dekoninck, Lyuben Baltadzhiev, Maria Drencheva, Kristian Minchev, Mislav Balunovi\'c, Nikola Jovanovi\'c, Martin Vechev
Abstract: Recent math benchmarks for large language models (LLMs) such as MathArena indicate that state-of-the-art reasoning models achieve impressive performance on mathematical competitions like AIME, with the leading model, Gemini-2.5-Pro, achieving scores comparable to top human competitors. However, these benchmarks evaluate models solely based on final numerical answers, neglecting rigorous reasoning and proof generation which are essential for real-world mathematical tasks. To address this, we introduce the first comprehensive evaluation of full-solution reasoning for challenging mathematical problems. Using expert human annotators, we evaluated several state-of-the-art reasoning models on the six problems from the 2025 USAMO within hours of their release. Our results reveal that all tested models struggled significantly: only Gemini-2.5-Pro achieves a non-trivial score of 25%, while all other models achieve less than 5%. Through detailed analysis of reasoning traces, we identify the most common failure modes and find several unwanted artifacts arising from the optimization strategies employed during model training. Overall, our results suggest that current LLMs are inadequate for rigorous mathematical reasoning tasks, highlighting the need for substantial improvements in reasoning and proof generation capabilities.
Authors: Chung-En Sun, Ge Yan, Tsui-Wei Weng
Abstract: Recent studies have shown that Large Language Models (LLMs) augmented with chain-of-thought (CoT) reasoning demonstrate impressive problem-solving abilities. However, in this work, we identify a recurring issue where these models occasionally generate overly short reasoning, leading to degraded performance on even simple mathematical problems. Specifically, we investigate how reasoning length is embedded in the hidden representations of reasoning models and its impact on accuracy. Our analysis reveals that reasoning length is governed by a linear direction in the representation space, allowing us to induce overly short reasoning by steering the model along this direction. Building on this insight, we introduce ThinkEdit, a simple yet effective weight-editing approach to mitigate the issue of overly short reasoning. We first identify a small subset of attention heads (approximately 2%) that predominantly drive short reasoning behavior. We then edit the output projection weights of these heads to suppress the short reasoning direction. With changes to only 0.1% of the model's parameters, ThinkEdit effectively reduces overly short reasoning and yields notable accuracy gains for short reasoning outputs (+5.44%), along with an overall improvement across multiple math benchmarks (+2.43%). Our findings provide new mechanistic insights into how reasoning length is controlled within LLMs and highlight the potential of fine-grained model interventions to improve reasoning quality. Our code is available at https://github.com/Trustworthy-ML-Lab/ThinkEdit
Authors: Muhammad Ahmad, Humaira Farid, Iqra Ameer, Maaz Amjad, Muhammad Muzamil, Ameer Hamza, Muhammad Jalal, Ildar Batyrshin, Grigori Sidorov
Abstract: The opioid overdose epidemic remains a critical public health crisis, particularly in the United States, leading to significant mortality and societal costs. Social media platforms like Reddit provide vast amounts of unstructured data that offer insights into public perceptions, discussions, and experiences related to opioid use. This study leverages Natural Language Processing (NLP), specifically Opioid Named Entity Recognition (ONER-2025), to extract actionable information from these platforms. Our research makes four key contributions. First, we created a unique, manually annotated dataset sourced from Reddit, where users share self-reported experiences of opioid use via different administration routes. This dataset contains 331,285 tokens and includes eight major opioid entity categories. Second, we detail our annotation process and guidelines while discussing the challenges of labeling the ONER-2025 dataset. Third, we analyze key linguistic challenges, including slang, ambiguity, fragmented sentences, and emotionally charged language, in opioid discussions. Fourth, we propose a real-time monitoring system to process streaming data from social media, healthcare records, and emergency services to identify overdose events. Using 5-fold cross-validation in 11 experiments, our system integrates machine learning, deep learning, and transformer-based language models with advanced contextual embeddings to enhance understanding. Our transformer-based models (bert-base-NER and roberta-base) achieved 97% accuracy and F1-score, outperforming baselines by 10.23% (RF=0.88).
Authors: Xiaoxuan Zhu, Zhouhong Gu, Baiqian Wu, Suhang Zheng, Tao Wang, Tianyu Li, Hongwei Feng, Yanghua Xiao
Abstract: Pre-training large language models (LLMs) necessitates enormous diverse textual corpora, making effective data selection a key challenge for balancing computational resources and model performance. Current methodologies primarily emphasize data quality metrics and mixing proportions, yet they fail to adequately capture the underlying semantic connections between training samples and quality disparities within individual domains. We introduce ToReMi (Topic-based Reweighting for Model improvement), a novel two-stage framework that dynamically adjusts training sample weights according to their topical associations and observed learning patterns. Our comprehensive experiments reveal that ToReMi variants consistently achieve superior performance over conventional pre-training approaches, demonstrating accelerated perplexity reduction across multiple domains and enhanced capabilities on downstream evaluation tasks. Code is available at https://github.com/zxx000728/ToReMi.
Authors: Jian Zhao, Runze Liu, Kaiyan Zhang, Zhimu Zhou, Junqi Gao, Dong Li, Jiafei Lyu, Zhouyi Qian, Biqing Qi, Xiu Li, Bowen Zhou
Abstract: Recent advancements in Large Language Models (LLMs) have shown that it is promising to utilize Process Reward Models (PRMs) as verifiers to enhance the performance of LLMs. However, current PRMs face three key challenges: (1) limited process supervision and generalization capabilities, (2) dependence on scalar value prediction without leveraging the generative abilities of LLMs, and (3) inability to scale the test-time compute of PRMs. In this work, we introduce GenPRM, a generative process reward model that performs explicit Chain-of-Thought (CoT) reasoning with code verification before providing judgment for each reasoning step. To obtain high-quality process supervision labels and rationale data, we propose Relative Progress Estimation (RPE) and a rationale synthesis framework that incorporates code verification. Experimental results on ProcessBench and several mathematical reasoning tasks show that GenPRM significantly outperforms prior PRMs with only 23K training data from MATH dataset. Through test-time scaling, a 1.5B GenPRM outperforms GPT-4o, and a 7B GenPRM surpasses Qwen2.5-Math-PRM-72B on ProcessBench. Additionally, GenPRM demonstrates strong abilities to serve as a critic model for policy model refinement. This work establishes a new paradigm for process supervision that bridges the gap between PRMs and critic models in LLMs. Our code, model, and data will be available in https://ryanliu112.github.io/GenPRM.
Authors: Juncheng Wu, Wenlong Deng, Xingxuan Li, Sheng Liu, Taomian Mi, Yifan Peng, Ziyang Xu, Yi Liu, Hyunjin Cho, Chang-In Choi, Yihan Cao, Hui Ren, Xiang Li, Xiaoxiao Li, Yuyin Zhou
Abstract: Medical tasks such as diagnosis and treatment planning require precise and complex reasoning, particularly in life-critical domains. Unlike mathematical reasoning, medical reasoning demands meticulous, verifiable thought processes to ensure reliability and accuracy. However, there is a notable lack of datasets that provide transparent, step-by-step reasoning to validate and enhance the medical reasoning ability of AI models. To bridge this gap, we introduce MedReason, a large-scale high-quality medical reasoning dataset designed to enable faithful and explainable medical problem-solving in large language models (LLMs). We utilize a structured medical knowledge graph (KG) to convert clinical QA pairs into logical chains of reasoning, or ``thinking paths'', which trace connections from question elements to answers via relevant KG entities. Each path is validated for consistency with clinical logic and evidence-based medicine. Our pipeline generates detailed reasoning for various medical questions from 7 medical datasets, resulting in a dataset of 32,682 question-answer pairs, each with detailed, step-by-step explanations. Experiments demonstrate that fine-tuning with our dataset consistently boosts medical problem-solving capabilities, achieving significant gains of up to 7.7% for DeepSeek-Ditill-8B. Our top-performing model, MedReason-8B, outperforms the Huatuo-o1-8B, a state-of-the-art medical reasoning model, by up to 4.2% on the clinical benchmark MedBullets. We also engage medical professionals from diverse specialties to assess our dataset's quality, ensuring MedReason offers accurate and coherent medical reasoning. Our data, models, and code is available at https://github.com/UCSC-VLAA/MedReason.
Authors: Souradip Chakraborty, Mohammadreza Pourreza, Ruoxi Sun, Yiwen Song, Nino Scherrer, Furong Huang, Amrit Singh Bedi, Ahmad Beirami, Jindong Gu, Hamid Palangi, Tomas Pfister
Abstract: While AI agents have shown remarkable performance at various tasks, they still struggle with complex multi-modal applications, structured generation and strategic planning. Improvements via standard fine-tuning is often impractical, as solving agentic tasks usually relies on black box API access without control over model parameters. Inference-time methods such as Best-of-N (BON) sampling offer a simple yet effective alternative to improve performance. However, BON lacks iterative feedback integration mechanism. Hence, we propose Iterative Agent Decoding (IAD) which combines iterative refinement with dynamic candidate evaluation and selection guided by a verifier. IAD differs in how feedback is designed and integrated, specifically optimized to extract maximal signal from reward scores. We conduct a detailed comparison of baselines across key metrics on Sketch2Code, Text2SQL, and Webshop where IAD consistently outperforms baselines, achieving 3--6% absolute gains on Sketch2Code and Text2SQL (with and without LLM judges) and 8--10% gains on Webshop across multiple metrics. To better understand the source of IAD's gains, we perform controlled experiments to disentangle the effect of adaptive feedback from stochastic sampling, and find that IAD's improvements are primarily driven by verifier-guided refinement, not merely sampling diversity. We also show that both IAD and BON exhibit inference-time scaling with increased compute when guided by an optimal verifier. Our analysis highlights the critical role of verifier quality in effective inference-time optimization and examines the impact of noisy and sparse rewards on scaling behavior. Together, these findings offer key insights into the trade-offs and principles of effective inference-time optimization.
Authors: Zijun Liu, Peiyi Wang, Runxin Xu, Shirong Ma, Chong Ruan, Peng Li, Yang Liu, Yu Wu
Abstract: Reinforcement learning (RL) has been widely adopted in post-training for large language models (LLMs) at scale. Recently, the incentivization of reasoning capabilities in LLMs from RL indicates that $\textit{proper learning methods could enable effective inference-time scalability}$. A key challenge of RL is to obtain accurate reward signals for LLMs in various domains beyond verifiable questions or artificial rules. In this work, we investigate how to improve reward modeling (RM) with more inference compute for general queries, i.e. the $\textbf{inference-time scalability of generalist RM}$, and further, how to improve the effectiveness of performance-compute scaling with proper learning methods. For the RM approach, we adopt pointwise generative reward modeling (GRM) to enable flexibility for different input types and potential for inference-time scaling. For the learning method, we propose Self-Principled Critique Tuning (SPCT) to foster scalable reward generation behaviors in GRMs through online RL, to generate principles adaptively and critiques accurately, resulting in $\textbf{DeepSeek-GRM}$ models. Furthermore, for effective inference-time scaling, we use parallel sampling to expand compute usage, and introduce a meta RM to guide voting process for better scaling performance. Empirically, we show that SPCT significantly improves the quality and scalability of GRMs, outperforming existing methods and models in various RM benchmarks without severe biases, and could achieve better performance compared to training-time scaling. DeepSeek-GRM still meets challenges in some tasks, which we believe can be addressed by future efforts in generalist reward systems. The models will be released and open-sourced.
Authors: Siddharth Khincha, Tushar Kataria, Ankita Anand, Dan Roth, Vivek Gupta
Abstract: The vast amount of online information today poses challenges for non-English speakers, as much of it is concentrated in high-resource languages such as English and French. Wikipedia reflects this imbalance, with content in low-resource languages frequently outdated or incomplete. Recent research has sought to improve cross-language synchronization of Wikipedia tables using rule-based methods. These approaches can be effective, but they struggle with complexity and generalization. This paper explores large language models (LLMs) for multilingual information synchronization, using zero-shot prompting as a scalable solution. We introduce the Information Updation dataset, simulating the real-world process of updating outdated Wikipedia tables, and evaluate LLM performance. Our findings reveal that single-prompt approaches often produce suboptimal results, prompting us to introduce a task decomposition strategy that enhances coherence and accuracy. Our proposed method outperforms existing baselines, particularly in Information Updation (1.79%) and Information Addition (20.58%), highlighting the model strength in dynamically updating and enriching data across architectures.
Authors: Abhishek Sharma, Dan Goldwasser
Abstract: LLMs' remarkable ability to tackle a wide range of language tasks opened new opportunities for collaborative human-AI problem solving. LLMs can amplify human capabilities by applying their intuitions and reasoning strategies at scale. We explore whether human guides can be simulated, by generalizing from human demonstrations of guiding an AI system to solve complex language problems. We introduce CoLa, a novel self-guided learning paradigm for training automated $\textit{guides}$ and evaluate it on two QA datasets, a puzzle-solving task, and a constrained text generation task. Our empirical results show that CoLa consistently outperforms competitive approaches across all domains. Moreover, a small-sized trained guide outperforms a strong model like GPT-4 when acting as a guide. We compare the strategies employed by humans and automated guides by conducting a human study on a QA dataset. We show that automated guides outperform humans by adapting their strategies to reasoners' capabilities and conduct qualitative analyses highlighting distinct differences in guiding strategies.
Authors: Jaewoo Park, Jungyang Park, Dongju Jang, Jiwan Chung, Byungwoo Yoo, Jaewoo Shin, Seonjoon Park, Taehyeong Kim, Youngjae Yu
Abstract: With the rapid advancement of mathematical reasoning capabilities in Large Language Models (LLMs), AI systems are increasingly being adopted in educational settings to support students' comprehension of problem-solving processes. However, a critical component remains underexplored in current LLM-generated explanations: visual explanation. In real-world instructional contexts, human tutors routinely employ visual aids - such as diagrams, markings, and highlights - to enhance conceptual clarity. To bridge this gap, we introduce a novel task of visual solution explanation, which requires generating explanations that incorporate newly introduced visual elements essential for understanding (e.g., auxiliary lines, annotations, or geometric constructions). To evaluate model performance on this task, we propose MathExplain, a multimodal benchmark consisting of 997 math problems annotated with visual keypoints and corresponding explanatory text that references those elements. Our empirical results show that while some closed-source models demonstrate promising capabilities on visual solution-explaining, current open-source general-purpose models perform inconsistently, particularly in identifying relevant visual components and producing coherent keypoint-based explanations. We expect that visual solution-explaining and the MathExplain dataset will catalyze further research on multimodal LLMs in education and advance their deployment as effective, explanation-oriented AI tutors. Code and data will be released publicly.
Authors: Xiaohan Yuan, Jinfeng Li, Dongxia Wang, Yuefeng Chen, Xiaofeng Mao, Longtao Huang, Jialuo Chen, Hui Xue, Xiaoxia Liu, Wenhai Wang, Kui Ren, Jingyi Wang
Abstract: Generative large language models (LLMs) have revolutionized natural language processing with their transformative and emergent capabilities. However, recent evidence indicates that LLMs can produce harmful content that violates social norms, raising significant concerns regarding the safety and ethical ramifications of deploying these advanced models. Thus, it is both critical and imperative to perform a rigorous and comprehensive safety evaluation of LLMs before deployment. Despite this need, owing to the extensiveness of LLM generation space, it still lacks a unified and standardized risk taxonomy to systematically reflect the LLM content safety, as well as automated safety assessment techniques to explore the potential risk efficiently. To bridge the striking gap, we propose S-Eval, a novel LLM-based automated Safety Evaluation framework with a newly defined comprehensive risk taxonomy. S-Eval incorporates two key components, i.e., an expert testing LLM ${M}_t$ and a novel safety critique LLM ${M}_c$. ${M}_t$ is responsible for automatically generating test cases in accordance with the proposed risk taxonomy. ${M}_c$ can provide quantitative and explainable safety evaluations for better risk awareness of LLMs. In contrast to prior works, S-Eval is efficient and effective in test generation and safety evaluation. Moreover, S-Eval can be flexibly configured and adapted to the rapid evolution of LLMs and accompanying new safety threats, test generation methods and safety critique methods thanks to the LLM-based architecture. S-Eval has been deployed in our industrial partner for the automated safety evaluation of multiple LLMs serving millions of users, demonstrating its effectiveness in real-world scenarios. Our benchmark is publicly available at https://github.com/IS2Lab/S-Eval.
Authors: Yichi Zhang, Zhuo Chen, Lingbing Guo, Yajing Xu, Binbin Hu, Ziqi Liu, Wen Zhang, Huajun Chen
Abstract: Learning high-quality multi-modal entity representations is an important goal of multi-modal knowledge graph (MMKG) representation learning, which can enhance reasoning tasks within the MMKGs, such as MMKG completion (MMKGC). The main challenge is to collaboratively model the structural information concealed in massive triples and the multi-modal features of the entities. Existing methods focus on crafting elegant entity-wise multi-modal fusion strategies, yet they overlook the utilization of multi-perspective features concealed within the modalities under diverse relational contexts. To address this issue, we introduce a novel framework with Mixture of Modality Knowledge experts (MoMoK for short) to learn adaptive multi-modal entity representations for better MMKGC. We design relation-guided modality knowledge experts to acquire relation-aware modality embeddings and integrate the predictions from multi-modalities to achieve joint decisions. Additionally, we disentangle the experts by minimizing their mutual information. Experiments on four public MMKG benchmarks demonstrate the outstanding performance of MoMoK under complex scenarios.
Authors: Tarun Khajuria, Braian Olmiro Dias, Marharyta Domnich, Jaan Aru
Abstract: In this work, we interpret the representations of multi-object scenes in vision encoders through the lens of structured representations. Structured representations allow modeling of individual objects distinctly and their flexible use based on the task context for both scene-level and object-specific tasks. These capabilities play a central role in human reasoning and generalization, allowing us to abstract away irrelevant details and focus on relevant information in a compact and usable form. We define structured representations as those that adhere to two specific properties: binding specific object information into discrete representation units and segregating object representations into separate sets of tokens to minimize cross-object entanglement. Based on these properties, we evaluated and compared image encoders pre-trained on classification (ViT), large vision-language models (CLIP, BLIP, FLAVA), and self-supervised methods (DINO, DINOv2). We examine the token representations by creating object-decoding tasks that measure the ability of specific tokens to capture individual objects in multi-object scenes from the COCO dataset. This analysis provides insights into how object-wise representations are distributed across tokens and layers within these vision encoders. Our findings highlight significant differences in the representation of objects depending on their relevance to the pre-training objective, with this effect particularly pronounced in the CLS token (often used for downstream tasks). Meanwhile, networks and layers that exhibit more structured representations retain better information about individual objects. To guide practical applications, we propose formal measures to quantify the two properties of structured representations, aiding in selecting and adapting vision encoders for downstream tasks.
Authors: Nikhil Abhyankar, Vivek Gupta, Dan Roth, Chandan K. Reddy
Abstract: Tabular reasoning involves interpreting natural language queries about tabular data, which presents a unique challenge of combining language understanding with structured data analysis. Existing methods employ either textual reasoning, which excels in semantic interpretation but struggles with mathematical operations, or symbolic reasoning, which handles computations well but lacks semantic understanding. This paper introduces a novel algorithm H-STAR that integrates both symbolic and semantic (textual) approaches in a two-stage process to address these limitations. H-STAR employs: (1) step-wise table extraction using `multi-view' column retrieval followed by row extraction, and (2) adaptive reasoning that adapts reasoning strategies based on question types, utilizing semantic reasoning for direct lookup and complex lexical queries while augmenting textual reasoning with symbolic reasoning support for quantitative and logical tasks. Our extensive experiments demonstrate that H-STAR significantly outperforms state-of-the-art methods across three tabular question-answering (QA) and fact-verification datasets, underscoring its effectiveness and efficiency.
Authors: Feiyang Kang, Yifan Sun, Bingbing Wen, Si Chen, Dawn Song, Rafid Mahmood, Ruoxi Jia
Abstract: Domain reweighting is an emerging research area aimed at adjusting the relative weights of different data sources to improve the effectiveness and efficiency of LLM pre-training. We show that data mixtures that perform well at smaller scales may not retain their advantage at larger scales, challenging the existing practice of determining competitive mixtures in small-scale experiments and directly applying them at much larger scales. To address this, we propose AutoScale, a two-stage, scale-aware data composition framework. First, AutoScale fits a parametric model that predicts the model's loss under different data compositions, then uses it to find an approximate best allocation at smaller, more manageable budgets. Next, leveraging a novel theoretical analysis of how optimal compositions evolve with scale, AutoScale extrapolates that composition to larger budgets without further retraining. Empirically, AutoScale accelerates convergence and improves downstream performance. For instance, when pre-training GPT-2 Large, it achieves a 28% faster perplexity reduction than baselines and up to a 38% speed-up over unweighted training, while yielding best-average results on various downstream tasks. Overall, our findings illustrate how domain importance shifts with training scale, underscoring the need for scale-dependent data curation in LLM training. Our code is open-sourced.
Authors: Zhichao Wang, Bin Bi, Can Huang, Shiva Kumar Pentyala, Zixu James Zhu, Sitaram Asur, Na Claire Cheng
Abstract: An LLM is pretrained on trillions of tokens, but the pretrained LLM may still generate undesired responses. To solve this problem, alignment techniques such as RLHF, DPO and KTO are proposed. However, these alignment techniques have limitations. For example, RLHF requires training the reward model and policy separately, which is complex, time-consuming, memory intensive and unstable during training processes. DPO proposes a mapping between an optimal policy and a reward, greatly simplifying the training process of RLHF. However, it can not take full advantages of a reward model and it is limited to pairwise preference data. In this paper, we propose \textbf{UN}ified \textbf{A}lignment (UNA) which unifies RLHF/PPO, DPO and KTO. Firstly, we mathematically prove that given the classical RLHF objective, the optimal policy is induced by a generalize implicit reward function. With this novel mapping between a reward model and an optimal policy, UNA can 1. unify RLHF/PPO, DPO and KTO into a supervised learning of minimizing the difference between an implicit reward and an explicit reward; 2. outperform RLHF/PPO while simplify, stabilize, speed up and reduce memory burden of RL fine-tuning process; 3. accommodate different feedback types including pairwise, binary and scalar feedback. Downstream experiments show UNA outperforms DPO, KTO and RLHF.
Authors: Weichen Dai, Yezeng Chen, Zijie Dai, Yubo Liu, Zhijie Huang, Yixuan Pan, Baiyang Song, Chengli Zhong, Xinhe Li, Zeyu Wang, Zhuoying Feng, Yi Zhou
Abstract: Artificial intelligence is gradually demonstrating its immense potential, and increasing attention is being given to how AI can be harnessed to advance scientific research. In this vision paper, we present our perspectives on how AI can better assist scientific inquiry and explore corresponding technical approach. We have proposed and open-sourced two large models of our KALE-LM model series, KALE-LM-Chem(-1.5), which have achieved outstanding performance in tasks related to the field of chemistry. We hope that our work serves as a strong starting point, helping to realize more intelligent AI and promoting the advancement of human science and technology, as well as societal development.
Authors: Payam Saeedi, Mahsa Goodarzi, M Abdullah Canbaz
Abstract: We investigate the presence of cognitive biases in three large language models (LLMs): GPT-4o, Gemma 2, and Llama 3.1. The study uses 1,500 experiments across nine established cognitive biases to evaluate the models' responses and consistency. GPT-4o demonstrated the strongest overall performance. Gemma 2 showed strengths in addressing the sunk cost fallacy and prospect theory, however its performance varied across different biases. Llama 3.1 consistently underperformed, relying on heuristics and exhibiting frequent inconsistencies and contradictions. The findings highlight the challenges of achieving robust and generalizable reasoning in LLMs, and underscore the need for further development to mitigate biases in artificial general intelligence (AGI). The study emphasizes the importance of integrating statistical reasoning and ethical considerations in future AI development.
Authors: Alex Iacob, Lorenzo Sani, Meghdad Kurmanji, William F. Shen, Xinchi Qiu, Dongqi Cai, Yan Gao, Nicholas D. Lane
Abstract: Language Model pre-training uses broad data mixtures to enhance performance across domains and languages. However, training on such heterogeneous text corpora requires extensive and expensive efforts. Since these data sources vary significantly in lexical, syntactic, and semantic aspects, they cause negative interference or the ``curse of multilinguality''. To address these challenges we propose a communication-efficient pre-training framework, DEPT. Our method decouples embeddings from the transformer body while simultaneously training the latter on multiple data sources without requiring a shared vocabulary. DEPT can: (1) train robustly and effectively under significant data heterogeneity, (2) minimize token embedding parameters to only what the data source vocabulary requires, while cutting communication costs in direct proportion to both the communication frequency and the reduction in parameters, (3) enhance transformer body plasticity and generalization, improving both average perplexity (up to 20%) and downstream task performance, and (4) enable training with custom optimized vocabularies per data source. We demonstrate DEPT's potential via the first vocabulary-agnostic federated pre-training of billion-scale models, reducing communication costs by orders of magnitude and embedding memory by 4-5x.
Authors: Sijun Tan, Siyuan Zhuang, Kyle Montgomery, William Y. Tang, Alejandro Cuadron, Chenguang Wang, Raluca Ada Popa, Ion Stoica
Abstract: LLM-based judges have emerged as a scalable alternative to human evaluation and are increasingly used to assess, compare, and improve models. However, the reliability of LLM-based judges themselves is rarely scrutinized. As LLMs become more advanced, their responses grow more sophisticated, requiring stronger judges to evaluate them. Existing benchmarks primarily focus on a judge's alignment with human preferences, but often fail to account for more challenging tasks where crowdsourced human preference is a poor indicator of factual and logical correctness. To address this, we propose a novel evaluation framework to objectively evaluate LLM-based judges. Based on this framework, we propose JudgeBench, a benchmark for evaluating LLM-based judges on challenging response pairs spanning knowledge, reasoning, math, and coding. JudgeBench leverages a novel pipeline for converting existing difficult datasets into challenging response pairs with preference labels reflecting objective correctness. Our comprehensive evaluation on a collection of prompted judges, fine-tuned judges, multi-agent judges, and reward models shows that JudgeBench poses a significantly greater challenge than previous benchmarks, with many strong models (e.g., GPT-4o) performing just slightly better than random guessing. Overall, JudgeBench offers a reliable platform for assessing increasingly advanced LLM-based judges. Data and code are available at https://github.com/ScalerLab/JudgeBench.
Authors: Zedian Shao, Hongbin Liu, Jaden Mu, Neil Zhenqiang Gong
Abstract: In a prompt injection attack, an attacker injects a prompt into the original one, aiming to make an LLM follow the injected prompt to perform an attacker-chosen task. Existing attacks primarily focus on how to blend the injected prompt into the original prompt without altering the LLM itself. Our experiments show that these attacks achieve some success, but there is still significant room for improvement. In this work, we show that an attacker can boost the success of prompt injection attacks by poisoning the LLM's alignment process. Specifically, we propose PoisonedAlign, a method to strategically create poisoned alignment samples. When even a small fraction of the alignment data is poisoned using our method, the aligned LLM becomes more vulnerable to prompt injection while maintaining its foundational capabilities. The code is available at https://github.com/Sadcardation/PoisonedAlign
Authors: Minhua Lin, Zhengzhang Chen, Yanchi Liu, Xujiang Zhao, Zongyu Wu, Junxiang Wang, Xiang Zhang, Suhang Wang, Haifeng Chen
Abstract: Time series data is ubiquitous across various domains, including manufacturing, finance, and healthcare. High-quality annotations are essential for effectively understanding time series and facilitating downstream tasks; however, obtaining such annotations is challenging, particularly in mission-critical domains. In this paper, we propose TESSA, a multi-agent system designed to automatically generate both general and domain-specific annotations for time series data. TESSA introduces two agents: a general annotation agent and a domain-specific annotation agent. The general agent captures common patterns and knowledge across multiple source domains, leveraging both time-series-wise and text-wise features to generate general annotations. Meanwhile, the domain-specific agent utilizes limited annotations from the target domain to learn domain-specific terminology and generate targeted annotations. Extensive experiments on multiple synthetic and real-world datasets demonstrate that TESSA effectively generates high-quality annotations, outperforming existing methods.
Authors: Chan Hee Song, Valts Blukis, Jonathan Tremblay, Stephen Tyree, Yu Su, Stan Birchfield
Abstract: Spatial understanding is a crucial capability that enables robots to perceive their surroundings, reason about their environment, and interact with it meaningfully. In modern robotics, these capabilities are increasingly provided by vision-language models. However, these models face significant challenges in spatial reasoning tasks, as their training data are based on general-purpose image datasets that often lack sophisticated spatial understanding. For example, datasets frequently do not capture reference frame comprehension, yet effective spatial reasoning requires understanding whether to reason from ego-, world-, or object-centric perspectives. To address this issue, we introduce RoboSpatial, a large-scale dataset for spatial understanding in robotics. It consists of real indoor and tabletop scenes, captured as 3D scans and egocentric images, and annotated with rich spatial information relevant to robotics. The dataset includes 1M images, 5k 3D scans, and 3M annotated spatial relationships, and the pairing of 2D egocentric images with 3D scans makes it both 2D- and 3D- ready. Our experiments show that models trained with RoboSpatial outperform baselines on downstream tasks such as spatial affordance prediction, spatial relationship prediction, and robot manipulation.
Authors: Guofeng Mei, Wei Lin, Luigi Riz, Yujiao Wu, Fabio Poiesi, Yiming Wang
Abstract: Enabling Large Language Models (LLMs) to understand the 3D physical world is an emerging yet challenging research direction. Current strategies for processing point clouds typically downsample the scene or divide it into smaller parts for separate analysis. However, both approaches risk losing key local details or global contextual information. In this paper, we introduce PerLA, a 3D language assistant designed to be more perceptive to both details and context, making visual representations more informative for the LLM. PerLA captures high-resolution (local) details in parallel from different point cloud areas and integrates them with (global) context obtained from a lower-resolution whole point cloud. We present a novel algorithm that preserves point cloud locality through the Hilbert curve and effectively aggregates local-to-global information via cross-attention and a graph neural network. Lastly, we introduce a novel loss for local representation consensus to promote training stability. PerLA outperforms state-of-the-art 3D language assistants, with gains of up to +1.34 CiDEr on ScanQA for question answering, and +4.22 on ScanRefer and +3.88 on Nr3D for dense captioning. https://gfmei.github.io/PerLA/
Authors: Luoxuan Weng, Yinghao Tang, Yingchaojie Feng, Zhuo Chang, Ruiqin Chen, Haozhe Feng, Chen Hou, Danqing Huang, Yang Li, Huaming Rao, Haonan Wang, Canshi Wei, Xiaofeng Yang, Yuhui Zhang, Yifeng Zheng, Xiuqi Huang, Minfeng Zhu, Yuxin Ma, Bin Cui, Peng Chen, Wei Chen
Abstract: Business intelligence (BI) transforms large volumes of data within modern organizations into actionable insights for informed decision-making. Recently, large language model (LLM)-based agents have streamlined the BI workflow by automatically performing task planning, reasoning, and actions in executable environments based on natural language (NL) queries. However, existing approaches primarily focus on individual BI tasks such as NL2SQL and NL2VIS. The fragmentation of tasks across different data roles and tools lead to inefficiencies and potential errors due to the iterative and collaborative nature of BI. In this paper, we introduce DataLab, a unified BI platform that integrates a one-stop LLM-based agent framework with an augmented computational notebook interface. DataLab supports various BI tasks for different data roles in data preparation, analysis, and visualization by seamlessly combining LLM assistance with user customization within a single environment. To achieve this unification, we design a domain knowledge incorporation module tailored for enterprise-specific BI tasks, an inter-agent communication mechanism to facilitate information sharing across the BI workflow, and a cell-based context management strategy to enhance context utilization efficiency in BI notebooks. Extensive experiments demonstrate that DataLab achieves state-of-the-art performance on various BI tasks across popular research benchmarks. Moreover, DataLab maintains high effectiveness and efficiency on real-world datasets from Tencent, achieving up to a 58.58% increase in accuracy and a 61.65% reduction in token cost on enterprise-specific BI tasks.
Authors: Zhenting Wang, Shuming Hu, Shiyu Zhao, Xiaowen Lin, Felix Juefei-Xu, Zhuowei Li, Ligong Han, Harihar Subramanyam, Li Chen, Jianfa Chen, Nan Jiang, Lingjuan Lyu, Shiqing Ma, Dimitris N. Metaxas, Ankit Jain
Abstract: Image content safety has become a significant challenge with the rise of visual media on online platforms. Meanwhile, in the age of AI-generated content (AIGC), many image generation models are capable of producing harmful content, such as images containing sexual or violent material. Thus, it becomes crucial to identify such unsafe images based on established safety rules. Pre-trained Multimodal Large Language Models (MLLMs) offer potential in this regard, given their strong pattern recognition abilities. Existing approaches typically fine-tune MLLMs with human-labeled datasets, which however brings a series of drawbacks. First, relying on human annotators to label data following intricate and detailed guidelines is both expensive and labor-intensive. Furthermore, users of safety judgment systems may need to frequently update safety rules, making fine-tuning on human-based annotation more challenging. This raises the research question: Can we detect unsafe images by querying MLLMs in a zero-shot setting using a predefined safety constitution (a set of safety rules)? Our research showed that simply querying pre-trained MLLMs does not yield satisfactory results. This lack of effectiveness stems from factors such as the subjectivity of safety rules, the complexity of lengthy constitutions, and the inherent biases in the models. To address these challenges, we propose a MLLM-based method includes objectifying safety rules, assessing the relevance between rules and images, making quick judgments based on debiased token probabilities with logically complete yet simplified precondition chains for safety rules, and conducting more in-depth reasoning with cascaded chain-of-thought processes if necessary. Experiment results demonstrate that our method is highly effective for zero-shot image safety judgment tasks.
Authors: Stefan Szeider
Abstract: The MCP Solver bridges Large Language Models (LLMs) with symbolic solvers through the Model Context Protocol (MCP), an open-source standard for AI system integration. Providing LLMs access to formal solving and reasoning capabilities addresses their key deficiency while leveraging their strengths. Our implementation offers interfaces for constraint programming (Minizinc), propositional satisfiability (PySAT), and SAT modulo Theories (Python Z3). The system employs an editing approach with iterated validation to ensure model consistency during modifications and enable structured refinement.
Authors: Yuxiao Hu, Qian Li, Dongxiao Zhang, Jinyue Yan, Yuntian Chen
Abstract: Recently, leveraging pre-trained Large Language Models (LLMs) for time series (TS) tasks has gained increasing attention, which involves activating and enhancing LLMs' capabilities. Many methods aim to activate LLMs' capabilities based on token-level alignment but overlook LLMs' inherent strength on natural language processing -- their deep understanding of linguistic logic and structure rather than superficial embedding processing. We propose Context-Alignment, a new paradigm that aligns TS with a linguistic component in the language environments familiar to LLMs to enable LLMs to contextualize and comprehend TS data, thereby activating their capabilities. Specifically, such context-level alignment comprises structural alignment and logical alignment, which is achieved by a Dual-Scale Context-Alignment GNNs (DSCA-GNNs) applied to TS-language multimodal inputs. Structural alignment utilizes dual-scale nodes to describe hierarchical structure in TS-language, enabling LLMs treat long TS data as a whole linguistic component while preserving intrinsic token features. Logical alignment uses directed edges to guide logical relationships, ensuring coherence in the contextual semantics. Demonstration examples prompt are employed to construct Demonstration Examples based Context-Alignment (DECA) following DSCA-GNNs framework. DECA can be flexibly and repeatedly integrated into various layers of pre-trained LLMs to improve awareness of logic and structure, thereby enhancing performance. Extensive experiments show the effectiveness of DECA and the importance of Context-Alignment across tasks, particularly in few-shot and zero-shot forecasting, confirming that Context-Alignment provide powerful prior knowledge on context.
Authors: Faria Huq, Zora Zhiruo Wang, Frank F. Xu, Tianyue Ou, Shuyan Zhou, Jeffrey P. Bigham, Graham Neubig
Abstract: While much work on web agents emphasizes the promise of autonomously performing tasks on behalf of users, in reality, agents often fall short on complex tasks in real-world contexts and modeling user preference. This presents an opportunity for humans to collaborate with the agent and leverage the agent's capabilities effectively. We propose CowPilot, a framework supporting autonomous as well as human-agent collaborative web navigation, and evaluation across task success and task efficiency. CowPilot reduces the number of steps humans need to perform by allowing agents to propose next steps, while users are able to pause, reject, or take alternative actions. During execution, users can interleave their actions with the agent by overriding suggestions or resuming agent control when needed. We conducted case studies on five common websites and found that the human-agent collaborative mode achieves the highest success rate of 95% while requiring humans to perform only 15.2% of the total steps. Even with human interventions during task execution, the agent successfully drives up to half of task success on its own. CowPilot can serve as a useful tool for data collection and agent evaluation across websites, which we believe will enable research in how users and agents can work together. Video demonstrations are available at https://oaishi.github.io/cowpilot.html
Authors: Haywood Gelman, John D. Hastings
Abstract: Insider threats wield an outsized influence on organizations, disproportionate to their small numbers. This is due to the internal access insiders have to systems, information, and infrastructure. %One example of this influence is where anonymous respondents submit web-based job search site reviews, an insider threat risk to organizations. Signals for such risks may be found in anonymous submissions to public web-based job search site reviews. This research studies the potential for large language models (LLMs) to analyze and detect insider threat sentiment within job site reviews. Addressing ethical data collection concerns, this research utilizes synthetic data generation using LLMs alongside existing job review datasets. A comparative analysis of sentiment scores generated by LLMs is benchmarked against expert human scoring. Findings reveal that LLMs demonstrate alignment with human evaluations in most cases, thus effectively identifying nuanced indicators of threat sentiment. The performance is lower on human-generated data than synthetic data, suggesting areas for improvement in evaluating real-world data. Text diversity analysis found differences between human-generated and LLM-generated datasets, with synthetic data exhibiting somewhat lower diversity. Overall, the results demonstrate the applicability of LLMs to insider threat detection, and a scalable solution for insider sentiment testing by overcoming ethical and logistical barriers tied to data acquisition.
Authors: Julien Siems, Timur Carstensen, Arber Zela, Frank Hutter, Massimiliano Pontil, Riccardo Grazzi
Abstract: Linear Recurrent Neural Networks (linear RNNs) have emerged as competitive alternatives to Transformers for sequence modeling, offering efficient training and linear-time inference. However, existing architectures face a fundamental trade-off between expressivity and efficiency, dictated by the structure of their state-transition matrices. While diagonal matrices used in architectures like Mamba, GLA, or mLSTM yield fast runtime, they suffer from severely limited expressivity. To address this, recent architectures such as (Gated) DeltaNet and RWKV-7 adopted a diagonal plus rank-1 structure, allowing simultaneous token-channel mixing, which overcomes some expressivity limitations with only a slight decrease in training efficiency. Building on the interpretation of DeltaNet's recurrence as performing one step of online gradient descent per token on an associative recall loss, we introduce DeltaProduct, which instead takes multiple ($n_h$) steps per token. This naturally leads to diagonal plus rank-$n_h$ state-transition matrices, formed as products of $n_h$ generalized Householder transformations, providing a tunable mechanism to balance expressivity and efficiency and a stable recurrence. Through extensive experiments, we demonstrate that DeltaProduct achieves superior state-tracking and language modeling capabilities while exhibiting significantly improved length extrapolation compared to DeltaNet. Additionally, we also strengthen the theoretical foundation of DeltaNet by proving that it can solve dihedral group word problems in just two layers.
Authors: Yuxuan Lu, Bingsheng Yao, Hansu Gu, Jing Huang, Jessie Wang, Yang Li, Jiri Gesi, Qi He, Toby Jia-Jun Li, Dakuo Wang
Abstract: Usability testing is a fundamental yet challenging (e.g., inflexible to iterate the study design flaws and hard to recruit study participants) research method for user experience (UX) researchers to evaluate a web design. Recent advances in Large Language Model-simulated Agent (LLM-Agent) research inspired us to design UXAgent to support UX researchers in evaluating and reiterating their usability testing study design before they conduct the real human subject study. Our system features an LLM-Agent module and a universal browser connector module so that UX researchers can automatically generate thousands of simulated users to test the target website. The results are shown in qualitative (e.g., interviewing how an agent thinks ), quantitative (e.g., # of actions), and video recording formats for UX researchers to analyze. Through a heuristic user evaluation with five UX researchers, participants praised the innovation of our system but also expressed concerns about the future of LLM Agent-assisted UX study.
Authors: William Rudman, Michal Golovanevsky, Amir Bar, Vedant Palit, Yann LeCun, Carsten Eickhoff, Ritambhara Singh
Abstract: Despite strong performance on vision-language tasks, Multimodal Large Language Models (MLLMs) struggle with mathematical problem-solving, with both open-source and state-of-the-art models falling short of human performance on visual-math benchmarks. To systematically examine visual-mathematical reasoning in MLLMs, we (1) evaluate their understanding of geometric primitives, (2) test multi-step reasoning, and (3) explore a potential solution to improve visual reasoning capabilities. Our findings reveal fundamental shortcomings in shape recognition, with top models achieving under 50% accuracy in identifying regular polygons. We analyze these failures through the lens of dual-process theory and show that MLLMs rely on System 1 (intuitive, memorized associations) rather than System 2 (deliberate reasoning). Consequently, MLLMs fail to count the sides of both familiar and novel shapes, suggesting they have neither learned the concept of sides nor effectively process visual inputs. Finally, we propose Visually Cued Chain-of-Thought (VC-CoT) prompting, which enhances multi-step mathematical reasoning by explicitly referencing visual annotations in diagrams, boosting GPT-4o's accuracy on an irregular polygon side-counting task from 7% to 93%. Our findings suggest that System 2 reasoning in MLLMs remains an open problem, and visually-guided prompting is essential for successfully engaging visual reasoning. Code available at: https://github.com/rsinghlab/Shape-Blind.
Authors: Xu Wang, Jiaju Kang, Puyu Han, Yubao Zhao, Qian Liu, Liwenfei He, Lingqiong Zhang, Lingyun Dai, Yongcheng Wang, Jie Tao
Abstract: We present ECG-Expert-QA, a comprehensive multimodal dataset for evaluating diagnostic capabilities in electrocardiogram (ECG) interpretation. It combines real-world clinical ECG data with systematically generated synthetic cases, covering 12 essential diagnostic tasks and totaling 47,211 expert-validated QA pairs. These encompass diverse clinical scenarios, from basic rhythm recognition to complex diagnoses involving rare conditions and temporal changes. A key innovation is the support for multi-turn dialogues, enabling the development of conversational medical AI systems that emulate clinician-patient or interprofessional interactions. This allows for more realistic assessment of AI models' clinical reasoning, diagnostic accuracy, and knowledge integration. Constructed through a knowledge-guided framework with strict quality control, ECG-Expert-QA ensures linguistic and clinical consistency, making it a high-quality resource for advancing AI-assisted ECG interpretation. It challenges models with tasks like identifying subtle ischemic changes and interpreting complex arrhythmias in context-rich scenarios. To promote research transparency and collaboration, the dataset, accompanying code, and prompts are publicly released at https://github.com/Zaozzz/ECG-Expert-QA
Authors: Qingpei Guo, Kaiyou Song, Zipeng Feng, Ziping Ma, Qinglong Zhang, Sirui Gao, Xuzheng Yu, Yunxiao Sun, Tai-Wei Chang, Jingdong Chen, Ming Yang, Jun Zhou
Abstract: We present M2-omni, a cutting-edge, open-source omni-MLLM that achieves competitive performance to GPT-4o. M2-omni employs a unified multimodal sequence modeling framework, which empowers Large Language Models(LLMs) to acquire comprehensive cross-modal understanding and generation capabilities. Specifically, M2-omni can process arbitrary combinations of audio, video, image, and text modalities as input, generating multimodal sequences interleaving with audio, image, or text outputs, thereby enabling an advanced and interactive real-time experience. The training of such an omni-MLLM is challenged by significant disparities in data quantity and convergence rates across modalities. To address these challenges, we propose a step balance strategy during pre-training to handle the quantity disparities in modality-specific data. Additionally, a dynamically adaptive balance strategy is introduced during the instruction tuning stage to synchronize the modality-wise training progress, ensuring optimal convergence. Notably, we prioritize preserving strong performance on pure text tasks to maintain the robustness of M2-omni's language understanding capability throughout the training process. To our best knowledge, M2-omni is currently a very competitive open-source model to GPT-4o, characterized by its comprehensive modality and task support, as well as its exceptional performance. We expect M2-omni will advance the development of omni-MLLMs, thus facilitating future research in this domain.
Authors: Yanshu Li
Abstract: Multimodal in-context learning (ICL) has emerged as a key capability of Large Vision-Language Models (LVLMs), driven by their increasing scale and applicability. Despite its promise, effective ICL in the multimodal setting remains challenging due to the inherent complexity of image-text inputs and the high sensitivity of ICL performance to input configurations. In this work, we shed light on the core mechanism underlying multimodal ICL, identifying task mapping as a crucial factor in configuring robust in-context demonstration (ICD) sequences. Building on these insights, we propose \textit{SabER}, a lightweight yet powerful decoder-only transformer equipped with task-aware attention, which intelligently selects and arranges ICDs from a demonstration library in an autoregressive fashion. This design enables fine-grained feature extraction and cross-modal reasoning, iteratively refining task mapping to generate high-quality ICD sequences. Through extensive experiments covering five LVLMs and nine benchmark datasets, SabER not only demonstrates strong empirical performance, but also provides deeper understanding of how task semantics interact with multimodal ICDs. Our findings highlight the importance of principled ICD sequence configuration and open new avenues to enhance multimodal ICL in a wide range of real-world scenarios.
Authors: Jaymari Chua, Chen Wang, Lina Yao
Abstract: As artificial intelligence surpasses human performance in select tasks, disclosing superhuman capabilities poses distinct challenges for fairness, accountability, and trust. However, the impact of such disclosures on diverse user attitudes and behaviors remains unclear, particularly concerning potential negative reactions like discouragement or overreliance. This paper investigates these effects by utilizing Persona Cards: a validated, standardized set of synthetic personas designed to simulate diverse user reactions and fairness perspectives. We conducted an ethics board-approved study (N=32), utilizing these personas to investigate how capability disclosure influenced behaviors with a superhuman game AI in competitive StarCraft II scenarios. Our results reveal transparency is double-edged: while disclosure could alleviate suspicion, it also provoked frustration and strategic defeatism among novices in cooperative scenarios, as well as overreliance in competitive contexts. Experienced and competitive players interpreted disclosure as confirmation of an unbeatable opponent, shifting to suboptimal goals. We release the Persona Cards Dataset, including profiles, prompts, interaction logs, and protocols, to foster reproducible research into human alignment AI design. This work demonstrates that transparency is not a cure-all; successfully leveraging disclosure to enhance trust and accountability requires careful tailoring to user characteristics, domain norms, and specific fairness objectives.
Authors: Haoyu Wang, Christopher M. Poskitt, Jun Sun
Abstract: Agents built on LLMs are increasingly deployed across diverse domains, automating complex decision-making and task execution. However, their autonomy introduces safety risks, including security vulnerabilities, legal violations, and unintended harmful actions. Existing mitigation methods, such as model-based safeguards and early enforcement strategies, fall short in robustness, interpretability, and adaptability. To address these challenges, we propose AgentSpec, a lightweight domain-specific language for specifying and enforcing runtime constraints on LLM agents. With AgentSpec, users define structured rules that incorporate triggers, predicates, and enforcement mechanisms, ensuring agents operate within predefined safety boundaries. We implement AgentSpec across multiple domains, including code execution, embodied agents, and autonomous driving, demonstrating its adaptability and effectiveness. Our evaluation shows that AgentSpec successfully prevents unsafe executions in over 90% of code agent cases, eliminates all hazardous actions in embodied agent tasks, and enforces 100% compliance by autonomous vehicles (AVs). Despite its strong safety guarantees, AgentSpec remains computationally lightweight, with overheads in milliseconds. By combining interpretability, modularity, and efficiency, AgentSpec provides a practical and scalable solution for enforcing LLM agent safety across diverse applications. We also automate the generation of rules using LLMs and assess their effectiveness. Our evaluation shows that the rules generated by OpenAI o1 achieve a precision of 95.56% and recall of 70.96% for embodied agents, successfully identifying 87.26% of the risky code, and prevent AVs from breaking laws in 5 out of 8 scenarios.
Authors: Songtao Sun, Jingyi Li, Yuanfei Dong, Haoguang Liu, Chenxin Xu, Fuyang Li, Qiang Liu
Abstract: This paper introduces a multi-agent application system designed to enhance office collaboration efficiency and work quality. The system integrates artificial intelligence, machine learning, and natural language processing technologies, achieving functionalities such as task allocation, progress monitoring, and information sharing. The agents within the system are capable of providing personalized collaboration support based on team members' needs and incorporate data analysis tools to improve decision-making quality. The paper also proposes an intelligent agent architecture that separates Plan and Solver, and through techniques such as multi-turn query rewriting and business tool retrieval, it enhances the agent's multi-intent and multi-turn dialogue capabilities. Furthermore, the paper details the design of tools and multi-turn dialogue in the context of office collaboration scenarios, and validates the system's effectiveness through experiments and evaluations. Ultimately, the system has demonstrated outstanding performance in real business applications, particularly in query understanding, task planning, and tool calling. Looking forward, the system is expected to play a more significant role in addressing complex interaction issues within dynamic environments and large-scale multi-agent systems.
Authors: Giulio Starace, Oliver Jaffe, Dane Sherburn, James Aung, Jun Shern Chan, Leon Maksin, Rachel Dias, Evan Mays, Benjamin Kinsella, Wyatt Thompson, Johannes Heidecke, Amelia Glaese, Tejal Patwardhan
Abstract: We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments. For objective evaluation, we develop rubrics that hierarchically decompose each replication task into smaller sub-tasks with clear grading criteria. In total, PaperBench contains 8,316 individually gradable tasks. Rubrics are co-developed with the author(s) of each ICML paper for accuracy and realism. To enable scalable evaluation, we also develop an LLM-based judge to automatically grade replication attempts against rubrics, and assess our judge's performance by creating a separate benchmark for judges. We evaluate several frontier models on PaperBench, finding that the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%. Finally, we recruit top ML PhDs to attempt a subset of PaperBench, finding that models do not yet outperform the human baseline. We open-source our code (https://github.com/openai/preparedness) to facilitate future research in understanding the AI engineering capabilities of AI agents.
Authors: Yuxiang Zheng, Dayuan Fu, Xiangkun Hu, Xiaojie Cai, Lyumanshan Ye, Pengrui Lu, Pengfei Liu
Abstract: Large Language Models (LLMs) equipped with web search capabilities have demonstrated impressive potential for deep research tasks. However, current approaches predominantly rely on either manually engineered prompts (prompt engineering-based) with brittle performance or reinforcement learning within controlled Retrieval-Augmented Generation (RAG) environments (RAG-based) that fail to capture the complexities of real-world interaction. In this paper, we introduce DeepResearcher, the first comprehensive framework for end-to-end training of LLM-based deep research agents through scaling reinforcement learning (RL) in real-world environments with authentic web search interactions. Unlike RAG-based approaches that assume all necessary information exists within a fixed corpus, our method trains agents to navigate the noisy, unstructured, and dynamic nature of the open web. We implement a specialized multi-agent architecture where browsing agents extract relevant information from various webpage structures and overcoming significant technical challenges. Extensive experiments on open-domain research tasks demonstrate that DeepResearcher achieves substantial improvements of up to 28.9 points over prompt engineering-based baselines and up to 7.2 points over RAG-based RL agents. Our qualitative analysis reveals emergent cognitive behaviors from end-to-end RL training, including the ability to formulate plans, cross-validate information from multiple sources, engage in self-reflection to redirect research, and maintain honesty when unable to find definitive answers. Our results highlight that end-to-end training in real-world web environments is not merely an implementation detail but a fundamental requirement for developing robust research capabilities aligned with real-world applications. We release DeepResearcher at https://github.com/GAIR-NLP/DeepResearcher.