Authors: Rita Sevastjanova, Robin Gerling, Thilo Spinner, Mennatallah El-Assady
Abstract: Large language models (LLMs) represent words through contextual word embeddings encoding different language properties like semantics and syntax. Understanding these properties is crucial, especially for researchers investigating language model capabilities, employing embeddings for tasks related to text similarity, or evaluating the reasons behind token importance as measured through attribution methods. Applications for embedding exploration frequently involve dimensionality reduction techniques, which reduce high-dimensional vectors to two dimensions used as coordinates in a scatterplot. This data transformation step introduces uncertainty that can be propagated to the visual representation and influence users' interpretation of the data. To communicate such uncertainties, we present LayerFlow - a visual analytics workspace that displays embeddings in an interlinked projection design and communicates the transformation, representation, and interpretation uncertainty. In particular, to hint at potential data distortions and uncertainties, the workspace includes several visual components, such as convex hulls showing 2D and HD clusters, data point pairwise distances, cluster summaries, and projection quality metrics. We show the usability of the presented workspace through replication and expert case studies that highlight the need to communicate uncertainty through multiple visual components and different data perspectives.
Authors: Thilo Hagendorff, Sarah Fabi
Abstract: Large language models (LLMs) can perform reasoning computations both internally within their latent space and externally by generating explicit token sequences like chains of thought. Significant progress in enhancing reasoning abilities has been made by scaling test-time compute. However, understanding and quantifying model-internal reasoning abilities - the inferential "leaps" models make between individual token predictions - remains crucial. This study introduces a benchmark (n = 4,000 items) designed to quantify model-internal reasoning in different domains. We achieve this by having LLMs indicate the correct solution to reasoning problems not through descriptive text, but by selecting a specific language of their initial response token that is different from English, the benchmark language. This not only requires models to reason beyond their context window, but also to overrise their default tendency to respond in the same language as the prompt, thereby posing an additional cognitive strain. We evaluate a set of 18 LLMs, showing significant performance variations, with GPT-4.5 achieving the highest accuracy (74.7%), outperforming models like Grok-2 (67.2%), and Llama 3.1 405B (65.6%). Control experiments and difficulty scaling analyses suggest that while LLMs engage in internal reasoning, we cannot rule out heuristic exploitations under certain conditions, marking an area for future investigation. Our experiments demonstrate that LLMs can "think" via latent-space computations, revealing model-internal inference strategies that need further understanding, especially regarding safety-related concerns such as covert planning, goal-seeking, or deception emerging without explicit token traces.
Authors: Afra Amini, Tim Vieira, Ryan Cotterell
Abstract: Estimating the Kullback--Leibler (KL) divergence between language models has many applications, e.g., reinforcement learning from human feedback (RLHF), interpretability, and knowledge distillation. However, computing the exact KL divergence between two arbitrary language models is intractable. Thus, practitioners often resort to the use of sampling-based estimators. While it is easy to fashion a simple Monte Carlo (MC) estimator that provides an unbiased estimate of the KL divergence between language models, this estimator notoriously suffers from high variance, and can even result in a negative estimate of the KL divergence, a non-negative quantity. In this paper, we introduce a Rao--Blackwellized estimator that is also unbiased and provably has variance less than or equal to that of the standard Monte Carlo estimator. In an empirical study on sentiment-controlled fine-tuning, we show that our estimator provides more stable KL estimates and reduces variance substantially in practice. Additionally, we derive an analogous Rao--Blackwellized estimator of the gradient of the KL divergence, which leads to more stable training and produces models that more frequently appear on the Pareto frontier of reward vs. KL compared to the ones trained with the MC estimator of the gradient.
Authors: Saif Punjwani, Larry Heck
Abstract: Large language models (LLMs) have demonstrated remarkable reasoning capabilities when prompted with strategies such as Chain-of-Thought (CoT). However, these approaches focus on token-level output without considering internal weight dynamics. We introduce Weight-of-Thought (WoT) reasoning, a novel approach that examines neural network weights before inference to identify reasoning pathways. Unlike existing methods, WoT explores the weight space through graph-based message passing, multi-step reasoning processes, and attention mechanisms. Our implementation creates an interconnected graph of reasoning nodes. Experiments on diverse reasoning tasks (syllogistic, mathematical, algebraic, combinatorial, and geometric) demonstrate that WoT achieves superior performance compared to traditional methods, particularly for complex problems. This approach leads to both improved performance and greater interpretability of the reasoning process, offering a promising direction for enhancing LLM reasoning capabilities.
Authors: Nafis Sadeq, Xin Xu, Zhouhang Xie, Julian McAuley, Byungkyu Kang, Prarit Lamba, Xiang Gao
Abstract: Language models rely on semantic priors to perform in-context learning, which leads to poor performance on tasks involving inductive reasoning. Instruction-tuning methods based on imitation learning can superficially enhance the in-context learning performance of language models, but they often fail to improve the model's understanding of the underlying rules that connect inputs and outputs in few-shot demonstrations. We propose ReDis, a reasoning distillation technique designed to improve the inductive reasoning capabilities of language models. Through a careful combination of data augmentation, filtering, supervised fine-tuning, and alignment, ReDis achieves significant performance improvements across a diverse range of tasks, including 1D-ARC, List Function, ACRE, and MiniSCAN. Experiments on three language model backbones show that ReDis outperforms equivalent few-shot prompting baselines across all tasks and even surpasses the teacher model, GPT-4o, in some cases. ReDis, based on the LLaMA-3 backbone, achieves relative improvements of 23.2%, 2.8%, and 66.6% over GPT-4o on 1D-ARC, ACRE, and MiniSCAN, respectively, within a similar hypothesis search space. The code, dataset, and model checkpoints will be made available at https://github.com/NafisSadeq/reasoning-distillation.git.
URLs: https://github.com/NafisSadeq/reasoning-distillation.git.
Authors: Paul Rosu
Abstract: This paper introduces an LLM-based Latin-to-English translation platform designed to address the challenges of translating Latin texts. We named the model LITERA, which stands for Latin Interpretation and Translations into English for Research Assistance. Through a multi-layered translation process utilizing a fine-tuned version of GPT-4o-mini and GPT-4o, LITERA offers an unprecedented level of accuracy, showcased by greatly improved BLEU scores, particularly in classical Latin, along with improved BLEURT scores. The development of LITERA involved close collaboration with Duke University's Classical Studies Department, which was instrumental in creating a small, high-quality parallel Latin-English dataset. This paper details the architecture, fine-tuning methodology, and prompting strategies used in LITERA, emphasizing its ability to produce literal translations.
Authors: Mykola Trokhymovych, Oleksandr Kosovan, Nathan Forrester, Pablo Arag\'on, Diego Saez-Trumper, Ricardo Baeza-Yates
Abstract: Wikipedia is powered by MediaWiki, a free and open-source software that is also the infrastructure for many other wiki-based online encyclopedias. These include the recently launched website Ruwiki, which has copied and modified the original Russian Wikipedia content to conform to Russian law. To identify practices and narratives that could be associated with different forms of knowledge manipulation, this article presents an in-depth analysis of this Russian Wikipedia fork. We propose a methodology to characterize the main changes with respect to the original version. The foundation of this study is a comprehensive comparative analysis of more than 1.9M articles from Russian Wikipedia and its fork. Using meta-information and geographical, temporal, categorical, and textual features, we explore the changes made by Ruwiki editors. Furthermore, we present a classification of the main topics of knowledge manipulation in this fork, including a numerical estimation of their scope. This research not only sheds light on significant changes within Ruwiki, but also provides a methodology that could be applied to analyze other Wikipedia forks and similar collaborative projects.
Authors: F. A. Rizvi, T. Navojith, A. M. N. H. Adhikari, W. P. U. Senevirathna, Dharshana Kasthurirathna, Lakmini Abeywardhana
Abstract: Brand reputation in the banking sector is maintained through insightful analysis of customer opinion on code-mixed and multilingual content. Conventional NLP models misclassify or ignore code-mixed text, when mix with low resource languages such as Sinhala-English and fail to capture domain-specific knowledge. This study introduces a hybrid NLP method to improve keyword extraction, content filtering, and aspect-based classification of banking content. Keyword extraction in English is performed with a hybrid approach comprising a fine-tuned SpaCy NER model, FinBERT-based KeyBERT embeddings, YAKE, and EmbedRank, which results in a combined accuracy of 91.2%. Code-mixed and Sinhala keywords are extracted using a fine-tuned XLM-RoBERTa model integrated with a domain-specific Sinhala financial vocabulary, and it results in an accuracy of 87.4%. To ensure data quality, irrelevant comment filtering was performed using several models, with the BERT-base-uncased model achieving 85.2% for English and XLM-RoBERTa 88.1% for Sinhala, which was better than GPT-4o, SVM, and keyword-based filtering. Aspect classification followed the same pattern, with the BERT-base-uncased model achieving 87.4% for English and XLM-RoBERTa 85.9% for Sinhala, both exceeding GPT-4 and keyword-based approaches. These findings confirm that fine-tuned transformer models outperform traditional methods in multilingual financial text analysis. The present framework offers an accurate and scalable solution for brand reputation monitoring in code-mixed and low-resource banking environments.
Authors: Soham Shah, Kumar Shridhar, Surojit Chatterjee, Souvik Sen
Abstract: While recent advances in large language models (LLMs) have significantly enhanced performance across diverse natural language tasks, the high computational and financial costs associated with their deployment remain substantial barriers. Existing routing strategies partially alleviate this challenge by assigning queries to cheaper or specialized models, but they frequently rely on extensive labeled data or fragile task-specific heuristics. Conversely, fusion techniques aggregate multiple LLM outputs to boost accuracy and robustness, yet they often exacerbate cost and may reinforce shared biases. We introduce EMAFusion, a new framework that self-optimizes for seamless LLM selection and reliable execution for a given query. Specifically, EMAFusion integrates a taxonomy-based router for familiar query types, a learned router for ambiguous inputs, and a cascading approach that progressively escalates from cheaper to more expensive models based on multi-judge confidence evaluations. Through extensive evaluations, we find EMAFusion outperforms the best individual models by over 2.6 percentage points (94.3% vs. 91.7%), while being 4X cheaper than the average cost. EMAFusion further achieves a remarkable 17.1 percentage point improvement over models like GPT-4 at less than 1/20th the cost. Our combined routing approach delivers 94.3% accuracy compared to taxonomy-based (88.1%) and learned model predictor-based (91.7%) methods alone, demonstrating the effectiveness of our unified strategy. Finally, EMAFusion supports flexible cost-accuracy trade-offs, allowing users to balance their budgetary constraints and performance needs.
Authors: Avinash Kumar, Shashank Nag, Jason Clemons, Lizy John, Poulami Das
Abstract: Deploying large language models (LLMs) presents critical challenges due to the inherent trade-offs associated with key performance metrics, such as latency, accuracy, and throughput. Typically, gains in one metric is accompanied with degradation in others. Early-Exit LLMs (EE-LLMs) efficiently navigate this trade-off space by skipping some of the later model layers when it confidently finds an output token early, thus reducing latency without impacting accuracy. However, as the early exits taken depend on the task and are unknown apriori to request processing, EE-LLMs conservatively load the entire model, limiting resource savings and throughput. Also, current frameworks statically select a model for a user task, limiting our ability to adapt to changing nature of the input queries. We propose HELIOS to address these challenges. First, HELIOS shortlists a set of candidate LLMs, evaluates them using a subset of prompts, gathering telemetry data in real-time. Second, HELIOS uses the early exit data from these evaluations to greedily load the selected model only up to a limited number of layers. This approach yields memory savings which enables us to process more requests at the same time, thereby improving throughput. Third, HELIOS monitors and periodically reassesses the performance of the candidate LLMs and if needed, switches to another model that can service incoming queries more efficiently (such as using fewer layers without lowering accuracy). Our evaluations show that HELIOS achieves 1.48$\times$ throughput, 1.10$\times$ energy-efficiency, 1.39$\times$ lower response time, and 3.7$\times$ improvements in inference batch sizes compared to the baseline, when optimizing for the respective service level objectives.
Authors: Ralf Schm\"alzle, Sue Lim, Yuetong Du, Gary Bente
Abstract: This paper examines the thin-slicing approach - the ability to make accurate judgments based on minimal information - in the context of scientific presentations. Drawing on research from nonverbal communication and personality psychology, we show that brief excerpts (thin slices) reliably predict overall presentation quality. Using a novel corpus of over one hundred real-life science talks, we employ Large Language Models (LLMs) to evaluate transcripts of full presentations and their thin slices. By correlating LLM-based evaluations of short excerpts with full-talk assessments, we determine how much information is needed for accurate predictions. Our results demonstrate that LLM-based evaluations align closely with human ratings, proving their validity, reliability, and efficiency. Critically, even very short excerpts (less than 10 percent of a talk) strongly predict overall evaluations. This suggests that the first moments of a presentation convey relevant information that is used in quality evaluations and can shape lasting impressions. The findings are robust across different LLMs and prompting strategies. This work extends thin-slicing research to public speaking and connects theories of impression formation to LLMs and current research on AI communication. We discuss implications for communication and social cognition research on message reception. Lastly, we suggest an LLM-based thin-slicing framework as a scalable feedback tool to enhance human communication.
Authors: Jessica Lin, Amir Zeldes
Abstract: Determining and ranking the most salient entities in a text is critical for user-facing systems, especially as users increasingly rely on models to interpret long documents they only partially read. Graded entity salience addresses this need by assigning entities scores that reflect their relative importance in a text. Existing approaches fall into two main categories: subjective judgments of salience, which allow for gradient scoring but lack consistency, and summarization-based methods, which define salience as mention-worthiness in a summary, promoting explainability but limiting outputs to binary labels (entities are either summary-worthy or not). In this paper, we introduce a novel approach for graded entity salience that combines the strengths of both approaches. Using an English dataset spanning 12 spoken and written genres, we collect 5 summaries per document and calculate each entity's salience score based on its presence across these summaries. Our approach shows stronger correlation with scores based on human summaries and alignments, and outperforms existing techniques, including LLMs. We release our data and code at https://github.com/jl908069/gum_sum_salience to support further research on graded salient entity extraction.
Authors: Annabella Sakunkoo, Jonathan Sakunkoo
Abstract: Across cultures, names tell a lot about their bearers as they carry deep personal and cultural significance. Names also serve as powerful signals of gender, race, and status in the social hierarchy - a pecking order in which individual positions shape others' expectations on their perceived competence and worth. With the widespread adoption of LLMs and as names are often an input for LLMs, it is crucial to evaluate whether LLMs may sort people into status positions based on first and last names and, if so, whether it is in an unfair, biased fashion. While prior work has primarily investigated biases in first names, little attention has been paid to last names and even less to the combined effects of first and last names. In this study, we conduct a large-scale analysis of name variations across 5 ethnicities to examine how AI exhibits name biases. Our study investigates three key characteristics of inequality and finds that LLMs reflect and reinforce status hierarchies based on names that signal gender and ethnicity as they encode differential expectations of competence, leadership, and economic potential. Contrary to the common assumption that AI tends to favor Whites, we show that East and, in some contexts, South Asian names receive higher rankings. We also disaggregate Asians, a population projected to be the largest immigrant group in the U.S. by 2055. Our results challenge the monolithic Asian model minority assumption, illustrating a more complex and stratified model of bias. Gender moderates biases, with girls facing unfair disadvantages in certain racial groups. Additionally, spanning cultural categories by adopting Western first names improves AI-perceived status for East and Southeast Asian students, particularly for girls. Our findings underscore the importance of intersectional and more nuanced understandings of race, gender, and mixed identities in the evaluation of LLMs.
Authors: Ayoung Lee, Ryan Sungmo Kwon, Peter Railton, Lu Wang
Abstract: Navigating high-stakes dilemmas involving conflicting values is challenging even for humans, let alone for AI. Yet prior work in evaluating the reasoning capabilities of large language models (LLMs) in such situations has been limited to everyday scenarios. To close this gap, this work first introduces CLASH (Character perspective-based LLM Assessments in Situations with High-stakes), a meticulously curated dataset consisting of 345 high-impact dilemmas along with 3,795 individual perspectives of diverse values. In particular, we design CLASH in a way to support the study of critical aspects of value-based decision-making processes which are missing from prior work, including understanding decision ambivalence and psychological discomfort as well as capturing the temporal shifts of values in characters' perspectives. By benchmarking 10 open and closed frontier models, we uncover several key findings. (1) Even the strongest models, such as GPT-4o and Claude-Sonnet, achieve less than 50% accuracy in identifying situations where the decision should be ambivalent, while they perform significantly better in clear-cut scenarios. (2) While LLMs reasonably predict psychological discomfort as marked by human, they inadequately comprehend perspectives involving value shifts, indicating a need for LLMs to reason over complex values. (3) Our experiments also reveal a significant correlation between LLMs' value preferences and their steerability towards a given value. (4) Finally, LLMs exhibit greater steerability when engaged in value reasoning from a third-party perspective, compared to a first-person setup, though certain value pairs benefit uniquely from the first-person framing.
Authors: Phill Kyu Rhee
Abstract: Large Language Models (LLMs), powered by Transformers, have demonstrated human-like intelligence capabilities, yet their underlying mechanisms remain poorly understood. This paper presents a novel framework for interpreting LLMs as probabilistic left context-sensitive languages (CSLs) generators. We hypothesize that Transformers can be effectively decomposed into three fundamental components: context windows, attention mechanisms, and autoregressive generation frameworks. This decomposition allows for the development of more flexible and interpretable computational models, moving beyond the traditional view of attention and autoregression as inseparable processes. We argue that next-token predictions can be understood as probabilistic, dynamic approximations of left CSL production rules, providing an intuitive explanation for how simple token predictions can yield human-like intelligence outputs. Given that all CSLs are left context-sensitive (Penttonen, 1974), we conclude that Transformers stochastically approximate CSLs, which are widely recognized as models of human-like intelligence. This interpretation bridges the gap between Formal Language Theory and the observed generative power of Transformers, laying a foundation for future advancements in generative AI theory and applications. Our novel perspective on Transformer architectures will foster a deeper understanding of LLMs and their future potentials.
Authors: Amanpreet Singh, Joseph Chee Chang, Chloe Anastasiades, Dany Haddad, Aakanksha Naik, Amber Tanaka, Angele Zamarron, Cecile Nguyen, Jena D. Hwang, Jason Dunkleberger, Matt Latzke, Smita Rao, Jaron Lochner, Rob Evans, Rodney Kinney, Daniel S. Weld, Doug Downey, Sergey Feldman
Abstract: Retrieval-augmented generation is increasingly effective in answering scientific questions from literature, but many state-of-the-art systems are expensive and closed-source. We introduce Ai2 Scholar QA, a free online scientific question answering application. To facilitate research, we make our entire pipeline public: as a customizable open-source Python package and interactive web app, along with paper indexes accessible through public APIs and downloadable datasets. We describe our system in detail and present experiments analyzing its key design decisions. In an evaluation on a recent scientific QA benchmark, we find that Ai2 Scholar QA outperforms competing systems.
Authors: Sicheng Feng, Gongfan Fang, Xinyin Ma, Xinchao Wang
Abstract: Reasoning models have demonstrated remarkable progress in solving complex and logic-intensive tasks by generating extended Chain-of-Thoughts (CoTs) prior to arriving at a final answer. Yet, the emergence of this "slow-thinking" paradigm, with numerous tokens generated in sequence, inevitably introduces substantial computational overhead. To this end, it highlights an urgent need for effective acceleration. This survey aims to provide a comprehensive overview of recent advances in efficient reasoning. It categorizes existing works into three key directions: (1) shorter - compressing lengthy CoTs into concise yet effective reasoning chains; (2) smaller - developing compact language models with strong reasoning capabilities through techniques such as knowledge distillation, other model compression techniques, and reinforcement learning; and (3) faster - designing efficient decoding strategies to accelerate inference. A curated collection of papers discussed in this survey is available in our GitHub repository.
Authors: Changjiang Gao, Hankun Lin, Shujian Huang, Xin Huang, Xue Han, Junlan Feng, Chao Deng, Jiajun Chen
Abstract: The ability of cross-lingual context retrieval is a fundamental aspect of cross-lingual alignment of large language models (LLMs), where the model extracts context information in one language based on requests in another language. Despite its importance in real-life applications, this ability has not been adequately investigated for state-of-the-art models. In this paper, we evaluate the cross-lingual context retrieval ability of over 40 LLMs across 12 languages to understand the source of this ability, using cross-lingual machine reading comprehension (xMRC) as a representative scenario. Our results show that several small, post-trained open LLMs show strong cross-lingual context retrieval ability, comparable to closed-source LLMs such as GPT-4o, and their estimated oracle performances greatly improve after post-training. Our interpretability analysis shows that the cross-lingual context retrieval process can be divided into two main phases: question encoding and answer retrieval, which are formed in pre-training and post-training, respectively. The phasing stability correlates with xMRC performance, and the xMRC bottleneck lies at the last model layers in the second phase, where the effect of post-training can be evidently observed. Our results also indicate that larger-scale pretraining cannot improve the xMRC performance. Instead, larger LLMs need further multilingual post-training to fully unlock their cross-lingual context retrieval potential. Our code and is available at https://github.com/NJUNLP/Cross-Lingual-Context-Retrieval
URLs: https://github.com/NJUNLP/Cross-Lingual-Context-Retrieval
Authors: Yingjian Chen, Feiyang Li, Xingyu Song, Tianxiao Li, Issey Sudeka, Irene Li
Abstract: Large language models (LLMs) perform well in medical QA, but their effectiveness in Japanese contexts is limited due to privacy constraints that prevent the use of commercial models like GPT-4 in clinical settings. As a result, recent efforts focus on instruction-tuning open-source LLMs, though the potential of combining them with retrieval-augmented generation (RAG) remains underexplored. To bridge this gap, we are the first to explore a knowledge graph-based (KG) RAG framework for Japanese medical QA small-scale open-source LLMs. Experimental results show that KG-based RAG has only a limited impact on Japanese medical QA using small-scale open-source LLMs. Further case studies reveal that the effectiveness of the RAG is sensitive to the quality and relevance of the external retrieved content. These findings offer valuable insights into the challenges and potential of applying RAG in Japanese medical QA, while also serving as a reference for other low-resource languages.
Authors: Alan Dao (Gia Tuan Dao), Thinh Le
Abstract: Retrieval-Augmented Generation (RAG) improves Large Language Model (LLM) performance on knowledge-intensive tasks but depends heavily on initial search query quality. Current methods, often using Reinforcement Learning (RL), typically focus on query formulation or reasoning over results, without explicitly encouraging persistence after a failed search. We introduce ReZero (Retry-Zero), a novel RL framework that directly rewards the act of retrying a search query following an initial unsuccessful attempt. This incentivizes the LLM to explore alternative queries rather than prematurely halting. ReZero demonstrates significant improvement, achieving 46.88% accuracy compared to a 25% baseline. By rewarding persistence, ReZero enhances LLM robustness in complex information-seeking scenarios where initial queries may prove insufficient.
Authors: Jinwu Hu, Wei Zhang, Yufeng Wang, Yu Hu, Bin Xiao, Mingkui Tan, Qing Du
Abstract: Large Language Models (LLMs) have shown outstanding performance across a variety of tasks, partly due to advanced prompting techniques. However, these techniques often require lengthy prompts, which increase computational costs and can hinder performance because of the limited context windows of LLMs. While prompt compression is a straightforward solution, existing methods confront the challenges of retaining essential information, adapting to context changes, and remaining effective across different tasks. To tackle these issues, we propose a task-agnostic method called Dynamic Compressing Prompts (LLM-DCP). Our method reduces the number of prompt tokens while aiming to preserve the performance as much as possible. We model prompt compression as a Markov Decision Process (MDP), enabling the DCP-Agent to sequentially remove redundant tokens by adapting to dynamic contexts and retaining crucial content. We develop a reward function for training the DCP-Agent that balances the compression rate, the quality of the LLM output, and the retention of key information. This allows for prompt token reduction without needing an external black-box LLM. Inspired by the progressive difficulty adjustment in curriculum learning, we introduce a Hierarchical Prompt Compression (HPC) training strategy that gradually increases the compression difficulty, enabling the DCP-Agent to learn an effective compression method that maintains information integrity. Experiments demonstrate that our method outperforms state-of-the-art techniques, especially at higher compression rates. The code for our approach will be available at https://github.com/Fhujinwu/DCP.
Authors: Sukannya Purkayastha, Zhuang Li, Anne Lauscher, Lizhen Qu, Iryna Gurevych
Abstract: Peer review is a cornerstone of quality control in scientific publishing. With the increasing workload, the unintended use of `quick' heuristics, referred to as lazy thinking, has emerged as a recurring issue compromising review quality. Automated methods to detect such heuristics can help improve the peer-reviewing process. However, there is limited NLP research on this issue, and no real-world dataset exists to support the development of detection tools. This work introduces LazyReview, a dataset of peer-review sentences annotated with fine-grained lazy thinking categories. Our analysis reveals that Large Language Models (LLMs) struggle to detect these instances in a zero-shot setting. However, instruction-based fine-tuning on our dataset significantly boosts performance by 10-20 performance points, highlighting the importance of high-quality training data. Furthermore, a controlled experiment demonstrates that reviews revised with lazy thinking feedback are more comprehensive and actionable than those written without such feedback. We will release our dataset and the enhanced guidelines that can be used to train junior reviewers in the community. (Code available here: https://github.com/UKPLab/arxiv2025-lazy-review)
Authors: Efthymios Georgiou, Vassilis Katsouros, Yannis Avrithis, Alexandros Potamianos
Abstract: While multimodal fusion has been extensively studied in Multimodal Sentiment Analysis (MSA), the role of fusion depth and multimodal capacity allocation remains underexplored. In this work, we position fusion depth, scalability, and dedicated multimodal capacity as primary factors for effective fusion. We introduce DeepMLF, a novel multimodal language model (LM) with learnable tokens tailored toward deep fusion. DeepMLF leverages an audiovisual encoder and a pretrained decoder LM augmented with multimodal information across its layers. We append learnable tokens to the LM that: 1) capture modality interactions in a controlled fashion and 2) preserve independent information flow for each modality. These fusion tokens gather linguistic information via causal self-attention in LM Blocks and integrate with audiovisual information through cross-attention MM Blocks. Serving as dedicated multimodal capacity, this design enables progressive fusion across multiple layers, providing depth in the fusion process. Our training recipe combines modality-specific losses and language modelling loss, with the decoder LM tasked to predict ground truth polarity. Across three MSA benchmarks with varying dataset characteristics, DeepMLF achieves state-of-the-art performance. Our results confirm that deeper fusion leads to better performance, with optimal fusion depths (5-7) exceeding those of existing approaches. Additionally, our analysis on the number of fusion tokens reveals that small token sets ($\sim$20) achieve optimal performance. We examine the importance of representation learning order (fusion curriculum) through audiovisual encoder initialization experiments. Our ablation studies demonstrate the superiority of the proposed fusion design and gating while providing a holistic examination of DeepMLF's scalability to LLMs, and the impact of each training objective and embedding regularization.
Authors: Ren\'e Peinl
Abstract: This study examines how Large Language Models (LLMs) can reduce biases in text-to-image generation systems by modifying user prompts. We define bias as a model's unfair deviation from population statistics given neutral prompts. Our experiments with Stable Diffusion XL, 3.5 and Flux demonstrate that LLM-modified prompts significantly increase image diversity and reduce bias without the need to change the image generators themselves. While occasionally producing results that diverge from original user intent for elaborate prompts, this approach generally provides more varied interpretations of underspecified requests rather than superficial variations. The method works particularly well for less advanced image generators, though limitations persist for certain contexts like disability representation. All prompts and generated images are available at https://iisys-hof.github.io/llm-prompt-img-gen/
Authors: Ren\'e Peinl, Vincent Tischler
Abstract: Similar to LLMs, the development of vision language models is mainly driven by English datasets and models trained in English and Chinese language, whereas support for other languages, even those considered high-resource languages such as German, remains significantly weaker. In this work we present an analysis of open-weight VLMs on factual knowledge in the German and English language. We disentangle the image-related aspects from the textual ones by analyzing accu-racy with jury-as-a-judge in both prompt languages and images from German and international contexts. We found that for celebrities and sights, VLMs struggle because they are lacking visual cognition of German image contents. For animals and plants, the tested models can often correctly identify the image contents ac-cording to the scientific name or English common name but fail in German lan-guage. Cars and supermarket products were identified equally well in English and German images across both prompt languages.
Authors: Laura De Grazia, Pol Pastells, Mauro V\'azquez Chas, Desmond Elliott, Danae S\'anchez Villegas, Mireia Farr\'us, Mariona Taul\'e
Abstract: Sexism is generally defined as prejudice and discrimination based on sex or gender, affecting every sector of society, from social institutions to relationships and individual behavior. Social media platforms amplify the impact of sexism by conveying discriminatory content not only through text but also across multiple modalities, highlighting the critical need for a multimodal approach to the analysis of sexism online. With the rise of social media platforms where users share short videos, sexism is increasingly spreading through video content. Automatically detecting sexism in videos is a challenging task, as it requires analyzing the combination of verbal, audio, and visual elements to identify sexist content. In this study, (1) we introduce MuSeD, a new Multimodal Spanish dataset for Sexism Detection consisting of $\approx$ 11 hours of videos extracted from TikTok and BitChute; (2) we propose an innovative annotation framework for analyzing the contribution of textual and multimodal labels in the classification of sexist and non-sexist content; and (3) we evaluate a range of large language models (LLMs) and multimodal LLMs on the task of sexism detection. We find that visual information plays a key role in labeling sexist content for both humans and models. Models effectively detect explicit sexism; however, they struggle with implicit cases, such as stereotypes, instances where annotators also show low agreement. This highlights the inherent difficulty of the task, as identifying implicit sexism depends on the social and cultural context.
Authors: Ej Zhou, Weiming Lu
Abstract: Social bias in language models can potentially exacerbate social inequalities. Despite it having garnered wide attention, most research focuses on English data. In a low-resource scenario, the models often perform worse due to insufficient training data. This study aims to leverage high-resource language corpora to evaluate bias and experiment with debiasing methods in low-resource languages. We evaluated the performance of recent multilingual models in five languages: English (\textsc{eng}), Chinese (\textsc{zho}), Russian (\textsc{rus}), Indonesian (\textsc{ind}) and Thai (\textsc{tha}), and analyzed four bias dimensions: \textit{gender}, \textit{religion}, \textit{nationality}, and \textit{race-color}. By constructing multilingual bias evaluation datasets, this study allows fair comparisons between models across languages. We have further investigated three debiasing methods-\texttt{CDA}, \texttt{Dropout}, \texttt{SenDeb}-and demonstrated that debiasing methods from high-resource languages can be effectively transferred to low-resource ones, providing actionable insights for fairness research in multilingual NLP.
Authors: Minjie Zou, Sahana Srinivasan, Thaddaeus Wai Soon Lo, Ke Zou, Gabriel Dawei Yang, Xuguang Ai, Hyunjae Kim, Maxwell Singer, Fares Antaki, Kelvin Li, Robert Chang, Marcus Tan, David Ziyou Chen, Dianbo Liu, Qingyu Chen, Yih Chung Tham
Abstract: Recent advances in reasoning-focused large language models (LLMs) mark a shift from general LLMs toward models designed for complex decision-making, a crucial aspect in medicine. However, their performance in specialized domains like ophthalmology remains underexplored. This study comprehensively evaluated and compared the accuracy and reasoning capabilities of four newly developed reasoning-focused LLMs, namely DeepSeek-R1, OpenAI o1, o3-mini, and Gemini 2.0 Flash-Thinking. Each model was assessed using 5,888 multiple-choice ophthalmology exam questions from the MedMCQA dataset in zero-shot setting. Quantitative evaluation included accuracy, Macro-F1, and five text-generation metrics (ROUGE-L, METEOR, BERTScore, BARTScore, and AlignScore), computed against ground-truth reasonings. Average inference time was recorded for a subset of 100 randomly selected questions. Additionally, two board-certified ophthalmologists qualitatively assessed clarity, completeness, and reasoning structure of responses to differential diagnosis questions.O1 (0.902) and DeepSeek-R1 (0.888) achieved the highest accuracy, with o1 also leading in Macro-F1 (0.900). The performance of models across the text-generation metrics varied: O3-mini excelled in ROUGE-L (0.151), o1 in METEOR (0.232), DeepSeek-R1 and o3-mini tied for BERTScore (0.673), DeepSeek-R1 (-4.105) and Gemini 2.0 Flash-Thinking (-4.127) performed best in BARTScore, while o3-mini (0.181) and o1 (0.176) led AlignScore. Inference time across the models varied, with DeepSeek-R1 being slowest (40.4 seconds) and Gemini 2.0 Flash-Thinking fastest (6.7 seconds). Qualitative evaluation revealed that DeepSeek-R1 and Gemini 2.0 Flash-Thinking tended to provide detailed and comprehensive intermediate reasoning, whereas o1 and o3-mini displayed concise and summarized justifications.
Authors: Guocong Li, Weize Liu, Yihang Wu, Ping Wang, Shuaihan Huang, Hongxia Xu, Jian Wu
Abstract: Large language models (LLMs) exhibit excellent performance in natural language processing (NLP), but remain highly sensitive to the quality of input queries, especially when these queries contain misleading or inaccurate information. Existing methods focus on correcting the output, but they often overlook the potential of improving the ability of LLMs to detect and correct misleading content in the input itself. In this paper, we propose a novel three-stage fine-tuning method that enhances the ability of LLMs to detect and correct misleading information in the input, further improving response accuracy and reducing hallucinations. Specifically, the three stages include (1) training LLMs to identify misleading information, (2) training LLMs to correct the misleading information using built-in or external knowledge, and (3) training LLMs to generate accurate answers based on the corrected queries. To evaluate our method, we conducted experiments on three datasets for the hallucination detection task and the question answering (QA) task, as well as two datasets containing misleading information that we constructed. The experimental results demonstrate that our method significantly improves the accuracy and factuality of LLM responses, while also enhancing the ability to detect hallucinations and reducing the generation of hallucinations in the output, particularly when the query contains misleading information. We will publicly release our code upon acceptance.
Authors: Joshua Otten, Antonios Anastasopoulos, Kevin Moran
Abstract: Python is one of the most commonly used programming languages in industry and education. Its English keywords and built-in functions/modules allow it to come close to pseudo-code in terms of its readability and ease of writing. However, those who do not speak English may not experience these advantages. In fact, they may even be hindered in their ability to understand Python code, as the English nature of its terms creates an additional layer of overhead. To that end, we introduce the task of automatically translating Python's natural modality (keywords, error types, identifiers, etc.) into other human languages. This presents a unique challenge, considering the abbreviated nature of these forms, as well as potential untranslatability of advanced mathematical/programming concepts across languages. We therefore create an automated pipeline to translate Python into other human languages, comparing strategies using machine translation and large language models. We then use this pipeline to acquire translations from five common Python libraries (pytorch, pandas, tensorflow, numpy, and random) in seven languages, and do a quality test on a subset of these terms in French, Greek, and Bengali. We hope this will provide a clearer path forward towards creating a universal Python, accessible to anyone regardless of nationality or language background.
Authors: Hao Liu, Lijun He, Jiaxi Liang, Zhihan Ren, Fan Li
Abstract: Multimodal Aspect-Based Sentiment Analysis (MABSA) seeks to extract fine-grained information from image-text pairs to identify aspect terms and determine their sentiment polarity. However, existing approaches often fall short in simultaneously addressing three core challenges: Sentiment Cue Perception (SCP), Multimodal Information Misalignment (MIM), and Semantic Noise Elimination (SNE). To overcome these limitations, we propose DASCO (\textbf{D}ependency Structure \textbf{A}ugmented \textbf{Sco}ping Framework), a fine-grained scope-oriented framework that enhances aspect-level sentiment reasoning by leveraging dependency parsing trees. First, we designed a multi-task pretraining strategy for MABSA on our base model, combining aspect-oriented enhancement, image-text matching, and aspect-level sentiment-sensitive cognition. This improved the model's perception of aspect terms and sentiment cues while achieving effective image-text alignment, addressing key challenges like SCP and MIM. Furthermore, we incorporate dependency trees as syntactic branch combining with semantic branch, guiding the model to selectively attend to critical contextual elements within a target-specific scope while effectively filtering out irrelevant noise for addressing SNE problem. Extensive experiments on two benchmark datasets across three subtasks demonstrate that DASCO achieves state-of-the-art performance in MABSA, with notable gains in JMASA (+3.1\% F1 and +5.4\% precision on Twitter2015).
Authors: Zhihao Xu, Yongqi Tong, Xin Zhang, Jun Zhou, Xiting Wang
Abstract: Multi-objective preference alignment in language models often encounters a challenging trade-off: optimizing for one human preference (e.g., helpfulness) frequently compromises others (e.g., harmlessness) due to the inherent conflicts between competing objectives. While prior work mainly focuses on algorithmic solutions, we explore a novel data-driven approach to uncover the types of data that can effectively mitigate these conflicts. Specifically, we propose the concept of Reward Consistency (RC), which identifies samples that align with multiple preference objectives, thereby reducing conflicts during training. Through gradient-based analysis, we demonstrate that RC-compliant samples inherently constrain performance degradation during multi-objective optimization. Building on these insights, we further develop Reward Consistency Sampling, a framework that automatically constructs preference datasets that effectively mitigate conflicts during multi-objective alignment. Our generated data achieves an average improvement of 13.37% in both the harmless rate and helpfulness win rate when optimizing harmlessness and helpfulness, and can consistently resolve conflicts in varying multi-objective scenarios.
Authors: Lucio La Cava, Andrea Tagarelli
Abstract: Open Large Language Models (OLLMs) are increasingly leveraged in generative AI applications, posing new challenges for detecting their outputs. We propose OpenTuringBench, a new benchmark based on OLLMs, designed to train and evaluate machine-generated text detectors on the Turing Test and Authorship Attribution problems. OpenTuringBench focuses on a representative set of OLLMs, and features a number of challenging evaluation tasks, including human/machine-manipulated texts, out-of-domain texts, and texts from previously unseen models. We also provide OTBDetector, a contrastive learning framework to detect and attribute OLLM-based machine-generated texts. Results highlight the relevance and varying degrees of difficulty of the OpenTuringBench tasks, with our detector achieving remarkable capabilities across the various tasks and outperforming most existing detectors. Resources are available on the OpenTuringBench Hugging Face repository at https://huggingface.co/datasets/MLNTeam-Unical/OpenTuringBench
URLs: https://huggingface.co/datasets/MLNTeam-Unical/OpenTuringBench
Authors: Wang Bill Zhu, Tianqi Chen, Ching Ying Lin, Jade Law, Mazen Jizzini, Jorge J. Nieva, Ruishan Liu, Robin Jia
Abstract: Cancer patients are increasingly turning to large language models (LLMs) as a new form of internet search for medical information, making it critical to assess how well these models handle complex, personalized questions. However, current medical benchmarks focus on medical exams or consumer-searched questions and do not evaluate LLMs on real patient questions with detailed clinical contexts. In this paper, we first evaluate LLMs on cancer-related questions drawn from real patients, reviewed by three hematology oncology physicians. While responses are generally accurate, with GPT-4-Turbo scoring 4.13 out of 5, the models frequently fail to recognize or address false presuppositions in the questions-posing risks to safe medical decision-making. To study this limitation systematically, we introduce Cancer-Myth, an expert-verified adversarial dataset of 585 cancer-related questions with false presuppositions. On this benchmark, no frontier LLM -- including GPT-4o, Gemini-1.Pro, and Claude-3.5-Sonnet -- corrects these false presuppositions more than 30% of the time. Even advanced medical agentic methods do not prevent LLMs from ignoring false presuppositions. These findings expose a critical gap in the clinical reliability of LLMs and underscore the need for more robust safeguards in medical AI systems.
Authors: Juan Diego Rodriguez, Wenxuan Ding, Katrin Erk, Greg Durrett
Abstract: Although large language models (LLMs) have become generally more capable and accurate across many tasks, some fundamental sources of unreliability remain in their behavior. One key limitation is their inconsistency at reporting the the same information when prompts are changed. In this paper, we consider the discrepancy between a model's generated answer and their own verification of that answer, the generator-validator gap. We define this gap in a more stringent way than prior work: we expect correlation of scores from a generator and a validator over the entire set of candidate answers. We show that according to this measure, a large gap exists in various settings, including question answering, lexical semantics tasks, and next-word prediction. We then propose RankAlign, a ranking-based training method, and show that it significantly closes the gap by 31.8% on average, surpassing all baseline methods. Moreover, this approach generalizes well to out-of-domain tasks and lexical items.
Authors: Ali Taghibakhshi, Sharath Turuvekere Sreenivas, Saurav Muralidharan, Marcin Chochowski, Yashaswi Karnati, Raviraj Joshi, Ameya Sunil Mahabaleshwarkar, Zijia Chen, Yoshi Suhara, Oluwatobi Olabiyi, Daniel Korzekwa, Mostofa Patwary, Mohammad Shoeybi, Jan Kautz, Bryan Catanzaro, Ashwath Aithal, Nima Tajbakhsh, Pavlo Molchanov
Abstract: Hybrid LLM architectures that combine Attention and State Space Models (SSMs) achieve state-of-the-art accuracy and runtime performance. Recent work has demonstrated that applying compression and distillation to Attention-only models yields smaller, more accurate models at a fraction of the training cost. In this work, we explore the effectiveness of compressing Hybrid architectures. We introduce a novel group-aware pruning strategy that preserves the structural integrity of SSM blocks and their sequence modeling capabilities. Furthermore, we demonstrate the necessity of such SSM pruning to achieve improved accuracy and inference speed compared to traditional approaches. Our compression recipe combines SSM, FFN, embedding dimension, and layer pruning, followed by knowledge distillation-based retraining, similar to the MINITRON technique. Using this approach, we compress the Nemotron-H 8B Hybrid model down to 4B parameters with up to 40x fewer training tokens. The resulting model surpasses the accuracy of similarly-sized models while achieving 2x faster inference, significantly advancing the Pareto frontier.
Authors: Quanyu Long, Jianda Chen, Zhengyuan Liu, Nancy F. Chen, Wenya Wang, Sinno Jialin Pan
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities across numerous tasks, yet they often rely on external context to handle complex tasks. While retrieval-augmented frameworks traditionally focus on selecting top-ranked documents in a single pass, many real-world scenarios demand compositional retrieval, where multiple sources must be combined in a coordinated manner. In this work, we propose a tri-encoder sequential retriever that models this process as a Markov Decision Process (MDP), decomposing the probability of retrieving a set of elements into a sequence of conditional probabilities and allowing each retrieval step to be conditioned on previously selected examples. We train the retriever in two stages: first, we efficiently construct supervised sequential data for initial policy training; we then refine the policy to align with the LLM's preferences using a reward grounded in the structural correspondence of generated programs. Experimental results show that our method consistently and significantly outperforms baselines, underscoring the importance of explicitly modeling inter-example dependencies. These findings highlight the potential of compositional retrieval for tasks requiring multiple pieces of evidence or examples.
Authors: Xue Zhang, Songming Zhang, Yunlong Liang, Fandong Meng, Yufeng Chen, Jinan Xu, Jie Zhou
Abstract: Knowledge distillation (KD) is a promising solution to compress large language models (LLMs) by transferring their knowledge to smaller models. During this process, white-box KD methods usually minimize the distance between the output distributions of the teacher model and the student model to transfer more information. However, we reveal that the current white-box KD framework exhibits two limitations: a) bridging probability distributions from different output spaces will limit the similarity between the teacher model and the student model; b) this framework cannot be applied to LLMs with different vocabularies. One of the root causes for these limitations is that the distributions from the teacher and the student for KD are output by different prediction heads, which yield distributions in different output spaces and dimensions. Therefore, in this paper, we propose a dual-space knowledge distillation (DSKD) framework that unifies the prediction heads of the teacher and the student models for KD. Specifically, we first introduce two projectors with ideal initialization to project the teacher/student hidden states into the student/teacher representation spaces. After this, the hidden states from different models can share the same head and unify the output spaces of the distributions. Furthermore, we develop an exact token alignment (ETA) algorithm to align the same tokens in two differently-tokenized sequences. Based on the above, our DSKD framework is a general KD framework that supports both off-policy and on-policy KD, and KD between any two LLMs regardless of their vocabularies. Extensive experiments on instruction-following, mathematical reasoning, and code generation benchmarks show that DSKD significantly outperforms existing methods based on the current white-box KD framework and surpasses other cross-tokenizer KD methods for LLMs with different vocabularies.
Authors: Maria Teleki, Xiangjue Dong, Haoran Liu, James Caverlee
Abstract: Masculine defaults are widely recognized as a significant type of gender bias, but they are often unseen as they are under-researched. Masculine defaults involve three key parts: (i) the cultural context, (ii) the masculine characteristics or behaviors, and (iii) the reward for, or simply acceptance of, those masculine characteristics or behaviors. In this work, we study discourse-based masculine defaults, and propose a twofold framework for (i) the large-scale discovery and analysis of gendered discourse words in spoken content via our Gendered Discourse Correlation Framework (GDCF); and (ii) the measurement of the gender bias associated with these gendered discourse words in LLMs via our Discourse Word-Embedding Association Test (D-WEAT). We focus our study on podcasts, a popular and growing form of social media, analyzing 15,117 podcast episodes. We analyze correlations between gender and discourse words -- discovered via LDA and BERTopic -- to automatically form gendered discourse word lists. We then study the prevalence of these gendered discourse words in domain-specific contexts, and find that gendered discourse-based masculine defaults exist in the domains of business, technology/politics, and video games. Next, we study the representation of these gendered discourse words from a state-of-the-art LLM embedding model from OpenAI, and find that the masculine discourse words have a more stable and robust representation than the feminine discourse words, which may result in better system performance on downstream tasks for men. Hence, men are rewarded for their discourse patterns with better system performance by one of the state-of-the-art language models -- and this embedding disparity is a representational harm and a masculine default.
Authors: Leon Guertler, Bobby Cheng, Simon Yu, Bo Liu, Leshem Choshen, Cheston Tan
Abstract: TextArena is an open-source collection of competitive text-based games for training and evaluation of agentic behavior in Large Language Models (LLMs). It spans 57+ unique environments (including single-player, two-player, and multi-player setups) and allows for easy evaluation of model capabilities via an online-play system (against humans and other submitted models) with real-time TrueSkill scores. Traditional benchmarks rarely assess dynamic social skills such as negotiation, theory of mind, and deception, creating a gap that TextArena addresses. Designed with research, community and extensibility in mind, TextArena emphasizes ease of adding new games, adapting the framework, testing models, playing against the models, and training models. Detailed documentation of environments, games, leaderboard, and examples are available on https://github.com/LeonGuertler/TextArena and https://www.textarena.ai/.
URLs: https://github.com/LeonGuertler/TextArena, https://www.textarena.ai/.
Authors: Zhiwei He, Tian Liang, Jiahao Xu, Qiuzhi Liu, Xingyu Chen, Yue Wang, Linfeng Song, Dian Yu, Zhenwen Liang, Wenxuan Wang, Zhuosheng Zhang, Rui Wang, Zhaopeng Tu, Haitao Mi, Dong Yu
Abstract: The capacity for complex mathematical reasoning is a key benchmark for artificial intelligence. While reinforcement learning (RL) applied to LLMs shows promise, progress is significantly hindered by the lack of large-scale training data that is sufficiently challenging, possesses verifiable answer formats suitable for RL, and is free from contamination with evaluation benchmarks. To address these limitations, we introduce DeepMath-103K, a new, large-scale dataset comprising approximately 103K mathematical problems, specifically designed to train advanced reasoning models via RL. DeepMath-103K is curated through a rigorous pipeline involving source analysis, stringent decontamination against numerous benchmarks, and filtering for high difficulty (primarily Levels 5-9), significantly exceeding existing open resources in challenge. Each problem includes a verifiable final answer, enabling rule-based RL, and three distinct R1-generated solutions suitable for diverse training paradigms like supervised fine-tuning or distillation. Spanning a wide range of mathematical topics, DeepMath-103K promotes the development of generalizable reasoning. We demonstrate that models trained on DeepMath-103K achieve significant improvements on challenging mathematical benchmarks, validating its effectiveness. We release DeepMath-103K publicly to facilitate community progress in building more capable AI reasoning systems: https://github.com/zwhe99/DeepMath.
Authors: Gabriel Bo, Marc Bernardino, Justin Gu
Abstract: We explore the potential of integrating learnable and interpretable modules--specifically Kolmogorov-Arnold Networks (KAN) and graph-based representations--within a pre-trained GPT-2 model to enhance multi-task learning accuracy. Motivated by the recent surge in using KAN and graph attention (GAT) architectures in chain-of-thought (CoT) models and debates over their benefits compared to simpler architectures like MLPs, we begin by enhancing a standard self-attention transformer using Low-Rank Adaptation (LoRA), fine-tuning hyperparameters, and incorporating L2 regularization. This approach yields significant improvements. To further boost interpretability and richer representations, we develop two variants that attempt to improve the standard KAN and GAT: Graph LoRA and Hybrid-KAN LoRA (Learnable GPT). However, systematic evaluations reveal that neither variant outperforms the optimized LoRA-enhanced transformer, which achieves 55.249% accuracy on the SST test set, 99.18% on the CFIMDB dev set, and 89.9% paraphrase detection test accuracy. On sonnet generation, we get a CHRF score of 42.097. These findings highlight that efficient parameter adaptation via LoRA remains the most effective strategy for our tasks: sentiment analysis, paraphrase detection, and sonnet generation.
Authors: Ning Li, Jingran Zhang, Justin Cui
Abstract: Large language models (LLMs) have demonstrated remarkable effectiveness in completing various tasks such as reasoning, translation, and question answering. However the issue of factual incorrect content in LLM-generated responses remains a persistent challenge. In this study, we evaluate both proprietary and open-source LLMs on their ability to respond with relevant research papers and accurate links to articles hosted on the arXiv platform, based on high level prompts. To facilitate this evaluation, we introduce arXivBench, a benchmark specifically designed to assess LLM performance across eight major subject categories on arXiv and five subfields within computer science, one of the most popular categories among them. Our findings reveal a concerning accuracy of LLM-generated responses depending on the subject, with some subjects experiencing significantly lower accuracy than others. Notably, Claude-3.5-Sonnet exhibits a substantial advantage in generating both relevant and accurate responses. And interestingly, most LLMs achieve a much higher accuracy in the Artificial Intelligence sub-field than other sub-fields. This benchmark provides a standardized tool for evaluating the reliability of LLM-generated scientific responses, promoting more dependable use of LLMs in academic and research environments. Our code is open-sourced at https://github.com/arxivBenchLLM/arXivBench and our dataset is available on huggingface at https://huggingface.co/datasets/arXivBenchLLM/arXivBench.
URLs: https://github.com/arxivBenchLLM/arXivBench, https://huggingface.co/datasets/arXivBenchLLM/arXivBench.
Authors: Zulun Zhu, Tiancheng Huang, Kai Wang, Junda Ye, Xinghe Chen, Siqiang Luo
Abstract: Large language models (LLMs) struggle with the factual error during inference due to the lack of sufficient training data and the most updated knowledge, leading to the hallucination problem. Retrieval-Augmented Generation (RAG) has gained attention as a promising solution to address the limitation of LLMs, by retrieving relevant information from external source to generate more accurate answers to the questions. Given the pervasive presence of structured knowledge in the external source, considerable strides in RAG have been made to employ the techniques related to graphs and achieve more complex reasoning based on the topological information between knowledge entities. However, there is currently neither unified review examining the diverse roles of graphs in RAG, nor a comprehensive resource to help researchers navigate and contribute to this evolving field. This survey offers a novel perspective on the functionality of graphs within RAG and their impact on enhancing performance across a wide range of graph-structured data. It provides a detailed breakdown of the roles that graphs play in RAG, covering database construction, algorithms, pipelines, and tasks. Finally, it identifies current challenges and outline future research directions, aiming to inspire further developments in this field. Our graph-centered analysis highlights the commonalities and differences in existing methods, setting the stage for future researchers in areas such as graph learning, database systems, and natural language processing.
Authors: Shravika Mittal, Darshi Shah, Shin Won Do, Mai ElSherief, Tanushree Mitra, Munmun De Choudhury
Abstract: Widespread stigma, both in the offline and online spaces, acts as a barrier to harm reduction efforts in the context of opioid use disorder (OUD). This stigma is prominently directed towards clinically approved medications for addiction treatment (MAT), people with the condition, and the condition itself. Given the potential of artificial intelligence based technologies in promoting health equity, and facilitating empathic conversations, this work examines whether large language models (LLMs) can help abate OUD-related stigma in online communities. To answer this, we conducted a series of pre-registered randomized controlled experiments, where participants read LLM-generated, human-written, or no responses to help seeking OUD-related content in online communities. The experiment was conducted under two setups, i.e., participants read the responses either once (N = 2,141), or repeatedly for 14 days (N = 107). We found that participants reported the least stigmatized attitudes toward MAT after consuming LLM-generated responses under both the setups. This study offers insights into strategies that can foster inclusive online discourse on OUD, e.g., based on our findings LLMs can be used as an education-based intervention to promote positive attitudes and increase people's propensity toward MAT.
Authors: Minh-Anh Nguyen, Dung D. Le
Abstract: Language representation learning has emerged as a promising approach for sequential recommendation, thanks to its ability to learn generalizable representations. However, despite its advantages, this approach still struggles with data sparsity and a limited understanding of common-sense user preferences. To address these limitations, we propose $\textbf{JEPA4Rec}$, a framework that combines $\textbf{J}$oint $\textbf{E}$mbedding $\textbf{P}$redictive $\textbf{A}$rchitecture with language modeling of item textual descriptions. JEPA4Rec captures semantically rich and transferable representations, improving recommendation performance and reducing reliance on large-scale pre-training data. Specifically, JEPA4Rec represents items as text sentences by flattening descriptive information such as $\textit{title, category}$, and other attributes. To encode these sentences, we employ a bidirectional Transformer encoder with modified embedding layers tailored for capturing item information in recommendation datasets. We apply masking to text sentences and use them to predict the representations of the unmasked sentences, helping the model learn generalizable item embeddings. To further improve recommendation performance and language understanding, we employ a two-stage training strategy incorporating self-supervised learning losses. Experiments on six real-world datasets demonstrate that JEPA4Rec consistently outperforms state-of-the-art methods, particularly in cross-domain, cross-platform, and low-resource scenarios.
Authors: Yijun Liang, Ming Li, Chenrui Fan, Ziyue Li, Dang Nguyen, Kwesi Cobbina, Shweta Bhardwaj, Jiuhai Chen, Fuxiao Liu, Tianyi Zhou
Abstract: Color plays an important role in human perception and usually provides critical clues in visual reasoning. However, it is unclear whether and how vision-language models (VLMs) can perceive, understand, and leverage color as humans. This paper introduces ColorBench, an innovative benchmark meticulously crafted to assess the capabilities of VLMs in color understanding, including color perception, reasoning, and robustness. By curating a suite of diverse test scenarios, with grounding in real applications, ColorBench evaluates how these models perceive colors, infer meanings from color-based cues, and maintain consistent performance under varying color transformations. Through an extensive evaluation of 32 VLMs with varying language models and vision encoders, our paper reveals some undiscovered findings: (i) The scaling law (larger models are better) still holds on ColorBench, while the language model plays a more important role than the vision encoder. (ii) However, the performance gaps across models are relatively small, indicating that color understanding has been largely neglected by existing VLMs. (iii) CoT reasoning improves color understanding accuracies and robustness, though they are vision-centric tasks. (iv) Color clues are indeed leveraged by VLMs on ColorBench but they can also mislead models in some tasks. These findings highlight the critical limitations of current VLMs and underscore the need to enhance color comprehension. Our ColorBenchcan serve as a foundational tool for advancing the study of human-level color understanding of multimodal AI.
Authors: Yuhang Yao, Haixin Wang, Yibo Chen, Jiawen Wang, Min Chang Jordan Ren, Bosheng Ding, Salman Avestimehr, Chaoyang He
Abstract: AI Agents powered by Large Language Models are transforming the world through enormous applications. A super agent has the potential to fulfill diverse user needs, such as summarization, coding, and research, by accurately understanding user intent and leveraging the appropriate tools to solve tasks. However, to make such an agent viable for real-world deployment and accessible at scale, significant optimizations are required to ensure high efficiency and low cost. This paper presents a design of the Super Agent System. Upon receiving a user prompt, the system first detects the intent of the user, then routes the request to specialized task agents with the necessary tools or automatically generates agentic workflows. In practice, most applications directly serve as AI assistants on edge devices such as phones and robots. As different language models vary in capability and cloud-based models often entail high computational costs, latency, and privacy concerns, we then explore the hybrid mode where the router dynamically selects between local and cloud models based on task complexity. Finally, we introduce the blueprint of an on-device super agent enhanced with cloud. With advances in multi-modality models and edge hardware, we envision that most computations can be handled locally, with cloud collaboration only as needed. Such architecture paves the way for super agents to be seamlessly integrated into everyday life in the near future.
Authors: \'Eva Sz\'ekely (Michaela), J\=ura Miniota (Michaela), M\'i\v{s}a (Michaela), Hejn\'a
Abstract: The growing prevalence of conversational voice interfaces, powered by developments in both speech and language technologies, raises important questions about their influence on human communication. While written communication can signal identity through lexical and stylistic choices, voice-based interactions inherently amplify socioindexical elements - such as accent, intonation, and speech style - which more prominently convey social identity and group affiliation. There is evidence that even passive media such as television is likely to influence the audience's linguistic patterns. Unlike passive media, conversational AI is interactive, creating a more immersive and reciprocal dynamic that holds a greater potential to impact how individuals speak in everyday interactions. Such heightened influence can be expected to arise from phenomena such as acoustic-prosodic entrainment and linguistic accommodation, which occur naturally during interaction and enable users to adapt their speech patterns in response to the system. While this phenomenon is still emerging, its potential societal impact could provide organisations, movements, and brands with a subtle yet powerful avenue for shaping and controlling public perception and social identity. We argue that the socioindexical influence of AI-generated speech warrants attention and should become a focus of interdisciplinary research, leveraging new and existing methodologies and technologies to better understand its implications.
Authors: Ankit Kumar Shaw (Tsinghua University), Kun Jiang (Tsinghua University), Tuopu Wen (Tsinghua University), Chandan Kumar Sah (Beihang University), Yining Shi (Tsinghua University), Mengmeng Yang (Tsinghua University), Diange Yang (Tsinghua University), Xiaoli Lian (Beihang University)
Abstract: The rapid growth of intelligent connected vehicles (ICVs) and integrated vehicle-road-cloud systems has increased the demand for accurate, real-time HD map updates. However, ensuring map reliability remains challenging due to inconsistencies in crowdsourced data, which suffer from motion blur, lighting variations, adverse weather, and lane marking degradation. This paper introduces CleanMAP, a Multimodal Large Language Model (MLLM)-based distillation framework designed to filter and refine crowdsourced data for high-confidence HD map updates. CleanMAP leverages an MLLM-driven lane visibility scoring model that systematically quantifies key visual parameters, assigning confidence scores (0-10) based on their impact on lane detection. A novel dynamic piecewise confidence-scoring function adapts scores based on lane visibility, ensuring strong alignment with human evaluations while effectively filtering unreliable data. To further optimize map accuracy, a confidence-driven local map fusion strategy ranks and selects the top-k highest-scoring local maps within an optimal confidence range (best score minus 10%), striking a balance between data quality and quantity. Experimental evaluations on a real-world autonomous vehicle dataset validate CleanMAP's effectiveness, demonstrating that fusing the top three local maps achieves the lowest mean map update error of 0.28m, outperforming the baseline (0.37m) and meeting stringent accuracy thresholds (<= 0.32m). Further validation with real-vehicle data confirms 84.88% alignment with human evaluators, reinforcing the model's robustness and reliability. This work establishes CleanMAP as a scalable and deployable solution for crowdsourced HD map updates, ensuring more precise and reliable autonomous navigation. The code will be available at https://Ankit-Zefan.github.io/CleanMap/
Authors: Ming Li, Yanhong Li, Ziyue Li, Tianyi Zhou
Abstract: As the post-training of large language models (LLMs) advances from instruction-following to complex reasoning tasks, understanding how different data affect finetuning dynamics remains largely unexplored. In this paper, we present a spectral analysis of layer-wise gradients induced by low/high-quality instruction and reasoning data for LLM post-training. Our analysis reveals that widely-studied metrics for data evaluation, e.g., IFD, InsTag, Difficulty, and Reward, can be explained and unified by spectral properties computed from gradients' singular value decomposition (SVD). Specifically, higher-quality data are usually associated with lower nuclear norms and higher effective ranks. Notably, effective rank exhibits better robustness and resolution than nuclear norm in capturing subtle quality differences. For example, reasoning data achieves substantially higher effective ranks than instruction data, implying richer gradient structures on more complex tasks. Our experiments also highlight that models within the same family share similar gradient patterns regardless of their sizes, whereas different model families diverge significantly. Providing a unified view on the effects of data quality across instruction and reasoning data, this work illuminates the interplay between data quality and training stability, shedding novel insights into developing better data exploration strategies for post-training.
Authors: Zhichao Xu, Aosong Feng, Yijun Tian, Haibo Ding, Lin Leee Cheong
Abstract: In recent years, dense retrieval has been the focus of information retrieval (IR) research. While effective, dense retrieval produces uninterpretable dense vectors, and suffers from the drawback of large index size. Learned sparse retrieval (LSR) has emerged as promising alternative, achieving competitive retrieval performance while also being able to leverage the classical inverted index data structure for efficient retrieval. However, limited works have explored scaling LSR beyond BERT scale. In this work, we identify two challenges in training large language models (LLM) for LSR: (1) training instability during the early stage of contrastive training; (2) suboptimal performance due to pre-trained LLM's unidirectional attention. To address these challenges, we propose two corresponding techniques: (1) a lightweight adaptation training phase to eliminate training instability; (2) two model variants to enable bidirectional information. With these techniques, we are able to train LSR models with 8B scale LLM, and achieve competitive retrieval performance with reduced index size. Furthermore, we are among the first to analyze the performance-efficiency tradeoff of LLM-based LSR model through the lens of model quantization. Our findings provide insights into adapting LLMs for efficient retrieval modeling.
Authors: Jiseon Kim, Jea Kwon, Luiz Felipe Vecchietti, Alice Oh, Meeyoung Cha
Abstract: Deploying large language models (LLMs) with agency in real-world applications raises critical questions about how these models will behave. In particular, how will their decisions align with humans when faced with moral dilemmas? This study examines the alignment between LLM-driven decisions and human judgment in various contexts of the moral machine experiment, including personas reflecting different sociodemographics. We find that the moral decisions of LLMs vary substantially by persona, showing greater shifts in moral decisions for critical tasks than humans. Our data also indicate an interesting partisan sorting phenomenon, where political persona predominates the direction and degree of LLM decisions. We discuss the ethical implications and risks associated with deploying these models in applications that involve moral decisions.
Authors: Yize Zhang, Tianshu Wang, Sirui Chen, Kun Wang, Xingyu Zeng, Hongyu Lin, Xianpei Han, Le Sun, Chaochao Lu
Abstract: Large language models (LLMs) have demonstrated impressive capabilities and are receiving increasing attention to enhance their reasoning through scaling test--time compute. However, their application in open--ended, knowledge--intensive, complex reasoning scenarios is still limited. Reasoning--oriented methods struggle to generalize to open--ended scenarios due to implicit assumptions of complete world knowledge. Meanwhile, knowledge--augmented reasoning (KAR) methods fail to address two core challenges: 1) error propagation, where errors in early steps cascade through the chain, and 2) verification bottleneck, where the explore--exploit tradeoff arises in multi--branch decision processes. To overcome these limitations, we introduce ARise, a novel framework that integrates risk assessment of intermediate reasoning states with dynamic retrieval--augmented generation (RAG) within a Monte Carlo tree search paradigm. This approach enables effective construction and optimization of reasoning plans across multiple maintained hypothesis branches. Experimental results show that ARise significantly outperforms the state--of--the--art KAR methods by up to 23.10%, and the latest RAG-equipped large reasoning models by up to 25.37%.
Authors: Anna Sofia Lippolis, Andrea Giovanni Nuzzolese, Aldo Gangemi
Abstract: Recent advances in Large Language Models have demonstrated their capabilities across a variety of tasks. However, automatically extracting implicit knowledge from natural language remains a significant challenge, as machines lack active experience with the physical world. Given this scenario, semantic knowledge graphs can serve as conceptual spaces that guide the automated text generation reasoning process to achieve more efficient and explainable results. In this paper, we apply a logic-augmented generation (LAG) framework that leverages the explicit representation of a text through a semantic knowledge graph and applies it in combination with prompt heuristics to elicit implicit analogical connections. This method generates extended knowledge graph triples representing implicit meaning, enabling systems to reason on unlabeled multimodal data regardless of the domain. We validate our work through three metaphor detection and understanding tasks across four datasets, as they require deep analogical reasoning capabilities. The results show that this integrated approach surpasses current baselines, performs better than humans in understanding visual metaphors, and enables more explainable reasoning processes, though still has inherent limitations in metaphor understanding, especially for domain-specific metaphors. Furthermore, we propose a thorough error analysis, discussing issues with metaphorical annotations and current evaluation methods.
Authors: Chang Yang, Ruiyu Wang, Junzhe Jiang, Qi Jiang, Qinggang Zhang, Yanchen Deng, Shuxin Li, Shuyue Hu, Bo Li, Florian T. Pokorny, Xiao Huang, Xinrun Wang
Abstract: Reasoning is the fundamental capability of large language models (LLMs). Due to the rapid progress of LLMs, there are two main issues of current benchmarks: i) these benchmarks can be crushed in a short time (less than 1 year), and ii) these benchmarks may be easily hacked. To handle these issues, we propose the ever-scalingness for building the benchmarks which are uncrushable, unhackable, auto-verifiable and general. This paper presents Nondeterministic Polynomial-time Problem Challenge (NPPC), an ever-scaling reasoning benchmark for LLMs. Specifically, the NPPC has three main modules: i) npgym, which provides a unified interface of 25 well-known NP-complete problems and can generate any number of instances with any levels of complexities, ii) npsolver: which provides a unified interface to evaluate the problem instances with both online and offline models via APIs and local deployments, respectively, and iii) npeval: which provides the comprehensive and ready-to-use tools to analyze the performances of LLMs over different problems, the number of tokens, the aha moments, the reasoning errors and the solution errors. Extensive experiments over widely-used LLMs demonstrate: i) NPPC can successfully decrease the performances of advanced LLMs' performances to below 10%, demonstrating that NPPC is uncrushable, ii) DeepSeek-R1, Claude-3.7-Sonnet, and o1/o3-mini are the most powerful LLMs, where DeepSeek-R1 outperforms Claude-3.7-Sonnet and o1/o3-mini in most NP-complete problems considered, and iii) the numbers of tokens, aha moments in the advanced LLMs, e.g., Claude-3.7-Sonnet and DeepSeek-R1, are observed first to increase and then decrease when the problem instances become more and more difficult. We believe that NPPC is the first ever-scaling reasoning benchmark, serving as the uncrushable and unhackable testbed for LLMs toward artificial general intelligence (AGI).
Authors: Balahari Vignesh Balu, Florian Geissler, Francesco Carella, Joao-Vitor Zacchi, Josef Jiru, Nuria Mata, Reinhard Stolle
Abstract: We study the automated derivation of safety requirements in a self-driving vehicle use case, leveraging LLMs in combination with agent-based retrieval-augmented generation. Conventional approaches that utilise pre-trained LLMs to assist in safety analyses typically lack domain-specific knowledge. Existing RAG approaches address this issue, yet their performance deteriorates when handling complex queries and it becomes increasingly harder to retrieve the most relevant information. This is particularly relevant for safety-relevant applications. In this paper, we propose the use of agent-based RAG to derive safety requirements and show that the retrieved information is more relevant to the queries. We implement an agent-based approach on a document pool of automotive standards and the Apollo case study, as a representative example of an automated driving perception system. Our solution is tested on a data set of safety requirement questions and answers, extracted from the Apollo data. Evaluating a set of selected RAG metrics, we present and discuss advantages of a agent-based approach compared to default RAG methods.
Authors: Xinyi Liu, Xiaoyi Zhang, Ziyun Zhang, Yan Lu
Abstract: Recent advancements in Large Vision-Language Models are accelerating the development of Graphical User Interface (GUI) agents that utilize human-like vision perception capabilities to enhance productivity on digital devices. Compared to approaches predicated on GUI metadata, which are platform-dependent and vulnerable to implementation variations, vision-based approaches offer broader applicability. In this vision-based paradigm, the GUI instruction grounding, which maps user instruction to the location of corresponding element on the given screenshot, remains a critical challenge, particularly due to limited public training dataset and resource-intensive manual instruction data annotation.In this paper, we delve into unexplored challenges in this task including element-to-screen ratio, unbalanced element type, and implicit instruction. To address these challenges, we introduce a large-scale data synthesis pipeline UI-E2I-Synth for generating varying complex instruction datasets using GPT-4o instead of human annotators. Furthermore, we propose a new GUI instruction grounding benchmark UI-I2E-Bench, which is designed to address the limitations of existing benchmarks by incorporating diverse annotation aspects. Our model, trained on the synthesized data, achieves superior performance in GUI instruction grounding, demonstrating the advancements of proposed data synthesis pipeline. The proposed benchmark, accompanied by extensive analyses, provides practical insights for future research in GUI grounding. We will release corresponding artifacts at https://colmon46.github.io/i2e-bench-leaderboard/
Authors: Chaoran Chen, Zhiping Zhang, Bingcan Guo, Shang Ma, Ibrahim Khalilov, Simret A Gebreegziabher, Yanfang Ye, Ziang Xiao, Yaxing Yao, Tianshi Li, Toby Jia-Jun Li
Abstract: A Large Language Model (LLM) powered GUI agent is a specialized autonomous system that performs tasks on the user's behalf according to high-level instructions. It does so by perceiving and interpreting the graphical user interfaces (GUIs) of relevant apps, often visually, inferring necessary sequences of actions, and then interacting with GUIs by executing the actions such as clicking, typing, and tapping. To complete real-world tasks, such as filling forms or booking services, GUI agents often need to process and act on sensitive user data. However, this autonomy introduces new privacy and security risks. Adversaries can inject malicious content into the GUIs that alters agent behaviors or induces unintended disclosures of private information. These attacks often exploit the discrepancy between visual saliency for agents and human users, or the agent's limited ability to detect violations of contextual integrity in task automation. In this paper, we characterized six types of such attacks, and conducted an experimental study to test these attacks with six state-of-the-art GUI agents, 234 adversarial webpages, and 39 human participants. Our findings suggest that GUI agents are highly vulnerable, particularly to contextually embedded threats. Moreover, human users are also susceptible to many of these attacks, indicating that simple human oversight may not reliably prevent failures. This misalignment highlights the need for privacy-aware agent design. We propose practical defense strategies to inform the development of safer and more reliable GUI agents.
Authors: Abitha Thankaraj, Yiding Jiang, J. Zico Kolter, Yonatan Bisk
Abstract: The structure of causal language model training assumes that each token can be accurately predicted from the previous context. This contrasts with humans' natural writing and reasoning process, where goals are typically known before the exact argument or phrasings. While this mismatch has been well studied in the literature, the working assumption has been that architectural changes are needed to address this mismatch. We argue that rearranging and processing the training data sequences can allow models to more accurately imitate the true data-generating process, and does not require any other changes to the architecture or training infrastructure. We demonstrate that this technique, Trelawney, and the inference algorithms derived from it allow us to improve performance on several key benchmarks that span planning, algorithmic reasoning, and story generation tasks. Finally, our method naturally enables the generation of long-term goals at no additional cost. We investigate how using the model's goal-generation capability can further improve planning and reasoning. Additionally, we believe Trelawney could potentially open doors to new capabilities beyond the current language modeling paradigm.
Authors: Wei Xiong, Jiarui Yao, Yuhui Xu, Bo Pang, Lei Wang, Doyen Sahoo, Junnan Li, Nan Jiang, Tong Zhang, Caiming Xiong, Hanze Dong
Abstract: Reinforcement learning (RL) has become a prevailing approach for fine-tuning large language models (LLMs) on complex reasoning tasks. Among recent methods, GRPO stands out for its empirical success in training models such as DeepSeek-R1, yet the sources of its effectiveness remain poorly understood. In this work, we revisit GRPO from a reinforce-like algorithm perspective and analyze its core components. Surprisingly, we find that a simple rejection sampling baseline, RAFT, which trains only on positively rewarded samples, yields competitive performance than GRPO and PPO. Our ablation studies reveal that GRPO's main advantage arises from discarding prompts with entirely incorrect responses, rather than from its reward normalization. Motivated by this insight, we propose Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples. Reinforce-Rej improves KL efficiency and stability, serving as a lightweight yet effective alternative to more complex RL algorithms. We advocate RAFT as a robust and interpretable baseline, and suggest that future advances should focus on more principled designs for incorporating negative samples, rather than relying on them indiscriminately. Our findings provide guidance for future work in reward-based LLM post-training.
Authors: Tianwei Ni, Allen Nie, Sapana Chaudhary, Yao Liu, Huzefa Rangwala, Rasool Fakoor
Abstract: Leveraging inference-time search in large language models has proven effective in further enhancing a trained model's capability to solve complex mathematical and reasoning problems. However, this approach significantly increases computational costs and inference time, as the model must generate and evaluate multiple candidate solutions to identify a viable reasoning path. To address this, we propose an effective approach that integrates search capabilities directly into the model by fine-tuning it using both successful (learning) and failed reasoning paths (forgetting) derived from diverse search methods. While fine-tuning the model with these data might seem straightforward, we identify a critical issue: the model's search capability tends to degrade rapidly if fine-tuning is performed naively. We show that this degradation can be substantially mitigated by employing a smaller learning rate. Extensive experiments on the challenging Game-of-24 and Countdown mathematical reasoning benchmarks show that our approach not only outperforms both standard fine-tuning and inference-time search baselines but also significantly reduces inference time by 180$\times$.
Authors: Rui Tang, Ziyun Yong, Shuyu Jiang, Xingshu Chen, Yaofang Liu, Yi-Cheng Zhang, Gui-Quan Sun, Wei Wang
Abstract: Complex networks are frequently employed to model physical or virtual complex systems. When certain entities exist across multiple systems simultaneously, unveiling their corresponding relationships across the networks becomes crucial. This problem, known as network alignment, holds significant importance. It enhances our understanding of complex system structures and behaviours, facilitates the validation and extension of theoretical physics research about studying complex systems, and fosters diverse practical applications across various fields. However, due to variations in the structure, characteristics, and properties of complex networks across different fields, the study of network alignment is often isolated within each domain, with even the terminologies and concepts lacking uniformity. This review comprehensively summarizes the latest advancements in network alignment research, focusing on analyzing network alignment characteristics and progress in various domains such as social network analysis, bioinformatics, computational linguistics and privacy protection. It provides a detailed analysis of various methods' implementation principles, processes, and performance differences, including structure consistency-based methods, network embedding-based methods, and graph neural network-based (GNN-based) methods. Additionally, the methods for network alignment under different conditions, such as in attributed networks, heterogeneous networks, directed networks, and dynamic networks, are presented. Furthermore, the challenges and the open issues for future studies are also discussed.
Authors: Ian Magnusson, Nguyen Tai, Ben Bogin, David Heineman, Jena D. Hwang, Luca Soldaini, Akshita Bhagia, Jiacheng Liu, Dirk Groeneveld, Oyvind Tafjord, Noah A. Smith, Pang Wei Koh, Jesse Dodge
Abstract: Because large language models are expensive to pretrain on different datasets, using smaller-scale experiments to decide on data is crucial for reducing costs. Which benchmarks and methods of making decisions from observed performance at small scale most accurately predict the datasets that yield the best large models? To empower open exploration of this question, we release models, data, and evaluations in DataDecide -- the most extensive open suite of models over differences in data and scale. We conduct controlled pretraining experiments across 25 corpora with differing sources, deduplication, and filtering up to 100B tokens, model sizes up to 1B parameters, and 3 random seeds. We find that the ranking of models at a single, small size (e.g., 150M parameters) is a strong baseline for predicting best models at our larger target scale (1B) (~80% of com parisons correct). No scaling law methods among 8 baselines exceed the compute-decision frontier of single-scale predictions, but DataDecide can measure improvement in future scaling laws. We also identify that using continuous likelihood metrics as proxies in small experiments makes benchmarks including MMLU, ARC, HellaSwag, MBPP, and HumanEval >80% predictable at the target 1B scale with just 0.01% of the compute.
Authors: Elizabeth Fons, Rachneet Kaur, Zhen Zeng, Soham Palande, Tucker Balch, Svitlana Vyetrenko, Manuela Veloso
Abstract: While image captioning has gained significant attention, the potential of captioning time-series images, prevalent in areas like finance and healthcare, remains largely untapped. Existing time-series captioning methods typically offer generic, domain-agnostic descriptions of time-series shapes and struggle to adapt to new domains without substantial retraining. To address these limitations, we introduce TADACap, a retrieval-based framework to generate domain-aware captions for time-series images, capable of adapting to new domains without retraining. Building on TADACap, we propose a novel retrieval strategy that retrieves diverse image-caption pairs from a target domain database, namely TADACap-diverse. We benchmarked TADACap-diverse against state-of-the-art methods and ablation variants. TADACap-diverse demonstrates comparable semantic accuracy while requiring significantly less annotation effort.
Authors: Hrishikesh Viswanath, Tianyi Zhang
Abstract: Recent studies have demonstrated that large pretrained language models (LLMs) such as BERT and GPT-2 exhibit biases in token prediction, often inherited from the data distributions present in their training corpora. In response, a number of mathematical frameworks have been proposed to quantify, identify, and mitigate these the likelihood of biased token predictions. In this paper, we present a comprehensive survey of such techniques tailored towards widely used LLMs such as BERT, GPT-2, etc. We additionally introduce Fairpy, a modular and extensible toolkit that provides plug-and-play interfaces for integrating these mathematical tools, enabling users to evaluate both pretrained and custom language models. Fairpy supports the implementation of existing debiasing algorithms. The toolkit is open-source and publicly available at: \href{https://github.com/HrishikeshVish/Fairpy}{https://github.com/HrishikeshVish/Fairpy}
URLs: https://github.com/HrishikeshVish/Fairpy, https://github.com/HrishikeshVish/Fairpy
Authors: Yuchi Liu, Jaskirat Singh, Gaowen Liu, Ali Payani, Liang Zheng
Abstract: Large language models (LLMs) have shown great progress in responding to user questions, allowing for a multitude of diverse applications. Yet, the quality of LLM outputs heavily depends on the prompt design, where a good prompt might enable the LLM to answer a very challenging question correctly. Therefore, recent works have developed many strategies for improving the prompt, including both manual crafting and in-domain optimization. However, their efficacy in unrestricted scenarios remains questionable, as the former depends on human design for specific questions and the latter usually generalizes poorly to unseen scenarios. To address these problems, we give LLMs the freedom to design the best prompts according to themselves. Specifically, we include a hierarchy of LLMs, first constructing a prompt with precise instructions and accurate wording in a hierarchical manner, and then using this prompt to generate the final answer to the user query. We term this pipeline Hierarchical Multi-Agent Workflow, or HMAW. In contrast with prior works, HMAW imposes no human restriction and requires no training, and is completely task-agnostic while capable of adjusting to the nuances of the underlying task. Through both quantitative and qualitative experiments across multiple benchmarks, we verify that despite its simplicity, the proposed approach can create detailed and suitable prompts, further boosting the performance of current LLMs.
Authors: Meiling Tao, Xuechen Liang, Xinyuan Song, Yangfan He, Yiling Tao, Jianhui Wang, Sun Li Tianyu Shi
Abstract: Recent advancements in large language models (LLMs) have unlocked the potential for generating high-quality game commentary. However, producing insightful and engaging commentary for complex games with incomplete information remains a significant challenge. In this paper, we introduce a novel commentary method that combine Reinforcement Learning (RL) and LLMs, tailored specifically for the Chinese card game \textit{Guandan}. Our system leverages RL to generate intricate card-playing scenarios and employs LLMs to generate corresponding commentary text, effectively emulating the strategic analysis and narrative prowess of professional commentators. The framework comprises a state commentary guide, a Theory of Mind (ToM)-based strategy analyzer, and a style retrieval module, which seamlessly collaborate to deliver detailed and context-relevant game commentary in the Chinese language environment. We empower LLMs with ToM capabilities and refine both retrieval and information filtering mechanisms. This facilitates the generation of personalized commentary content. Our experimental results showcase the substantial enhancement in performance achieved by the proposed commentary framework when applied to open-source LLMs, surpassing the performance of GPT-4 across multiple evaluation metrics.
Authors: Lihu Chen, Ga\"el Varoquaux
Abstract: Large Language Models (LLMs) have made significant progress in advancing artificial general intelligence (AGI), leading to the development of increasingly large models such as GPT-4 and LLaMA-405B. However, scaling up model sizes results in exponentially higher computational costs and energy consumption, making these models impractical for academic researchers and businesses with limited resources. At the same time, Small Models (SMs) are frequently used in practical settings, although their significance is currently underestimated. This raises important questions about the role of small models in the era of LLMs, a topic that has received limited attention in prior research. In this work, we systematically examine the relationship between LLMs and SMs from two key perspectives: Collaboration and Competition. We hope this survey provides valuable insights for practitioners, fostering a deeper understanding of the contribution of small models and promoting more efficient use of computational resources. The code is available at https://github.com/tigerchen52/role_of_small_models
Authors: Zheng Hui, Zhaoxiao Guo, Hang Zhao, Juanyong Duan, Congrui Huang
Abstract: In different NLP tasks, detecting harmful content is crucial for online environments, especially with the growing influence of social media. However, previous research has two main issues: 1) a lack of data in low-resource settings, and 2) inconsistent definitions and criteria for judging harmful content, requiring classification models to be robust to spurious features and diverse. We propose Toxicraft, a novel framework for synthesizing datasets of harmful information to address these weaknesses. With only a small amount of seed data, our framework can generate a wide variety of synthetic, yet remarkably realistic, examples of toxic information. Experimentation across various datasets showcases a notable enhancement in detection model robustness and adaptability, surpassing or close to the gold labels.
Authors: Aiwei Liu, Haoping Bai, Zhiyun Lu, Yanchao Sun, Xiang Kong, Simon Wang, Jiulong Shan, Albin Madappally Jose, Xiaojiang Liu, Lijie Wen, Philip S. Yu, Meng Cao
Abstract: Direct Preference Optimization (DPO) has been widely adopted for preference alignment of Large Language Models (LLMs) due to its simplicity and effectiveness. However, DPO is derived as a bandit problem in which the whole response is treated as a single arm, ignoring the importance differences between tokens, which may affect optimization efficiency and make it difficult to achieve optimal results. In this work, we propose that the optimal data for DPO has equal expected rewards for each token in winning and losing responses, as there is no difference in token importance. However, since the optimal dataset is unavailable in practice, we propose using the original dataset for importance sampling to achieve unbiased optimization. Accordingly, we propose a token-level importance sampling DPO objective named TIS-DPO that assigns importance weights to each token based on its reward. Inspired by previous works, we estimate the token importance weights using the difference in prediction probabilities from a pair of contrastive LLMs. We explore three methods to construct these contrastive LLMs: (1) guiding the original LLM with contrastive prompts, (2) training two separate LLMs using winning and losing responses, and (3) performing forward and reverse DPO training with winning and losing responses. Experiments show that TIS-DPO significantly outperforms various baseline methods on harmlessness and helpfulness alignment and summarization tasks. We also visualize the estimated weights, demonstrating their ability to identify key token positions.
Authors: Christos Xypolopoulos, Guokan Shang, Xiao Fei, Giannis Nikolentzos, Hadi Abdine, Iakovos Evdaimon, Michail Chatzianastasis, Giorgos Stamou, Michalis Vazirgiannis
Abstract: Large language models have evolved to process multiple modalities beyond text, such as images and audio, which motivates us to explore how to effectively leverage them for graph reasoning tasks. The key question, therefore, is how to transform graphs into linear sequences of tokens, a process we term "graph linearization", so that LLMs can handle graphs naturally. We consider that graphs should be linearized meaningfully to reflect certain properties of natural language text, such as local dependency and global alignment, in order to ease contemporary LLMs, trained on trillions of textual tokens, better understand graphs. To achieve this, we developed several graph linearization methods based on graph centrality and degeneracy. These methods are further enhanced using node relabeling techniques. The experimental results demonstrate the effectiveness of our methods compared to the random linearization baseline. Our work introduces novel graph representations suitable for LLMs, contributing to the potential integration of graph machine learning with the trend of multimodal processing using a unified transformer model.
Authors: Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V. Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D. Hwang, Jiangjiang Yang, Ronan Le Bras, Oyvind Tafjord, Chris Wilhelm, Luca Soldaini, Noah A. Smith, Yizhong Wang, Pradeep Dasigi, Hannaneh Hajishirzi
Abstract: Language model post-training is applied to refine behaviors and unlock new skills across a wide range of recent language models, but open recipes for applying these techniques lag behind proprietary ones. The underlying training data and recipes for post-training are simultaneously the most important pieces of the puzzle and the portion with the least transparency. To bridge this gap, we introduce Tulu 3, a family of fully-open state-of-the-art post-trained models, alongside its data, code, and training recipes, serving as a comprehensive guide for modern post-training techniques. Tulu 3, which builds on Llama 3.1 base models, achieves results surpassing the instruct versions of Llama 3.1, Qwen 2.5, Mistral, and even closed models such as GPT-4o-mini and Claude 3.5-Haiku. The training algorithms for our models include supervised finetuning (SFT), Direct Preference Optimization (DPO), and a novel method we call Reinforcement Learning with Verifiable Rewards (RLVR). With Tulu 3, we introduce a multi-task evaluation scheme for post-training recipes with development and unseen evaluations, standard benchmark implementations, and substantial decontamination of existing open datasets on said benchmarks. We conclude with analysis and discussion of training methods that did not reliably improve performance. In addition to the Tulu 3 model weights and demo, we release the complete recipe -- including datasets for diverse core skills, a robust toolkit for data curation and evaluation, the training code and infrastructure, and, most importantly, a detailed report for reproducing and further adapting the Tulu 3 approach to more domains.
Authors: Chang-Jin Li, Jiyuan Zhang, Yun Tang, Jian Li
Abstract: Personality assessment, particularly through situational judgment tests (SJTs), is a vital tool for psychological research, talent selection, and educational evaluation. This study explores the potential of GPT-4, a state-of-the-art large language model (LLM), to automate the generation of personality situational judgment tests (PSJTs) in Chinese. Traditional SJT development is labor-intensive and prone to biases, while GPT-4 offers a scalable, efficient alternative. Two studies were conducted: Study 1 evaluated the impact of prompt design and temperature settings on content validity, finding that optimized prompts with a temperature of 1.0 produced creative and accurate items. Study 2 assessed the psychometric properties of GPT-4-generated PSJTs, revealing that they demonstrated satisfactory reliability and validity, surpassing the performance of manually developed tests in measuring the Big Five personality traits. This research highlights GPT-4's effectiveness in developing high-quality PSJTs, providing a scalable and innovative method for psychometric test development. These findings expand the possibilities of automatic item generation and the application of LLMs in psychology, and offer practical implications for streamlining test development processes in resource-limited settings.
Authors: Vincenzo Timmel, Claudio Paonessa, Reza Kakooee, Manfred Vogel, Daniel Perruchoud
Abstract: This paper presents a new approach to fine-tuning OpenAI's Whisper model for low-resource languages by introducing a novel data generation method that converts sentence-level data into a long-form corpus, using Swiss German as a case study. Non-sentence-level data, which could improve the performance of long-form audio, is difficult to obtain and often restricted by copyright laws. Our method bridges this gap by transforming more accessible sentence-level data into a format that preserves the model's ability to handle long-form audio and perform segmentation without requiring non-sentence-level data. Our data generation process improves performance in several real-world applications and leads to the development of a new state-of-the-art speech-to-text (STT) model for Swiss German. We compare our model with a non-fine-tuned Whisper and our previous state-of-the-art Swiss German STT models, where our new model achieves higher BLEU scores. Our results also indicate that the proposed method is adaptable to other low-resource languages, supported by written guidance and code that allows the creation of fine-tuned Whisper models, which keep segmentation capabilities and allow the transcription of longer audio files using only sentence-level data with high quality.
Authors: Arina Razmyslovich, Kseniia Murasheva, Sofia Sedlova, Julien Capitaine, Eugene Dmitriev
Abstract: We introduce Efficient LLM Token Extraction (ELTEX), a framework addressing the critical challenge of LLM domain specialization by systematically extracting and integrating domain indicators throughout synthetic data generation. Unlike approaches relying on implicit knowledge transfer, ELTEX explicitly leverages domain signals to maintain specialized knowledge integrity. In our cybersecurity case study, ELTEX-enhanced data enables a fine-tuned Gemma-2B model to achieve performance competitive with GPT-4o on blockchain cyberattack classification while reducing computational requirements. Our Google Sheets implementation makes ELTEX accessible to non-technical users. Our contributions include: (1) the ELTEX framework; (2) Google Sheets Add-on implementation; (3) empirical validation showing how ELTEX bridges performance gaps between small and large models; and (4) a synthetic dataset of 11,448 texts for blockchain cyberattack detection.
Authors: Zixin Chen, Sicheng Song, Kashun Shum, Yanna Lin, Rui Sheng, Huamin Qu
Abstract: Misleading chart visualizations, which intentionally manipulate data representations to support specific claims, can distort perceptions and lead to incorrect conclusions. Despite decades of research, misleading visualizations remain a widespread and pressing issue. Recent advances in multimodal large language models (MLLMs) have demonstrated strong chart comprehension capabilities, yet no existing work has systematically evaluated their ability to detect and interpret misleading charts. This paper introduces the Misleading Chart Question Answering (Misleading ChartQA) Benchmark, a large-scale multimodal dataset designed to assess MLLMs in identifying and reasoning about misleading charts. It contains over 3,000 curated examples, covering 21 types of misleaders and 10 chart types. Each example includes standardized chart code, CSV data, and multiple-choice questions with labeled explanations, validated through multi-round MLLM checks and exhausted expert human review. We benchmark 16 state-of-the-art MLLMs on our dataset, revealing their limitations in identifying visually deceptive practices. We also propose a novel pipeline that detects and localizes misleaders, enhancing MLLMs' accuracy in misleading chart interpretation. Our work establishes a foundation for advancing MLLM-driven misleading chart comprehension. We publicly release the sample dataset to support further research in this critical area.
Authors: Magdalena Kaiser, Gerhard Weikum
Abstract: Conversational Question Answering (ConvQA) involves multiple subtasks, i) to understand incomplete questions in their context, ii) to retrieve relevant information, and iii) to generate answers. This work presents PRAISE, a pipeline-based approach for ConvQA that trains LLM adapters for each of the three subtasks. As labeled training data for individual subtasks is unavailable in practice, PRAISE learns from its own generations using the final answering performance as feedback signal without human intervention and treats intermediate information, like relevant evidence, as weakly labeled data. We apply Direct Preference Optimization by contrasting successful and unsuccessful samples for each subtask. In our experiments, we show the effectiveness of this training paradigm: PRAISE shows improvements per subtask and achieves new state-of-the-art performance on a popular ConvQA benchmark, by gaining 15.5 percentage points increase in precision over baselines.
Authors: NVIDIA, :, Aaron Blakeman, Aarti Basant, Abhinav Khattar, Adithya Renduchintala, Akhiad Bercovich, Aleksander Ficek, Alexis Bjorlin, Ali Taghibakhshi, Amala Sanjay Deshmukh, Ameya Sunil Mahabaleshwarkar, Andrew Tao, Anna Shors, Ashwath Aithal, Ashwin Poojary, Ayush Dattagupta, Balaram Buddharaju, Bobby Chen, Boris Ginsburg, Boxin Wang, Brandon Norick, Brian Butterfield, Bryan Catanzaro, Carlo del Mundo, Chengyu Dong, Christine Harvey, Christopher Parisien, Dan Su, Daniel Korzekwa, Danny Yin, Daria Gitman, David Mosallanezhad, Deepak Narayanan, Denys Fridman, Dima Rekesh, Ding Ma, Dmytro Pykhtar, Dong Ahn, Duncan Riach, Dusan Stosic, Eileen Long, Elad Segal, Ellie Evans, Eric Chung, Erick Galinkin, Evelina Bakhturina, Ewa Dobrowolska, Fei Jia, Fuxiao Liu, Gargi Prasad, Gerald Shen, Guilin Liu, Guo Chen, Haifeng Qian, Helen Ngo, Hongbin Liu, Hui Li, Igor Gitman, Ilia Karmanov, Ivan Moshkov, Izik Golan, Jan Kautz, Jane Polak Scowcroft, Jared Casper, Jarno Seppanen, Jason Lu, Jason Sewall, Jiaqi Zeng, Jiaxuan You, Jimmy Zhang, Jing Zhang, Jining Huang, Jinze Xue, Jocelyn Huang, Joey Conway, John Kamalu, Jon Barker, Jonathan Cohen, Joseph Jennings, Jupinder Parmar, Karan Sapra, Kari Briski, Kateryna Chumachenko, Katherine Luna, Keshav Santhanam, Kezhi Kong, Kirthi Sivamani, Krzysztof Pawelec, Kumar Anik, Kunlun Li, Lawrence McAfee, Leon Derczynski, Lindsey Pavao, Luis Vega, Lukas Voegtle, Maciej Bala, Maer Rodrigues de Melo, Makesh Narsimhan Sreedhar, Marcin Chochowski, Markus Kliegl, Marta Stepniewska-Dziubinska, Matthieu Le, Matvei Novikov, Mehrzad Samadi, Michael Andersch, Michael Evans, Miguel Martinez, Mike Chrzanowski, Mike Ranzinger, Mikolaj Blaz, Misha Smelyanskiy, Mohamed Fawzy, Mohammad Shoeybi, Mostofa Patwary, Nayeon Lee, Nima Tajbakhsh, Ning Xu, Oleg Rybakov, Oleksii Kuchaiev, Olivier Delalleau, Osvald Nitski, Parth Chadha, Pasha Shamis, Paulius Micikevicius, Pavlo Molchanov, Peter Dykas, Philipp Fischer, Pierre-Yves Aquilanti, Piotr Bialecki, Prasoon Varshney, Pritam Gundecha, Przemek Tredak, Rabeeh Karimi, Rahul Kandu, Ran El-Yaniv, Raviraj Joshi, Roger Waleffe, Ruoxi Zhang, Sabrina Kavanaugh, Sahil Jain, Samuel Kriman, Sangkug Lym, Sanjeev Satheesh, Saurav Muralidharan, Sean Narenthiran, Selvaraj Anandaraj, Seonmyeong Bak, Sergey Kashirsky, Seungju Han, Shantanu Acharya, Shaona Ghosh, Sharath Turuvekere Sreenivas, Sharon Clay, Shelby Thomas, Shrimai Prabhumoye, Shubham Pachori, Shubham Toshniwal, Shyamala Prayaga, Siddhartha Jain, Sirshak Das, Slawek Kierat, Somshubra Majumdar, Song Han, Soumye Singhal, Sriharsha Niverty, Stefania Alborghetti, Suseella Panguluri, Swetha Bhendigeri, Syeda Nahida Akter, Szymon Migacz, Tal Shiri, Terry Kong, Timo Roman, Tomer Ronen, Trisha Saar, Tugrul Konuk, Tuomas Rintamaki, Tyler Poon, Ushnish De, Vahid Noroozi, Varun Singh, Vijay Korthikanti, Vitaly Kurin, Wasi Uddin Ahmad, Wei Du, Wei Ping, Wenliang Dai, Wonmin Byeon, Xiaowei Ren, Yao Xu, Yejin Choi, Yian Zhang, Ying Lin, Yoshi Suhara, Zhiding Yu, Zhiqi Li, Zhiyu Li, Zhongbo Zhu, Zhuolin Yang, Zijia Chen
Abstract: As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.
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: 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: Haoyu Wang, Yujia Fu, Zhu Zhang, Shuo Wang, Zirui Ren, Xiaorong Wang, Zhili Li, Chaoqun He, Bo An, Zhiyuan Liu, Maosong Sun
Abstract: Long-form generation is crucial for a wide range of practical applications, typically categorized into short-to-long and long-to-long generation. While short-to-long generations have received considerable attention, generating long texts from extremely long resources remains relatively underexplored. The primary challenge in long-to-long generation lies in effectively integrating and analyzing relevant information from extensive inputs, which remains difficult for current large language models (LLMs). In this paper, we propose LLM$\times$MapReduce-V2, a novel test-time scaling strategy designed to enhance the ability of LLMs to process extremely long inputs. Drawing inspiration from convolutional neural networks, which iteratively integrate local features into higher-level global representations, LLM$\times$MapReduce-V2 utilizes stacked convolutional scaling layers to progressively expand the understanding of input materials. Both quantitative and qualitative experimental results demonstrate that our approach substantially enhances the ability of LLMs to process long inputs and generate coherent, informative long-form articles, outperforming several representative baselines. Both LLM$\times$MapReduce-V2 and SurveyEval are publicly available at https://github.com/thunlp/LLMxMapReduce .
Authors: Raymond Ng, Thanh Ngan Nguyen, Yuli Huang, Ngee Chia Tai, Wai Yi Leong, Wei Qi Leong, Xianbin Yong, Jian Gang Ngui, Yosephine Susanto, Nicholas Cheng, Hamsawardhini Rengarajan, Peerat Limkonchotiwat, Adithya Venkatadri Hulagadri, Kok Wai Teng, Yeo Yeow Tong, Bryan Siow, Wei Yi Teo, Wayne Lau, Choon Meng Tan, Brandon Ong, Zhi Hao Ong, Jann Railey Montalan, Adwin Chan, Sajeban Antonyrex, Ren Lee, Esther Choa, David Ong Tat-Wee, Bing Jie Darius Liu, William Chandra Tjhi, Erik Cambria, Leslie Teo
Abstract: Recently, Large Language Models (LLMs) have dominated much of the artificial intelligence scene with their ability to process and generate natural languages. However, the majority of LLM research and development remains English-centric, leaving low-resource languages such as those in the Southeast Asian (SEA) region under-represented. To address this representation gap, we introduce Llama-SEA-LION-v3-8B-IT and Gemma-SEA-LION-v3-9B-IT, two cutting-edge multilingual LLMs designed for SEA languages. The SEA-LION family of LLMs supports 11 SEA languages, namely English, Chinese, Indonesian, Vietnamese, Malay, Thai, Burmese, Lao, Filipino, Tamil, and Khmer. Our work leverages large-scale multilingual continued pre-training with a comprehensive post-training regime involving multiple stages of instruction fine-tuning, alignment, and model merging. Evaluation results on multilingual benchmarks indicate that our models achieve state-of-the-art performance across LLMs supporting SEA languages. We open-source the models to benefit the wider SEA community.
Authors: Biplav Srivastava, Kausik Lakkaraju, Nitin Gupta, Vansh Nagpal, Bharath C. Muppasani, Sara E. Jones
Abstract: Collaborative assistants, or chatbots, are data-driven decision support systems that enable natural interaction for task completion. While they can meet critical needs in modern society, concerns about their reliability and trustworthiness persist. In particular, Large Language Model (LLM)-based chatbots like ChatGPT, Gemini, and DeepSeek are becoming more accessible. However, such chatbots have limitations, including their inability to explain response generation, the risk of generating problematic content, the lack of standardized testing for reliability, and the need for deep AI expertise and extended development times. These issues make chatbots unsuitable for trust-sensitive applications like elections or healthcare. To address these concerns, we introduce SafeChat, a general architecture for building safe and trustworthy chatbots, with a focus on information retrieval use cases. Key features of SafeChat include: (a) safety, with a domain-agnostic design where responses are grounded and traceable to approved sources (provenance), and 'do-not-respond' strategies to prevent harmful answers; (b) usability, with automatic extractive summarization of long responses, traceable to their sources, and automated trust assessments to communicate expected chatbot behavior, such as sentiment; and (c) fast, scalable development, including a CSV-driven workflow, automated testing, and integration with various devices. We implemented SafeChat in an executable framework using the open-source chatbot platform Rasa. A case study demonstrates its application in building ElectionBot-SC, a chatbot designed to safely disseminate official election information. SafeChat is being used in many domains, validating its potential, and is available at: https://github.com/ai4society/trustworthy-chatbot.
Authors: Yuyang Xu, Renjun Hu, Haochao Ying, Jian Wu, Xing Shi, Wei Lin
Abstract: Multiple-choice question (MCQ) benchmarks are widely used for evaluating Large Language Models (LLMs), yet their reliability is undermined by benchmark contamination. In this study, we reframe contamination as an inherent aspect of learning and seek to disentangle genuine capability acquisition from superficial memorization in LLM evaluation. First, by analyzing model performance under different memorization conditions, we uncover a counterintuitive trend: LLMs perform worse on memorized MCQs than on non-memorized ones, indicating the coexistence of two distinct learning phenomena, i.e., rote memorization and genuine capability learning. To disentangle them, we propose TrinEval, a novel evaluation framework that reformulates MCQs into an alternative trinity format, reducing memorization while preserving knowledge assessment. Experiments validate TrinEval's effectiveness in reformulation, and its evaluation reveals that common LLMs may memorize by rote 20.5% of knowledge points (in MMLU on average).
Authors: Kartik Ravisankar, Hyojung Han, Marine Carpuat
Abstract: Large language models (LLMs) pre-trained predominantly on English text exhibit surprising multilingual capabilities, yet the mechanisms driving cross-lingual generalization remain poorly understood. This work investigates how the alignment of representations for text written in different languages correlates with LLM performance on natural language understanding tasks and translation tasks, both at the language and the instance level. For this purpose, we introduce cross-lingual alignment metrics such as the Discriminative Alignment Index (DALI) to quantify the alignment at an instance level for discriminative tasks. Through experiments on three natural language understanding tasks (Belebele, XStoryCloze, XCOPA), and machine translation, we find that while cross-lingual alignment metrics strongly correlate with task accuracy at the language level, the sample-level alignment often fails to distinguish correct from incorrect predictions, exposing alignment as a necessary but insufficient condition for success.
Authors: Wuyang Lan, Wenzheng Wang, Changwei Ji, Guoxing Yang, Yongbo Zhang, Xiaohong Liu, Song Wu, Guangyu Wang
Abstract: Recent advances in reasoning with large language models (LLMs)has shown remarkable reasoning capabilities in domains such as mathematics and coding, yet their application to clinical diagnosis remains underexplored. Here, we introduce ClinicalGPT-R1, a reasoning enhanced generalist large language model for disease diagnosis. Trained on a dataset of 20,000 real-world clinical records, ClinicalGPT-R1 leverages diverse training strategies to enhance diagnostic reasoning. To benchmark performance, we curated MedBench-Hard, a challenging dataset spanning seven major medical specialties and representative diseases. Experimental results demonstrate that ClinicalGPT-R1 outperforms GPT-4o in Chinese diagnostic tasks and achieves comparable performance to GPT-4 in English settings. This comparative study effectively validates the superior performance of ClinicalGPT-R1 in disease diagnosis tasks. Resources are available at https://github.com/medfound/medfound.
Authors: Yueqi Song, Tianyue Ou, Yibo Kong, Zecheng Li, Graham Neubig, Xiang Yue
Abstract: Current multimodal benchmarks often conflate reasoning with domain-specific knowledge, making it difficult to isolate and evaluate general reasoning abilities in non-expert settings. To address this, we introduce VisualPuzzles, a benchmark that targets visual reasoning while deliberately minimizing reliance on specialized knowledge. VisualPuzzles consists of diverse questions spanning five categories: algorithmic, analogical, deductive, inductive, and spatial reasoning. One major source of our questions is manually translated logical reasoning questions from the Chinese Civil Service Examination. Experiments show that VisualPuzzles requires significantly less intensive domain-specific knowledge and more complex reasoning compared to benchmarks like MMMU, enabling us to better evaluate genuine multimodal reasoning. Evaluations show that state-of-the-art multimodal large language models consistently lag behind human performance on VisualPuzzles, and that strong performance on knowledge-intensive benchmarks does not necessarily translate to success on reasoning-focused, knowledge-light tasks. Additionally, reasoning enhancements such as scaling up inference compute (with "thinking" modes) yield inconsistent gains across models and task types, and we observe no clear correlation between model size and performance. We also found that models exhibit different reasoning and answering patterns on VisualPuzzles compared to benchmarks with heavier emphasis on knowledge. VisualPuzzles offers a clearer lens through which to evaluate reasoning capabilities beyond factual recall and domain knowledge.
Authors: Dieuwke Hupkes, Nikolay Bogoychev
Abstract: We present MultiLoKo, a new benchmark for evaluating multilinguality in LLMs covering 31 languages. MultiLoKo consists of three partitions: a main partition consisting of 500 questions per language, separately sourced to be locally relevant to the specific language, and two translated partitions, containing human-authored translations from 30 non-English languages to English and vice versa. For comparison, we also release corresponding machine-authored translations. The data is equally distributed over two splits: a dev split and a blind, out-of-distribution test split. MultiLoKo can be used to study a variety of questions regarding the multilinguality of LLMs as well as meta-questions about multilingual benchmark creation. We compute MultiLoKo scores for 11 base and chat models marketed to be multilingual and study their average performance, their performance parity across languages, how much their ability to answer questions depends on the question language, and which languages are most difficult. None of the models we studied performs well on MultiLoKo, as indicated by low average scores as well as large differences between the best and worst scoring languages. Furthermore, we find a substantial effect of the question language, indicating sub-optimal knowledge transfer between languages. Lastly, we find that using local vs English-translated data can result in differences more than 20 points for the best performing models, drastically change the estimated difficulty of some languages. For using machines instead of human translations, we find a weaker effect on ordering of language difficulty, a larger difference in model rankings, and a substantial drop in estimated performance for all models.
Authors: Micha{\l} Turski, Mateusz Chili\'nski, {\L}ukasz Borchmann
Abstract: Checkboxes are critical in real-world document processing where the presence or absence of ticks directly informs data extraction and decision-making processes. Yet, despite the strong performance of Large Vision and Language Models across a wide range of tasks, they struggle with interpreting checkable content. This challenge becomes particularly pressing in industries where a single overlooked checkbox may lead to costly regulatory or contractual oversights. To address this gap, we introduce the CheckboxQA dataset, a targeted resource designed to evaluate and improve model performance on checkbox-related tasks. It reveals the limitations of current models and serves as a valuable tool for advancing document comprehension systems, with significant implications for applications in sectors such as legal tech and finance. The dataset is publicly available at: https://github.com/Snowflake-Labs/CheckboxQA
Authors: Hao Sha, Yao Mu, Yuxuan Jiang, Li Chen, Chenfeng Xu, Ping Luo, Shengbo Eben Li, Masayoshi Tomizuka, Wei Zhan, Mingyu Ding
Abstract: Existing learning-based autonomous driving (AD) systems face challenges in comprehending high-level information, generalizing to rare events, and providing interpretability. To address these problems, this work employs Large Language Models (LLMs) as a decision-making component for complex AD scenarios that require human commonsense understanding. We devise cognitive pathways to enable comprehensive reasoning with LLMs, and develop algorithms for translating LLM decisions into actionable driving commands. Through this approach, LLM decisions are seamlessly integrated with low-level controllers by guided parameter matrix adaptation. Extensive experiments demonstrate that our proposed method not only consistently surpasses baseline approaches in single-vehicle tasks, but also helps handle complex driving behaviors even multi-vehicle coordination, thanks to the commonsense reasoning capabilities of LLMs. This paper presents an initial step toward leveraging LLMs as effective decision-makers for intricate AD scenarios in terms of safety, efficiency, generalizability, and interoperability. We aspire for it to serve as inspiration for future research in this field. Project page: https://sites.google.com/view/llm-mpc
Authors: Mazal Bethany, Athanasios Galiopoulos, Emet Bethany, Mohammad Bahrami Karkevandi, Nicole Beebe, Nishant Vishwamitra, Peyman Najafirad
Abstract: The emergence of Large Language Models (LLMs) has heightened the threat of phishing emails by enabling the generation of highly targeted, personalized, and automated attacks. Traditionally, many phishing emails have been characterized by typos, errors, and poor language. These errors can be mitigated by LLMs, potentially lowering the barrier for attackers. Despite this, there is a lack of large-scale studies comparing the effectiveness of LLM-generated lateral phishing emails to those crafted by humans. Current literature does not adequately address the comparative effectiveness of LLM and human-generated lateral phishing emails in a real-world, large-scale organizational setting, especially considering the potential for LLMs to generate more convincing and error-free phishing content. To address this gap, we conducted a pioneering study within a large university, targeting its workforce of approximately 9,000 individuals including faculty, staff, administrators, and student workers. Our results indicate that LLM-generated lateral phishing emails are as effective as those written by communications professionals, emphasizing the critical threat posed by LLMs in leading phishing campaigns. We break down the results of the overall phishing experiment, comparing vulnerability between departments and job roles. Furthermore, to gather qualitative data, we administered a detailed questionnaire, revealing insights into the reasons and motivations behind vulnerable employee's actions. This study contributes to the understanding of cyber security threats in educational institutions and provides a comprehensive comparison of LLM and human-generated phishing emails' effectiveness, considering the potential for LLMs to generate more convincing content. The findings highlight the need for enhanced user education and system defenses to mitigate the growing threat of AI-powered phishing attacks.
Authors: Lucio Dery, Steven Kolawole, Jean-Fran\c{c}ois Kagy, Virginia Smith, Graham Neubig, Ameet Talwalkar
Abstract: Structured pruning is a promising approach to create smaller, faster LLMs. However, existing methods typically rely on backward passes, which can inflate memory requirements and compute costs. In this work we introduce Bonsai, a gradient-free structured pruning method that eliminates the need for backpropagation, significantly reducing memory requirements and compute costs while achieving state-of-the-art pruning performance. Bonsai uses forward-pass-only perturbative pruning to enable efficient compression of large models on a broader range of hardware configurations. Unlike existing structured pruning approaches, Bonsai not only achieves better compression with fewer resources, but also produces models that are twice as fast as those generated by semi-structured pruning. As a concrete demonstration, we use Bonsai to prune an 8B LLaMA-3 model to 50% sparsity on a single A6000 GPU -- a task infeasible with backprop-based methods, which require 2-3x memory. Our results show that removing backprop as a requirement not only enables pruning larger models on constrained hardware but can also lead to state-of-the-art efficiency and performance.
Authors: Aniket Vashishtha, Abhinav Kumar, Atharva Pandey, Abbavaram Gowtham Reddy, Kabir Ahuja, Vineeth N Balasubramanian, Amit Sharma
Abstract: For text-based AI systems to interact in the real world, causal reasoning is an essential skill. Since active interventions are costly, we study to what extent a system can learn causal reasoning from symbolic demonstrations of causal axioms. Specifically, we present an axiomatic training method where the system learns from multiple demonstrations of a causal axiom (or rule), rather than incorporating the axiom as an inductive bias or inferring it from data values. A key question is whether the system would learn to generalize from the axiom demonstrations to more complex scenarios. Our results, based on applying axiomatic training to learn the transitivity axiom and d-separation rule, indicate that such generalization is possible. To avoid data contamination issues, we start with a 67 million parameter transformer model and train it from scratch. On both tasks, we find that a model trained on linear causal chains (along with some noisy variations) can generalize well to complex graphs, including longer causal chains, causal chains with reversed order, and graphs with branching.To handle diverse text inputs, the same method is extended to finetune language models. Finetuning Llama-3.1 8B model on our axiomatic data leads to significant gains on causal benchmarks such as Corr2Cause and CLEAR, in some cases providing state-of-the-art performance surpassing GPT-4.
Authors: Swarnadeep Saha, Archiki Prasad, Justin Chih-Yao Chen, Peter Hase, Elias Stengel-Eskin, Mohit Bansal
Abstract: Language models can be used to solve long-horizon planning problems in two distinct modes: a fast 'System-1' mode, directly generating plans without any explicit search or backtracking, and a slow 'System-2' mode, planning step-by-step by explicitly searching over possible actions. While System-2 is typically more effective, it is also more computationally expensive, making it infeasible for long plans or large action spaces. Moreover, isolated System-1 or 2 ignores the user's end goals, failing to provide ways to control the model's behavior. To this end, we propose the System-1.x Planner, a controllable planning framework with LLMs that is capable of generating hybrid plans and balancing between the two planning modes based on the difficulty of the problem at hand. System-1.x consists of (i) a controller, (ii) a System-1 Planner, and (iii) a System-2 Planner. Based on a user-specified hybridization factor (x) governing the mixture between System-1 and 2, the controller decomposes a problem into sub-goals, and classifies them as easy or hard to be solved by either System-1 or 2, respectively. We fine-tune all three components on top of a single base LLM, requiring only search traces as supervision. Experiments with two diverse planning tasks -- Maze Navigation and Blocksworld -- show that our System-1.x Planner outperforms a System-1 Planner, a System-2 Planner trained to approximate A* search, and also a symbolic planner (A*). We demonstrate the following key properties of our planner: (1) controllability: increasing the hybridization factor (e.g., System-1.75 vs 1.5) performs more search, improving performance, (2) flexibility: by building a neuro-symbolic variant with a neural System-1 and a symbolic System-2, we can use existing symbolic methods, and (3) generalizability: by being able to learn from different search algorithms, our method is robust to the choice of search algorithm.
Authors: Bin Wang, Chunyu Xie, Dawei Leng, Yuhui Yin
Abstract: In the field of multimodal large language models (MLLMs), common methods typically involve unfreezing the language model during training to foster profound visual understanding. However, the fine-tuning of such models with vision-language data often leads to a diminution of their natural language processing (NLP) capabilities. To avoid this performance degradation, a straightforward solution is to freeze the language model while developing multimodal competencies. Unfortunately, previous works have not attained satisfactory outcomes. Building on the strategy of freezing the language model, we conduct thorough structural exploration and introduce the Inner-Adaptor Architecture (IAA). Specifically, the architecture incorporates multiple multimodal adaptors at varying depths within the large language model to facilitate direct interaction with the inherently text-oriented transformer layers, thereby enabling the frozen language model to acquire multimodal capabilities. Unlike previous approaches of freezing language models that require large-scale aligned data, our proposed architecture is able to achieve superior performance on small-scale datasets. We conduct extensive experiments to improve the general multimodal capabilities and visual grounding abilities of the MLLM. Our approach remarkably outperforms previous state-of-the-art methods across various vision-language benchmarks without sacrificing performance on NLP tasks. Code and models are available at https://github.com/360CVGroup/Inner-Adaptor-Architecture.
URLs: https://github.com/360CVGroup/Inner-Adaptor-Architecture.
Authors: Jiayi Zhang, Jinyu Xiang, Zhaoyang Yu, Fengwei Teng, Xionghui Chen, Jiaqi Chen, Mingchen Zhuge, Xin Cheng, Sirui Hong, Jinlin Wang, Bingnan Zheng, Bang Liu, Yuyu Luo, Chenglin Wu
Abstract: Large language models (LLMs) have demonstrated remarkable potential in solving complex tasks across diverse domains, typically by employing agentic workflows that follow detailed instructions and operational sequences. However, constructing these workflows requires significant human effort, limiting scalability and generalizability. Recent research has sought to automate the generation and optimization of these workflows, but existing methods still rely on initial manual setup and fall short of achieving fully automated and effective workflow generation. To address this challenge, we reformulate workflow optimization as a search problem over code-represented workflows, where LLM-invoking nodes are connected by edges. We introduce AFlow, an automated framework that efficiently explores this space using Monte Carlo Tree Search, iteratively refining workflows through code modification, tree-structured experience, and execution feedback. Empirical evaluations across six benchmark datasets demonstrate AFlow's efficacy, yielding a 5.7% average improvement over state-of-the-art baselines. Furthermore, AFlow enables smaller models to outperform GPT-4o on specific tasks at 4.55% of its inference cost in dollars. The code is available at https://github.com/FoundationAgents/AFlow.
Authors: Huming Qiu, Guanxu Chen, Mi Zhang, Xiaohan Zhang, Xiaoyu You, Min Yang
Abstract: In recent years, text-to-image (T2I) generation models have made significant progress in generating high-quality images that align with text descriptions. However, these models also face the risk of unsafe generation, potentially producing harmful content that violates usage policies, such as explicit material. Existing safe generation methods typically focus on suppressing inappropriate content by erasing undesired concepts from visual representations, while neglecting to sanitize the textual representation. Although these methods help mitigate the risk of misuse to some extent, their robustness remains insufficient when dealing with adversarial attacks. Given that semantic consistency between input text and output image is a core requirement of T2I models, we identify that textual representations are likely the primary source of unsafe generation. To this end, we propose Embedding Sanitizer (ES), which enhances the safety of T2I models by sanitizing inappropriate concepts in prompt embeddings. To our knowledge, ES is the first interpretable safe generation framework that assigns a score to each token in the prompt to indicate its potential harmfulness. In addition, ES adopts a plug-and-play modular design, offering compatibility for seamless integration with various T2I models and other safeguards. Evaluations on five prompt benchmarks show that ES outperforms eleven existing safeguard baselines, achieving state-of-the-art robustness while maintaining high-quality image generation.
Authors: Andrea Caria
Abstract: This perspective article aims at providing an outline of the state of the art and future developments towards the integration of cutting-edge predictive language models with BCI. A synthetic overview of early and more recent linguistic models, from natural language processing (NLP) models to recent LLM, that to a varying extent improved predictive writing systems, is first provided. Second, a summary of previous BCI implementations integrating language models is presented. The few preliminary studies investigating the possible combination of LLM with BCI spellers to efficiently support fast communication and control are then described. Finally, current challenges and limitations towards the full integration of LLM with BCI systems are discussed. Recent investigations suggest that the combination of LLM with BCI might drastically improve human-computer interaction in patients with motor or language disorders as well as in healthy individuals. In particular, the pretrained autoregressive transformer models, such as GPT, that capitalize from parallelization, learning through pre-training and fine-tuning, promise a substantial improvement of BCI for communication with respect to previous systems incorporating simpler language models. Indeed, among various models, the GPT-2 was shown to represent an excellent candidate for its integration into BCI although testing was only perfomed on simulated conversations and not on real BCI scenarios. Prospectively, the full integration of LLM with advanced BCI systems might lead to a big leap forward towards fast, efficient and user-adaptive neurotechnology.
Authors: Fiorenzo Parascandolo, Nicholas Moratelli, Enver Sangineto, Lorenzo Baraldi, Rita Cucchiara
Abstract: Recent work has empirically shown that Vision-Language Models (VLMs) struggle to fully understand the compositional properties of the human language, usually modeling an image caption as a "bag of words". As a result, they perform poorly on compositional tasks, which require a deeper understanding of the different entities of a sentence (subject, verb, etc.) jointly with their mutual relationships in order to be solved. In this paper, we model the dependency relations among textual and visual tokens using a Causal Graphical Model (CGM), built using a dependency parser, and we train a decoder conditioned by the VLM visual encoder. Differently from standard autoregressive or parallel predictions, our decoder's generative process is partially-ordered following the CGM structure. This structure encourages the decoder to learn only the main causal dependencies in a sentence discarding spurious correlations. Using extensive experiments on five compositional benchmarks, we show that our method significantly outperforms all the state-of-the-art compositional approaches by a large margin, and it also improves over methods trained using much larger datasets.
Authors: Zhihang Liu, Chen-Wei Xie, Bin Wen, Feiwu Yu, Jixuan Chen, Boqiang Zhang, Nianzu Yang, Pandeng Li, Yinglu Li, Zuan Gao, Yun Zheng, Hongtao Xie
Abstract: Visual captioning benchmarks have become outdated with the emergence of modern multimodal large language models (MLLMs), as the brief ground-truth sentences and traditional metrics fail to assess detailed captions effectively. While recent benchmarks attempt to address this by focusing on keyword extraction or object-centric evaluation, they remain limited to vague-view or object-view analyses and incomplete visual element coverage. In this paper, we introduce CAPability, a comprehensive multi-view benchmark for evaluating visual captioning across 12 dimensions spanning six critical views. We curate nearly 11K human-annotated images and videos with visual element annotations to evaluate the generated captions. CAPability stably assesses both the correctness and thoroughness of captions using F1-score. By converting annotations to QA pairs, we further introduce a heuristic metric, \textit{know but cannot tell} ($K\bar{T}$), indicating a significant performance gap between QA and caption capabilities. Our work provides the first holistic analysis of MLLMs' captioning abilities, as we identify their strengths and weaknesses across various dimensions, guiding future research to enhance specific aspects of capabilities.
Authors: Weigao Sun, Disen Lan, Tong Zhu, Xiaoye Qu, Yu Cheng
Abstract: Linear Sequence Modeling (LSM) like linear attention, state space models and linear RNNs, and Mixture-of-Experts (MoE) have recently emerged as significant architectural improvements. In this paper, we introduce Linear-MoE, a production-level system for modeling and training large-scale models that integrate LSM with MoE. Linear-MoE leverages the advantages of both LSM modules for linear-complexity sequence modeling and MoE layers for sparsely activation, aiming to offer high performance with efficient training. The Linear-MoE system comprises: 1) Modeling subsystem, which provides a unified framework supporting all instances of LSM. and 2) Training subsystem, which facilitates efficient training by incorporating various advanced parallelism technologies, particularly Sequence Parallelism designed for Linear-MoE models. Additionally, we explore hybrid models that combine Linear-MoE layers with standard Transformer-MoE layers with its Sequence Parallelism to further enhance model flexibility and performance. Evaluations on two model series, A0.3B-2B and A1B-7B, demonstrate Linear-MoE achieves efficiency gains while maintaining competitive performance on various benchmarks, showcasing its potential as a next-generation foundational model architecture. Code: https://github.com/OpenSparseLLMs/Linear-MoE.
Authors: Tabia Tanzin Prama, Jannatul Ferdaws Amrin, Md. Mushfique Anwar, Iqbal H. Sarker
Abstract: The rise of social media has significantly increased the prevalence of cyberbullying (CB), posing serious risks to both mental and physical well-being. Effective detection systems are essential for mitigating its impact. While several machine learning (ML) models have been developed, few incorporate victims' psychological, demographic, and behavioral factors alongside bullying comments to assess severity. In this study, we propose an AI model intregrating user-specific attributes, including psychological factors (self-esteem, anxiety, depression), online behavior (internet usage, disciplinary history), and demographic attributes (race, gender, ethnicity), along with social media comments. Additionally, we introduce a re-labeling technique that categorizes social media comments into three severity levels: Not Bullying, Mild Bullying, and Severe Bullying, considering user-specific factors.Our LSTM model is trained using 146 features, incorporating emotional, topical, and word2vec representations of social media comments as well as user-level attributes and it outperforms existing baseline models, achieving the highest accuracy of 98\% and an F1-score of 0.97. To identify key factors influencing the severity of cyberbullying, we employ explainable AI techniques (SHAP and LIME) to interpret the model's decision-making process. Our findings reveal that, beyond hate comments, victims belonging to specific racial and gender groups are more frequently targeted and exhibit higher incidences of depression, disciplinary issues, and low self-esteem. Additionally, individuals with a prior history of bullying are at a greater risk of becoming victims of cyberbullying.
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.
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: Takehiro Takayanagi, Kiyoshi Izumi, Javier Sanz-Cruzado, Richard McCreadie, Iadh Ounis
Abstract: Large language model-based agents are becoming increasingly popular as a low-cost mechanism to provide personalized, conversational advice, and have demonstrated impressive capabilities in relatively simple scenarios, such as movie recommendations. But how do these agents perform in complex high-stakes domains, where domain expertise is essential and mistakes carry substantial risk? This paper investigates the effectiveness of LLM-advisors in the finance domain, focusing on three distinct challenges: (1) eliciting user preferences when users themselves may be unsure of their needs, (2) providing personalized guidance for diverse investment preferences, and (3) leveraging advisor personality to build relationships and foster trust. Via a lab-based user study with 64 participants, we show that LLM-advisors often match human advisor performance when eliciting preferences, although they can struggle to resolve conflicting user needs. When providing personalized advice, the LLM was able to positively influence user behavior, but demonstrated clear failure modes. Our results show that accurate preference elicitation is key, otherwise, the LLM-advisor has little impact, or can even direct the investor toward unsuitable assets. More worryingly, users appear insensitive to the quality of advice being given, or worse these can have an inverse relationship. Indeed, users reported a preference for and increased satisfaction as well as emotional trust with LLMs adopting an extroverted persona, even though those agents provided worse advice.
Authors: Zhiqiu Xia, Lang Zhu, Bingzhe Li, Feng Chen, Qiannan Li, Chunhua Liao, Feiyi Wang, Hang Liu
Abstract: Large Language Models (LLMs) are reshaping the landscape of computer science research, driving significant shifts in research priorities across diverse conferences and fields. This study provides a comprehensive analysis of the publication trend of LLM-related papers in 77 top-tier computer science conferences over the past six years (2019-2024). We approach this analysis from four distinct perspectives: (1) We investigate how LLM research is driving topic shifts within major conferences. (2) We adopt a topic modeling approach to identify various areas of LLM-related topic growth and reveal the topics of concern at different conferences. (3) We explore distinct contribution patterns of academic and industrial institutions. (4) We study the influence of national origins on LLM development trajectories. Synthesizing the findings from these diverse analytical angles, we derive ten key insights that illuminate the dynamics and evolution of the LLM research ecosystem.
Authors: Junlei Zhang, Zichen Ding, Chang Ma, Zijie Chen, Qiushi Sun, Zhenzhong Lan, Junxian He
Abstract: Graphical User Interface (GUI) agents offer cross-platform solutions for automating complex digital tasks, with significant potential to transform productivity workflows. However, their performance is often constrained by the scarcity of high-quality trajectory data. To address this limitation, we propose training Vision Language Models (VLMs) on data-rich, reasoning-intensive tasks during a dedicated mid-training stage, and then examine how incorporating these tasks facilitates generalization to GUI planning scenarios. Specifically, we explore a range of tasks with readily available instruction-tuning data, including GUI perception, multimodal reasoning, and textual reasoning. Through extensive experiments across 11 mid-training tasks, we demonstrate that: (1) Task generalization proves highly effective, yielding substantial improvements across most settings. For instance, multimodal mathematical reasoning enhances performance on AndroidWorld by an absolute 6.3%. Remarkably, text-only mathematical data significantly boosts GUI web agent performance, achieving a 5.6% improvement on WebArena and 5.4% improvement on AndroidWorld, underscoring notable cross-modal generalization from text-based to visual domains; (2) Contrary to prior assumptions, GUI perception data - previously considered closely aligned with GUI agent tasks and widely utilized for training - has a comparatively limited impact on final performance; (3) Building on these insights, we identify the most effective mid-training tasks and curate optimized mixture datasets, resulting in absolute performance gains of 8.0% on WebArena and 12.2% on AndroidWorld. Our work provides valuable insights into cross-domain knowledge transfer for GUI agents and offers a practical approach to addressing data scarcity challenges in this emerging field. The code, data and models will be available at https://github.com/hkust-nlp/GUIMid.
Authors: Xiaobo Xia, Run Luo
Abstract: Existing efforts in building Graphical User Interface (GUI) agents largely rely on the training paradigm of supervised fine-tuning on Large Vision-Language Models (LVLMs). However, this approach not only demands extensive amounts of training data but also struggles to effectively understand GUI screenshots and generalize to unseen interfaces. The issue significantly limits its application in real-world scenarios, especially for high-level tasks. Inspired by Reinforcement Fine-Tuning (RFT) in large reasoning models (e.g., DeepSeek-R1), which efficiently enhances the problem-solving capabilities of large language models in real-world settings, we propose \name, the first reinforcement learning framework designed to enhance the GUI capabilities of LVLMs in high-level real-world task scenarios, through unified action space rule modeling. By leveraging a small amount of carefully curated high-quality data across multiple platforms (including Windows, Linux, MacOS, Android, and Web) and employing policy optimization algorithms such as Group Relative Policy Optimization (GRPO) to update the model, \name achieves superior performance using only 0.02\% of the data (3K vs. 13M) compared to previous state-of-the-art methods like OS-Atlas across eight benchmarks spanning three different platforms (mobile, desktop, and web). These results demonstrate the immense potential of reinforcement learning based on unified action space rule modeling in improving the execution capabilities of LVLMs for real-world GUI agent tasks.