Authors: Adilzhan Adilkhanov, Amir Yelenov, Assylkhan Seitzhanov, Ayan Mazhitov, Azamat Abdikarimov, Danissa Sandykbayeva, Daryn Kenzhebek, Daulet Baimukashev, Dinmukhammed Mukashev, Ilyas Umurbekov, Jabrail Chumakov, Kamila Spanova, Karina Burunchina, Rasul Yermagambet, Rustam Chibar, Saltanat Seitzhan, Soibkhon Khajikhanov, Tasbolat Taunyazov, Temirlan Galimzhanov, Temirlan Kaiyrbay, Tleukhan Mussin, Togzhan Syrymova, Valeriya Kostyukova, Yermakhan Kassym, Madina Yergibay, Margulan Issa, Moldir Zabirova, Nurdaulet Zhuzbay, Nurlan Kabdyshev, Nurlan Zhaniyar, Yerkebulan Massalim, Zerde Nurbayeva, Zhanat Kappassov
Abstract: This paper presents an AI-generated review of Vision-Language-Action (VLA) models, summarizing key methodologies, findings, and future directions. The content is produced using large language models (LLMs) and is intended only for demonstration purposes. This work does not represent original research, but highlights how AI can help automate literature reviews. As AI-generated content becomes more prevalent, ensuring accuracy, reliability, and proper synthesis remains a challenge. Future research will focus on developing a structured framework for AI-assisted literature reviews, exploring techniques to enhance citation accuracy, source credibility, and contextual understanding. By examining the potential and limitations of LLM in academic writing, this study aims to contribute to the broader discussion of integrating AI into research workflows. This work serves as a preliminary step toward establishing systematic approaches for leveraging AI in literature review generation, making academic knowledge synthesis more efficient and scalable.
Authors: Jinyu Xiang, Jiayi Zhang, Zhaoyang Yu, Fengwei Teng, Jinhao Tu, Xinbing Liang, Sirui Hong, Chenglin Wu, Yuyu Luo
Abstract: Well-designed prompts are crucial for enhancing Large language models' (LLMs) reasoning capabilities while aligning their outputs with task requirements across diverse domains. However, manually designed prompts require expertise and iterative experimentation. While existing prompt optimization methods aim to automate this process, they rely heavily on external references such as ground truth or by humans, limiting their applicability in real-world scenarios where such data is unavailable or costly to obtain. To address this, we propose Self-Supervised Prompt Optimization (SPO), a cost-efficient framework that discovers effective prompts for both closed and open-ended tasks without requiring external reference. Motivated by the observations that prompt quality manifests directly in LLM outputs and LLMs can effectively assess adherence to task requirements, we derive evaluation and optimization signals purely from output comparisons. Specifically, SPO selects superior prompts through pairwise output comparisons evaluated by an LLM evaluator, followed by an LLM optimizer that aligns outputs with task requirements. Extensive experiments demonstrate that SPO outperforms state-of-the-art prompt optimization methods, achieving comparable or superior results with significantly lower costs (e.g., 1.1% to 5.6% of existing methods) and fewer samples (e.g., three samples). The code is available at https://github.com/geekan/MetaGPT.
Authors: Finnian Westenfelder, Erik Hemberg, Miguel Tulla, Stephen Moskal, Una-May O'Reilly, Silviu Chiricescu
Abstract: The Bourne-Again Shell (Bash) command-line interface for Linux systems has complex syntax and requires extensive specialized knowledge. Using the natural language to Bash command (NL2SH) translation capabilities of large language models (LLMs) for command composition circumvents these issues. However, the NL2SH performance of LLMs is difficult to assess due to inaccurate test data and unreliable heuristics for determining the functional equivalence of Bash commands. We present a manually verified test dataset of 600 instruction-command pairs and a training dataset of 40,939 pairs, increasing the size of previous datasets by 441% and 135%, respectively. Further, we present a novel functional equivalence heuristic that combines command execution with LLM evaluation of command outputs. Our heuristic can determine the functional equivalence of two Bash commands with 95% confidence, a 16% increase over previous heuristics. Evaluation of popular LLMs using our test dataset and heuristic demonstrates that parsing, in-context learning, in-weight learning, and constrained decoding can improve NL2SH accuracy by up to 32%. Our findings emphasize the importance of dataset quality, execution-based evaluation and translation method for advancing NL2SH translation. Our code is available at https://github.com/westenfelder/NL2SH
Authors: Xiangrong Zhu, Yuexiang Xie, Yi Liu, Yaliang Li, Wei Hu
Abstract: Retrieval-augmented generation (RAG) has emerged as a promising technology for addressing hallucination issues in the responses generated by large language models (LLMs). Existing studies on RAG primarily focus on applying semantic-based approaches to retrieve isolated relevant chunks, which ignore their intrinsic relationships. In this paper, we propose a novel Knowledge Graph-Guided Retrieval Augmented Generation (KG$^2$RAG) framework that utilizes knowledge graphs (KGs) to provide fact-level relationships between chunks, improving the diversity and coherence of the retrieved results. Specifically, after performing a semantic-based retrieval to provide seed chunks, KG$^2$RAG employs a KG-guided chunk expansion process and a KG-based chunk organization process to deliver relevant and important knowledge in well-organized paragraphs. Extensive experiments conducted on the HotpotQA dataset and its variants demonstrate the advantages of KG$^2$RAG compared to existing RAG-based approaches, in terms of both response quality and retrieval quality.
Authors: David Noever, Forrest McKee
Abstract: The development of robust safety benchmarks for large language models requires open, reproducible datasets that can measure both appropriate refusal of harmful content and potential over-restriction of legitimate scientific discourse. We present an open-source dataset and testing framework for evaluating LLM safety mechanisms across mainly controlled substance queries, analyzing four major models' responses to systematically varied prompts. Our results reveal distinct safety profiles: Claude-3.5-sonnet demonstrated the most conservative approach with 73% refusals and 27% allowances, while Mistral attempted to answer 100% of queries. GPT-3.5-turbo showed moderate restriction with 10% refusals and 90% allowances, and Grok-2 registered 20% refusals and 80% allowances. Testing prompt variation strategies revealed decreasing response consistency, from 85% with single prompts to 65% with five variations. This publicly available benchmark enables systematic evaluation of the critical balance between necessary safety restrictions and potential over-censorship of legitimate scientific inquiry, while providing a foundation for measuring progress in AI safety implementation. Chain-of-thought analysis reveals potential vulnerabilities in safety mechanisms, highlighting the complexity of implementing robust safeguards without unduly restricting desirable and valid scientific discourse.
Authors: Zenghao Duan, Wenbin Duan, Zhiyi Yin, Yinghan Shen, Shaoling Jing, Jie Zhang, Huawei Shen, Xueqi Cheng
Abstract: Knowledge editing has become a promising approach for efficiently and precisely updating knowledge embedded in large language models (LLMs). In this work, we focus on Same-Subject Editing, which involves modifying multiple attributes of a single entity to ensure comprehensive and consistent updates to entity-centric knowledge. Through preliminary observation, we identify a significant challenge: Current state-of-the-art editing methods struggle when tasked with editing multiple related knowledge pieces for the same subject. To address the lack of relevant editing data for identical subjects in traditional benchmarks, we introduce the $\text{S}^2\text{RKE}$(Same-Subject Related Knowledge Editing) benchmark. Our extensive experiments reveal that only mainstream locate-then-edit methods, such as ROME and MEMIT, exhibit "related knowledge perturbation," where subsequent edits interfere with earlier ones. Further analysis reveals that these methods over-rely on subject information, neglecting other critical factors, resulting in reduced editing effectiveness.
Authors: Bo Ni, Zheyuan Liu, Leyao Wang, Yongjia Lei, Yuying Zhao, Xueqi Cheng, Qingkai Zeng, Luna Dong, Yinglong Xia, Krishnaram Kenthapadi, Ryan Rossi, Franck Dernoncourt, Md Mehrab Tanjim, Nesreen Ahmed, Xiaorui Liu, Wenqi Fan, Erik Blasch, Yu Wang, Meng Jiang, Tyler Derr
Abstract: Retrieval-Augmented Generation (RAG) is an advanced technique designed to address the challenges of Artificial Intelligence-Generated Content (AIGC). By integrating context retrieval into content generation, RAG provides reliable and up-to-date external knowledge, reduces hallucinations, and ensures relevant context across a wide range of tasks. However, despite RAG's success and potential, recent studies have shown that the RAG paradigm also introduces new risks, including robustness issues, privacy concerns, adversarial attacks, and accountability issues. Addressing these risks is critical for future applications of RAG systems, as they directly impact their trustworthiness. Although various methods have been developed to improve the trustworthiness of RAG methods, there is a lack of a unified perspective and framework for research in this topic. Thus, in this paper, we aim to address this gap by providing a comprehensive roadmap for developing trustworthy RAG systems. We place our discussion around five key perspectives: reliability, privacy, safety, fairness, explainability, and accountability. For each perspective, we present a general framework and taxonomy, offering a structured approach to understanding the current challenges, evaluating existing solutions, and identifying promising future research directions. To encourage broader adoption and innovation, we also highlight the downstream applications where trustworthy RAG systems have a significant impact.
Authors: Subin Kim, Hoonrae Kim, Heejin Do, Gary Geunbae Lee
Abstract: Previous research has revealed the potential of large language models (LLMs) to support cognitive reframing therapy; however, their focus was primarily on text-based methods, often overlooking the importance of non-verbal evidence crucial in real-life therapy. To alleviate this gap, we extend the textual cognitive reframing to multimodality, incorporating visual clues. Specifically, we present a new dataset called Multi Modal-Cognitive Support Conversation (M2CoSC), which pairs each GPT-4-generated dialogue with an image that reflects the virtual client's facial expressions. To better mirror real psychotherapy, where facial expressions lead to interpreting implicit emotional evidence, we propose a multi-hop psychotherapeutic reasoning approach that explicitly identifies and incorporates subtle evidence. Our comprehensive experiments with both LLMs and vision-language models (VLMs) demonstrate that the VLMs' performance as psychotherapists is significantly improved with the M2CoSC dataset. Furthermore, the multi-hop psychotherapeutic reasoning method enables VLMs to provide more thoughtful and empathetic suggestions, outperforming standard prompting methods.
Authors: Yanming Guo, Xiao Qian, Kevin Credit, Jin Ma
Abstract: Accurate greenhouse gas (GHG) emission reporting is critical for governments, businesses, and investors. However, adoption remains limited particularly among small and medium enterprises due to high implementation costs, fragmented emission factor databases, and a lack of robust sector classification methods. To address these challenges, we introduce Group Reasoning Emission Estimation Networks (GREEN), an AI-driven carbon accounting framework that standardizes enterprise-level emission estimation, constructs a large-scale benchmark dataset, and leverages a novel reasoning approach with large language models (LLMs). Specifically, we compile textual descriptions for 20,850 companies with validated North American Industry Classification System (NAICS) labels and align these with an economic model of carbon intensity factors. By reframing sector classification as an information retrieval task, we fine-tune Sentence-BERT models using a contrastive learning loss. To overcome the limitations of single-stage models in handling thousands of hierarchical categories, we propose a Group Reasoning method that ensembles LLM classifiers based on the natural NAICS ontology, decomposing the task into multiple sub-classification steps. We theoretically prove that this approach reduces classification uncertainty and computational complexity. Experiments on 1,114 NAICS categories yield state-of-the-art performance (83.68% Top-1, 91.47% Top-10 accuracy), and case studies on 20 companies report a mean absolute percentage error (MAPE) of 45.88%. The project is available at: https://huggingface.co/datasets/Yvnminc/ExioNAICS.
Authors: Jinluan Yang, Dingnan Jin, Anke Tang, Li Shen, Didi Zhu, Zhengyu Chen, Daixin Wang, Qing Cui, Zhiqiang Zhang, Jun Zhou, Fei Wu, Kun Kuang
Abstract: Achieving balanced alignment of large language models (LLMs) in terms of Helpfulness, Honesty, and Harmlessness (3H optimization) constitutes a cornerstone of responsible AI, with existing methods like data mixture strategies facing limitations including reliance on expert knowledge and conflicting optimization signals. While model merging offers a promising alternative by integrating specialized models, its potential for 3H optimization remains underexplored. This paper establishes the first comprehensive benchmark for model merging in 3H-aligned LLMs, systematically evaluating 15 methods (12 training-free merging and 3 data mixture techniques) across 10 datasets associated with 5 annotation dimensions, 2 LLM families, and 2 training paradigms. Our analysis reveals three pivotal insights: (i) previously overlooked collaborative/conflicting relationships among 3H dimensions, (ii) the consistent superiority of model merging over data mixture approaches in balancing alignment trade-offs, and (iii) the critical role of parameter-level conflict resolution through redundant component pruning and outlier mitigation. Building on these findings, we propose R-TSVM, a Reweighting-enhanced Task Singular Vector Merging method that incorporates outlier-aware parameter weighting and sparsity-adaptive rank selection strategies adapted to the heavy-tailed parameter distribution and sparsity for LLMs, further improving LLM alignment across multiple evaluations. Our models will be available at https://huggingface.co/Jinluan.
Authors: Shengbin Yue, Ting Huang, Zheng Jia, Siyuan Wang, Shujun Liu, Yun Song, Xuanjing Huang, Zhongyu Wei
Abstract: Large Language Models (LLMs) have significantly advanced legal intelligence, but the scarcity of scenario data impedes the progress toward interactive legal scenarios. This paper introduces a Multi-agent Legal Simulation Driver (MASER) to scalably generate synthetic data by simulating interactive legal scenarios. Leveraging real-legal case sources, MASER ensures the consistency of legal attributes between participants and introduces a supervisory mechanism to align participants' characters and behaviors as well as addressing distractions. A Multi-stage Interactive Legal Evaluation (MILE) benchmark is further constructed to evaluate LLMs' performance in dynamic legal scenarios. Extensive experiments confirm the effectiveness of our framework.
Authors: Peng Cui, Mrinmaya Sachan
Abstract: In this paper, we introduce a learning analytics framework to analyze the in-context learning (ICL) behavior of large language models (LLMs) through the lens of the Zone of Proximal Development (ZPD), an established theory in educational psychology. ZPD delineates the space between what a learner is capable of doing unsupported and what the learner cannot do even with support. We adapt this concept to ICL, measuring the ZPD of LLMs based on model performance on individual examples with and without ICL. Furthermore, we propose an item response theory (IRT) model to predict the distribution of zones for LLMs. Our findings reveal a series of intricate and multifaceted behaviors of ICL, providing new insights into understanding and leveraging this technique. Finally, we demonstrate how our framework can enhance LLM in both inference and fine-tuning scenarios: (1) By predicting a model's zone of proximal development, we selectively apply ICL to queries that are most likely to benefit from demonstrations, achieving a better balance between inference cost and performance; (2) We propose a human-like curriculum for fine-tuning, which prioritizes examples within the model's ZPD. The curriculum results in improved performance, and we explain its effectiveness through an analysis of the training dynamics of LLMs.
Authors: Haoqi Wang, Tong Zhang, Mathieu Salzmann
Abstract: Large transformer models are known to produce high-norm tokens. In vision transformers (ViTs), such tokens have been mathematically modeled through the singular vectors of the linear approximations of layers. However, in large language models (LLMs), the underlying causes of high-norm tokens remain largely unexplored, and their different properties from those of ViTs require a new analysis framework. In this paper, we provide both theoretical insights and empirical validation across a range of recent models, leading to the following observations: i) The layer-wise singular direction predicts the abrupt explosion of token norms in LLMs. ii) The negative eigenvalues of a layer explain its sudden decay. iii) The computational pathways leading to high-norm tokens differ between initial and noninitial tokens. iv) High-norm tokens are triggered by the right leading singular vector of the matrix approximating the corresponding modules. We showcase two practical applications of these findings: the improvement of quantization schemes and the design of LLM signatures. Our findings not only advance the understanding of singular defects in LLMs but also open new avenues for their application. We expect that this work will stimulate further research into the internal mechanisms of LLMs and will therefore publicly release our code.
Authors: Hotaka Maeda, Yikai Lu
Abstract: We fine-tuned and compared several encoder-based Transformer large language models (LLM) to predict differential item functioning (DIF) from the item text. We then applied explainable artificial intelligence (XAI) methods to these models to identify specific words associated with DIF. The data included 42,180 items designed for English language arts and mathematics summative state assessments among students in grades 3 to 11. Prediction $R^2$ ranged from .04 to .32 among eight focal and reference group pairs. Our findings suggest that many words associated with DIF reflect minor sub-domains included in the test blueprint by design, rather than construct-irrelevant item content that should be removed from assessments. This may explain why qualitative reviews of DIF items often yield confusing or inconclusive results. Our approach can be used to screen words associated with DIF during the item-writing process for immediate revision, or help review traditional DIF analysis results by highlighting key words in the text. Extensions of this research can enhance the fairness of assessment programs, especially those that lack resources to build high-quality items, and among smaller subpopulations where we do not have sufficient sample sizes for traditional DIF analyses.
Authors: Adriana Eufrosiana Bora, Pierre-Luc St-Charles, Mirko Bronzi, Ars\`ene Fansi Tchango, Bruno Rousseau, Kerrie Mengersen
Abstract: Despite over a decade of legislative efforts to address modern slavery in the supply chains of large corporations, the effectiveness of government oversight remains hampered by the challenge of scrutinizing thousands of statements annually. While Large Language Models (LLMs) can be considered a well established solution for the automatic analysis and summarization of documents, recognizing concrete modern slavery countermeasures taken by companies and differentiating those from vague claims remains a challenging task. To help evaluate and fine-tune LLMs for the assessment of corporate statements, we introduce a dataset composed of 5,731 modern slavery statements taken from the Australian Modern Slavery Register and annotated at the sentence level. This paper details the construction steps for the dataset that include the careful design of annotation specifications, the selection and preprocessing of statements, and the creation of high-quality annotation subsets for effective model evaluations. To demonstrate our dataset's utility, we propose a machine learning methodology for the detection of sentences relevant to mandatory reporting requirements set by the Australian Modern Slavery Act. We then follow this methodology to benchmark modern language models under zero-shot and supervised learning settings.
Authors: Kwanghee Choi, Eunjung Yeo, Kalvin Chang, Shinji Watanabe, David Mortensen
Abstract: Allophony refers to the variation in the phonetic realization of a phoneme based on its phonetic environment. Modeling allophones is crucial for atypical pronunciation assessment, which involves distinguishing atypical from typical pronunciations. However, recent phoneme classifier-based approaches often simplify this by treating various realizations as a single phoneme, bypassing the complexity of modeling allophonic variation. Motivated by the acoustic modeling capabilities of frozen self-supervised speech model (S3M) features, we propose MixGoP, a novel approach that leverages Gaussian mixture models to model phoneme distributions with multiple subclusters. Our experiments show that MixGoP achieves state-of-the-art performance across four out of five datasets, including dysarthric and non-native speech. Our analysis further suggests that S3M features capture allophonic variation more effectively than MFCCs and Mel spectrograms, highlighting the benefits of integrating MixGoP with S3M features.
Authors: M. Ali Bayram, Ali Arda Fincan, Ahmet Semih G\"um\"u\c{s}, Sercan Karaka\c{s}, Banu Diri, Sava\c{s} Y{\i}ld{\i}r{\i}m
Abstract: Tokenization is a fundamental preprocessing step in NLP, directly impacting large language models' (LLMs) ability to capture syntactic, morphosyntactic, and semantic structures. This paper introduces a novel framework for systematically evaluating tokenization strategies, addressing challenges in morphologically rich and low-resource languages. Using a Turkish dataset of 6,200 multiple-choice questions from the Massive Multitask Language Understanding (MMLU) benchmark, the framework assesses tokenizers across five key metrics: vocabulary size, token count, processing time, language-specific token percentages (\%TR), and token purity. These metrics provide a structured approach to evaluating how well tokenizers preserve linguistic structures. While \%TR measures the proportion of valid words in the target language, \%Pure assesses the alignment of tokens with meaningful linguistic units, such as roots and valid morphemes, minimizing semantic fragmentation. The findings reveal that \%TR, introduced as a critical metric, exhibits a stronger correlation with downstream performance (e.g., MMLU scores) than token purity, emphasizing its role in improving model accuracy. Additionally, larger model parameters do not necessarily yield better tokenization quality or enhanced results, highlighting the importance of tailored tokenization strategies that prioritize linguistic alignment. This framework sets a new standard for developing robust tokenization methods optimized for morphologically complex and low-resource languages. Future work will refine morphological analysis, explore domain-specific customizations, and conduct cross-linguistic evaluations to further enhance tokenization practices.
Authors: Zixin Tang, Chieh-Yang Huang, Tsung-Chi Li, Ho Yim Sam Ng, Hen-Hsen Huang, Ting-Hao 'Kenneth' Huang
Abstract: A language can have different varieties. These varieties can affect the performance of natural language processing (NLP) models, including large language models (LLMs), which are often trained on data from widely spoken varieties. This paper introduces a novel and cost-effective approach to benchmark model performance across language varieties. We argue that international online review platforms, such as Booking.com, can serve as effective data sources for constructing datasets that capture comments in different language varieties from similar real-world scenarios, like reviews for the same hotel with the same rating using the same language (e.g., Mandarin Chinese) but different language varieties (e.g., Taiwan Mandarin, Mainland Mandarin). To prove this concept, we constructed a contextually aligned dataset comprising reviews in Taiwan Mandarin and Mainland Mandarin and tested six LLMs in a sentiment analysis task. Our results show that LLMs consistently underperform in Taiwan Mandarin.
Authors: Yong Cao, Haijiang Liu, Arnav Arora, Isabelle Augenstein, Paul R\"ottger, Daniel Hershcovich
Abstract: Large-scale surveys are essential tools for informing social science research and policy, but running surveys is costly and time-intensive. If we could accurately simulate group-level survey results, this would therefore be very valuable to social science research. Prior work has explored the use of large language models (LLMs) for simulating human behaviors, mostly through prompting. In this paper, we are the first to specialize LLMs for the task of simulating survey response distributions. As a testbed, we use country-level results from two global cultural surveys. We devise a fine-tuning method based on first-token probabilities to minimize divergence between predicted and actual response distributions for a given question. Then, we show that this method substantially outperforms other methods and zero-shot classifiers, even on unseen questions, countries, and a completely unseen survey. While even our best models struggle with the task, especially on unseen questions, our results demonstrate the benefits of specialization for simulation, which may accelerate progress towards sufficiently accurate simulation in the future.
Authors: Sayem Mohammad Imtiaz, Astha Singh, Fraol Batole, Hridesh Rajan
Abstract: Not a day goes by without hearing about the impressive feats of large language models (LLMs), and equally, not a day passes without hearing about their challenges. LLMs are notoriously vulnerable to biases in their dataset, leading to issues such as toxicity. While domain-adaptive training has been employed to mitigate these issues, these techniques often address all model parameters indiscriminately during the repair process, resulting in poor repair quality and reduced model versatility. In this paper, we introduce a novel dynamic slicing-based intent-aware LLM repair strategy, IRepair. This approach selectively targets the most error-prone sections of the model for repair. Specifically, we propose dynamically slicing the model's most sensitive layers that require immediate attention, concentrating repair efforts on those areas. This method enables more effective repairs with potentially less impact on the model's overall performance by altering a smaller portion of the model. We evaluated our technique on three models from the GPT2 and GPT-Neo families, with parameters ranging from 800M to 1.6B, in a toxicity mitigation setup. Our results show that IRepair repairs errors 43.6% more effectively while causing 46% less disruption to general performance compared to the closest baseline, direct preference optimization. Our empirical analysis also reveals that errors are more concentrated in a smaller section of the model, with the top 20% of layers exhibiting 773% more error density than the remaining 80\%. This highlights the need for selective repair. Additionally, we demonstrate that a dynamic selection approach is essential for addressing errors dispersed throughout the model, ensuring a robust and efficient repair.
Authors: Lujain Ibrahim, Canfer Akbulut, Rasmi Elasmar, Charvi Rastogi, Minsuk Kahng, Meredith Ringel Morris, Kevin R. McKee, Verena Rieser, Murray Shanahan, Laura Weidinger
Abstract: The tendency of users to anthropomorphise large language models (LLMs) is of growing interest to AI developers, researchers, and policy-makers. Here, we present a novel method for empirically evaluating anthropomorphic LLM behaviours in realistic and varied settings. Going beyond single-turn static benchmarks, we contribute three methodological advances in state-of-the-art (SOTA) LLM evaluation. First, we develop a multi-turn evaluation of 14 anthropomorphic behaviours. Second, we present a scalable, automated approach by employing simulations of user interactions. Third, we conduct an interactive, large-scale human subject study (N=1101) to validate that the model behaviours we measure predict real users' anthropomorphic perceptions. We find that all SOTA LLMs evaluated exhibit similar behaviours, characterised by relationship-building (e.g., empathy and validation) and first-person pronoun use, and that the majority of behaviours only first occur after multiple turns. Our work lays an empirical foundation for investigating how design choices influence anthropomorphic model behaviours and for progressing the ethical debate on the desirability of these behaviours. It also showcases the necessity of multi-turn evaluations for complex social phenomena in human-AI interaction.
Authors: Marina Ribeiro, B\'arbara Malcorra, Diego Pintor, Nat\'alia Bezerra Mota
Abstract: Selecting an appropriate book is crucial for fostering reading habits in children. While children exhibit varying levels of complexity when generating oral narratives, the question arises: do children's books also differ in narrative complexity? This study explores the narrative dynamics of literary texts used in schools, focusing on how their complexity evolves across different grade levels. Using Word-Recurrence Graph Analysis, we examined a dataset of 1,627 literary texts spanning 13 years of education. The findings reveal significant exponential growth in connectedness, particularly during the first three years of schooling, mirroring patterns observed in children's oral narratives. These results highlight the potential of literary texts as a tool to support the development of literacy skills.
Authors: Saurabh Kumar Pandey, Sachin Vashistha, Debrup Das, Somak Aditya, Monojit Choudhury
Abstract: To understand the complexity of sequence classification tasks, Hahn et al. (2021) proposed sensitivity as the number of disjoint subsets of the input sequence that can each be individually changed to change the output. Though effective, calculating sensitivity at scale using this framework is costly because of exponential time complexity. Therefore, we introduce a Sensitivity-based Multi-Armed Bandit framework (SMAB), which provides a scalable approach for calculating word-level local (sentence-level) and global (aggregated) sensitivities concerning an underlying text classifier for any dataset. We establish the effectiveness of our approach through various applications. We perform a case study on CHECKLIST generated sentiment analysis dataset where we show that our algorithm indeed captures intuitively high and low-sensitive words. Through experiments on multiple tasks and languages, we show that sensitivity can serve as a proxy for accuracy in the absence of gold data. Lastly, we show that guiding perturbation prompts using sensitivity values in adversarial example generation improves attack success rate by 15.58%, whereas using sensitivity as an additional reward in adversarial paraphrase generation gives a 12.00% improvement over SOTA approaches. Warning: Contains potentially offensive content.
Authors: Denis Bakushev, Gideon Boultinghouse, Harriet Oppenheimer, Sebastian Gillingwater, Valentina Ashington, Wilfred Stanborough
Abstract: Structured neuron encapsulation introduces a modular framework that enables more effective aggregation and specialization of information within deep learning architectures. A model modified through this framework demonstrated improved perplexity scores, greater lexical variability, and enhanced consistency in logical reasoning, suggesting that structured parameter distribution contributes to more efficient language representation. Statistical analyses of generated text highlighted a wider range of sentence structures and reduced redundancy in token selection, indicating that encapsulation fosters more adaptable language generation. A detailed evaluation of attention weight distributions revealed that the experimental model exhibited greater divergence in cross-layer activations, supporting the hypothesis that encapsulated neurons assume specialized processing roles. Logical consistency assessments further demonstrated that modular architectures mitigate contradictory outputs, reducing internal conflicts in inferred relationships between linguistic constructs. Computational trade-offs were analyzed, with results showing a minor increase in processing overhead, though improvements in parameter efficiency and structured decision-making compensated for the additional complexity. The mathematical formulation of the encapsulation mechanism confirmed that modular aggregation maintains stable convergence properties while promoting distinct functional roles for different neuron clusters.
Authors: Danrui Li, Sen Zhang, Sam S. Sohn, Kaidong Hu, Muhammad Usman, Mubbasir Kapadia
Abstract: The prototyping of computer games, particularly card games, requires extensive human effort in creative ideation and gameplay evaluation. Recent advances in Large Language Models (LLMs) offer opportunities to automate and streamline these processes. However, it remains challenging for LLMs to design novel game mechanics beyond existing databases, generate consistent gameplay environments, and develop scalable gameplay AI for large-scale evaluations. This paper addresses these challenges by introducing a comprehensive automated card game prototyping framework. The approach highlights a graph-based indexing method for generating novel game designs, an LLM-driven system for consistent game code generation validated by gameplay records, and a gameplay AI constructing method that uses an ensemble of LLM-generated action-value functions optimized through self-play. These contributions aim to accelerate card game prototyping, reduce human labor, and lower barriers to entry for game developers.
Authors: Yewon Hwang, Sungbum Jung, Hanwool Lee, Sara Yu
Abstract: Domain specificity of embedding models is critical for effective performance. However, existing benchmarks, such as FinMTEB, are primarily designed for high-resource languages, leaving low-resource settings, such as Korean, under-explored. Directly translating established English benchmarks often fails to capture the linguistic and cultural nuances present in low-resource domains. In this paper, titled TWICE: What Advantages Can Low-Resource Domain-Specific Embedding Models Bring? A Case Study on Korea Financial Texts, we introduce KorFinMTEB, a novel benchmark for the Korean financial domain, specifically tailored to reflect its unique cultural characteristics in low-resource languages. Our experimental results reveal that while the models perform robustly on a translated version of FinMTEB, their performance on KorFinMTEB uncovers subtle yet critical discrepancies, especially in tasks requiring deeper semantic understanding, that underscore the limitations of direct translation. This discrepancy highlights the necessity of benchmarks that incorporate language-specific idiosyncrasies and cultural nuances. The insights from our study advocate for the development of domain-specific evaluation frameworks that can more accurately assess and drive the progress of embedding models in low-resource settings.
Authors: Quyu Kong, Yixuan Zhang, Yang Liu, Panrong Tong, Enqi Liu, Feng Zhou
Abstract: Temporal Point Processes (TPPs) have been widely used for event sequence modeling, but they often struggle to incorporate rich textual event descriptions effectively. Conversely, while Large Language Models (LLMs) have been shown remarkable capabilities in processing textual data, they lack mechanisms for handling temporal dynamics. To bridge this gap, we introduce Language-TPP, a unified framework that integrates TPPs with LLMs for enhanced event sequence modeling. Language-TPP introduces a novel temporal encoding mechanism that converts continuous time intervals into specialized byte-tokens, enabling seamless integration with standard LLM architectures. This approach allows Language-TPP to achieve state-of-the-art performance across multiple TPP tasks, including event time prediction, type prediction, and intensity estimation, on five datasets. Additionally, we demonstrate that incorporating temporal information significantly improves the quality of generated event descriptions.
Authors: Jiayuan Zhu, Junde Wu
Abstract: Accurate and efficient diagnosis in online medical consultations remains a challenge for current large language models. These models often rely on single-turn interactions and lack the ability to refine their predictions through follow-up questions. Additionally, their responses frequently contain complex medical terminology, making them less accessible to non-medical users and creating barriers to effective communication. In this paper, we introduce Ask Patients with Patience (APP), the first multi-turn dialogue that enables LLMs to iteratively refine diagnoses based on grounded reasoning. By integrating medical guidelines and entropy minimization, APP improves both diagnostic accuracy and efficiency. Furthermore, it features human-centric communication that bridges the gap between user comprehension and medical terminology, significantly enhancing user accessibility and engagement. We evaluated APP using a subset of the ReMeDi dataset, comparing it with single-turn and traditional multi-turn LLM baselines. APP achieved higher similarity scores in diagnosis predictions, demonstrating better alignment with ground truth diagnoses. Entropy analysis showed that APP reduces diagnostic uncertainty more rapidly across iterations, increasing confidence in its predictions. APP also excels in user accessibility and empathy, further bridging the gap between complex medical language and user understanding. Code will be released at: https://github.com/SuperMedIntel/AskPatients.
Authors: Lingze Zhang, Ellie Pavlick
Abstract: An increasingly common practice is to train large language models (LLMs) using synthetic data. Often this synthetic data is produced by the same or similar LLMs as those it is being used to train. This raises the question of whether the synthetic data might in fact exacerbate certain "blindspots" by reinforcing heuristics that the LLM already encodes. In this paper, we conduct simulated experiments on the natural language inference (NLI) task with Llama-2-7B-hf models. We use MultiNLI as the general task and HANS, a targeted evaluation set designed to measure the presence of specific heuristic strategies for NLI, as our "blindspot" task. Our goal is to determine whether performance disparities between the general and blind spot tasks emerge. Our results indicate that synthetic data does not reinforce blindspots in the way we expected. Specifically, we see that, while fine-tuning with synthetic data doesn't necessarily reduce the use of the heuristic, it also does not make it worse as we hypothesized.
Authors: Peipei Wei, Dimitris Dimitriadis, Yan Xu, Mingwei Shen
Abstract: We present PRINCIPLE-BASED PROMPTING, a simple but effective multi-agent prompting strategy for text classification. It first asks multiple LLM agents to independently generate candidate principles based on analysis of demonstration samples with or without labels, consolidates them into final principles via a finalizer agent, and then sends them to a classifier agent to perform downstream classification tasks. Extensive experiments on binary and multi-class classification datasets with different sizes of LLMs show that our approach not only achieves substantial performance gains (1.55% - 19.37%) over zero-shot prompting on macro-F1 score but also outperforms other strong baselines (CoT and stepback prompting). Principles generated by our approach help LLMs perform better on classification tasks than human crafted principles on two private datasets. Our multi-agent PRINCIPLE-BASED PROMPTING approach also shows on-par or better performance compared to demonstration-based few-shot prompting approaches, yet with substantially lower inference costs. Ablation studies show that label information and the multi-agent cooperative LLM framework play an important role in generating high-quality principles to facilitate downstream classification tasks.
Authors: Yinghui Li, Haojing Huang, Jiayi Kuang, Yangning Li, Shu-Yu Guo, Chao Qu, Xiaoyu Tan, Hai-Tao Zheng, Ying Shen, Philip S. Yu
Abstract: How to alleviate the hallucinations of Large Language Models (LLMs) has always been the fundamental goal pursued by the LLMs research community. Looking through numerous hallucination-related studies, a mainstream category of methods is to reduce hallucinations by optimizing the knowledge representation of LLMs to change their output. Considering that the core focus of these works is the knowledge acquired by models, and knowledge has long been a central theme in human societal progress, we believe that the process of models refining knowledge can greatly benefit from the way humans learn. In our work, by imitating the human learning process, we design an Adaptive Contrastive Learning strategy. Our method flexibly constructs different positive and negative samples for contrastive learning based on LLMs' actual mastery of knowledge. This strategy helps LLMs consolidate the correct knowledge they already possess, deepen their understanding of the correct knowledge they have encountered but not fully grasped, forget the incorrect knowledge they previously learned, and honestly acknowledge the knowledge they lack. Extensive experiments and detailed analyses on widely used datasets demonstrate the effectiveness of our method.
Authors: Sina Salimian, Gias Uddin, Most Husne Jahan, Shaina Raza
Abstract: Zero-shot LLMs are now also used for textual classification tasks, e.g., sentiment/emotion detection of a given input as a sentence/article. However, their performance can be suboptimal in such data annotation tasks. We introduce a novel technique Perceived Confidence Scoring (PCS) that evaluates LLM's confidence for its classification of an input by leveraging Metamorphic Relations (MRs). The MRs generate semantically equivalent yet textually mutated versions of the input. Following the principles of Metamorphic Testing (MT), the mutated versions are expected to have annotation labels similar to the input. By analyzing the consistency of LLM responses across these variations, PCS computes a confidence score based on the frequency of predicted labels. PCS can be used both for single LLM and multiple LLM settings (e.g., majority voting). We introduce an algorithm Perceived Differential Evolution (PDE) that determines the optimal weights assigned to the MRs and the LLMs for a classification task. Empirical evaluation shows PCS significantly improves zero-shot accuracy for Llama-3-8B-Instruct (4.96%) and Mistral-7B-Instruct-v0.3 (10.52%), with Gemma-2-9b-it showing a 9.39% gain. When combining all three models, PCS significantly outperforms majority voting by 7.75%.
Authors: Sang Quang Nguyen, Kiet Van Nguyen
Abstract: This paper presents ViSP, a high-quality Vietnamese dataset for sentence paraphrasing, consisting of 1.2M original-paraphrase pairs collected from various domains. The dataset was constructed using a hybrid approach that combines automatic paraphrase generation with manual evaluation to ensure high quality. We conducted experiments using methods such as back-translation, EDA, and baseline models like BART and T5, as well as large language models (LLMs), including GPT-4o, Gemini-1.5, Aya, Qwen-2.5, and Meta-Llama-3.1 variants. To the best of our knowledge, this is the first large-scale study on Vietnamese paraphrasing. We hope that our dataset and findings will serve as a valuable foundation for future research and applications in Vietnamese paraphrase tasks.
Authors: Elias Lumer, Pradeep Honaganahalli Basavaraju, Myles Mason, James A. Burke, Vamse Kumar Subbiah
Abstract: Recent developments in retrieval-augmented generation (RAG) for selecting relevant tools from a tool knowledge base enable LLM agents to scale their complex tool calling capabilities to hundreds or thousands of external tools, APIs, or agents-as-tools. However, traditional RAG-based tool retrieval fails to capture structured dependencies between tools, limiting the retrieval accuracy of a retrieved tool's dependencies. For example, among a vector database of tools, a "get stock price" API requires a "stock ticker" parameter from a "get stock ticker" API, and both depend on OS-level internet connectivity tools. In this paper, we address this limitation by introducing Graph RAG-Tool Fusion, a novel plug-and-play approach that combines the strengths of vector-based retrieval with efficient graph traversal to capture all relevant tools (nodes) along with any nested dependencies (edges) within the predefined tool knowledge graph. We also present ToolLinkOS, a new tool selection benchmark of 573 fictional tools, spanning over 15 industries, each with an average of 6.3 tool dependencies. We demonstrate that Graph RAG-Tool Fusion achieves absolute improvements of 71.7% and 22.1% over na\"ive RAG on ToolLinkOS and ToolSandbox benchmarks, respectively (mAP@10). ToolLinkOS dataset is available at https://github.com/EliasLumer/Graph-RAG-Tool-Fusion-ToolLinkOS
URLs: https://github.com/EliasLumer/Graph-RAG-Tool-Fusion-ToolLinkOS
Authors: Wei Wu, Qiuyi Li, Mingyang Li, Kun Fu, Fuli Feng, Jieping Ye, Hui Xiong, Zheng Wang
Abstract: Advancements in DNA sequencing technologies have significantly improved our ability to decode genomic sequences. However, the prediction and interpretation of these sequences remain challenging due to the intricate nature of genetic material. Large language models (LLMs) have introduced new opportunities for biological sequence analysis. Recent developments in genomic language models have underscored the potential of LLMs in deciphering DNA sequences. Nonetheless, existing models often face limitations in robustness and application scope, primarily due to constraints in model structure and training data scale. To address these limitations, we present GENERator, a generative genomic foundation model featuring a context length of 98k base pairs (bp) and 1.2B parameters. Trained on an expansive dataset comprising 386B bp of eukaryotic DNA, the GENERator demonstrates state-of-the-art performance across both established and newly proposed benchmarks. The model adheres to the central dogma of molecular biology, accurately generating protein-coding sequences that translate into proteins structurally analogous to known families. It also shows significant promise in sequence optimization, particularly through the prompt-responsive generation of promoter sequences with specific activity profiles. These capabilities position the GENERator as a pivotal tool for genomic research and biotechnological advancement, enhancing our ability to interpret and predict complex biological systems and enabling precise genomic interventions.
Authors: Yelin Chen, Fanjin Zhang, Jie Tang
Abstract: Named Entity Recognition (NER) is a fundamental problem in natural language processing (NLP). However, the task of extracting longer entity spans (e.g., awards) from extended texts (e.g., homepages) is barely explored. Current NER methods predominantly fall into two categories: span-based methods and generation-based methods. Span-based methods require the enumeration of all possible token-pair spans, followed by classification on each span, resulting in substantial redundant computations and excessive GPU memory usage. In contrast, generation-based methods involve prompting or fine-tuning large language models (LLMs) to adapt to downstream NER tasks. However, these methods struggle with the accurate generation of longer spans and often incur significant time costs for effective fine-tuning. To address these challenges, this paper introduces a lightweight span-based NER method called SeNER, which incorporates a bidirectional arrow attention mechanism coupled with LogN-Scaling on the [CLS] token to embed long texts effectively, and comprises a novel bidirectional sliding-window plus-shaped attention (BiSPA) mechanism to reduce redundant candidate token-pair spans significantly and model interactions between token-pair spans simultaneously. Extensive experiments demonstrate that our method achieves state-of-the-art extraction accuracy on three long NER datasets and is capable of extracting entities from long texts in a GPU-memory-friendly manner. Code: https://github.com/THUDM/scholar-profiling/tree/main/sener
URLs: https://github.com/THUDM/scholar-profiling/tree/main/sener
Authors: Junlong Li, Daya Guo, Dejian Yang, Runxin Xu, Yu Wu, Junxian He
Abstract: Reasoning is a fundamental capability of Large Language Models. While prior research predominantly focuses on enhancing narrow skills like math or code generation, improving performance on many other reasoning tasks remains challenging due to sparse and fragmented training data. To address this issue, we propose CodeI/O, a novel approach that systematically condenses diverse reasoning patterns inherently embedded in contextually-grounded codes, through transforming the original code into a code input-output prediction format. By training models to predict inputs/outputs given code and test cases entirely in natural language as Chain-of-Thought (CoT) rationales, we expose them to universal reasoning primitives -- like logic flow planning, state-space searching, decision tree traversal, and modular decomposition -- while decoupling structured reasoning from code-specific syntax and preserving procedural rigor. Experimental results demonstrate CodeI/O leads to consistent improvements across symbolic, scientific, logic, math & numerical, and commonsense reasoning tasks. By matching the existing ground-truth outputs or re-executing the code with predicted inputs, we can verify each prediction and further enhance the CoTs through multi-turn revision, resulting in CodeI/O++ and achieving higher performance. Our data and models are available at https://github.com/hkust-nlp/CodeIO.
Authors: Zilu Dong, Xiangqing Shen, Rui Xia
Abstract: As large language models continue to scale up, knowledge editing techniques that modify models' internal knowledge without full retraining have gained significant attention. MEMIT, a prominent batch editing algorithm, stands out for its capability to perform mass knowledge modifications. However, we uncover a critical limitation that MEMIT's editing efficacy significantly deteriorates when processing batches containing multiple edits sharing the same subject. Our analysis reveals that the root cause lies in MEMIT's key value modeling framework: When multiple facts with the same subject in a batch are modeled through MEMIT's key value mechanism, identical keys (derived from the shared subject) are forced to represent different values (corresponding to different knowledge), resulting in updates conflicts during editing. Addressing this issue, we propose MEMIT-Merge, an enhanced approach that merges value computation processes for facts sharing the same subject, effectively resolving the performance degradation in same-subject batch editing scenarios. Experimental results demonstrate that when MEMIT's edit success rate drops to around 50% at larger batch sizes, MEMIT-Merge maintains a success rate exceeding 90%, showcasing remarkable robustness to subject entity collisions.
Authors: Shuzheng Si, Haozhe Zhao, Gang Chen, Cheng Gao, Yuzhuo Bai, Zhitong Wang, Kaikai An, Kangyang Luo, Chen Qian, Fanchao Qi, Baobao Chang, Maosong Sun
Abstract: Training LLMs on data that contains unfamiliar knowledge during the instruction tuning stage can make LLMs overconfident and encourage hallucinations. To address this challenge, we introduce a novel framework, NOVA, which identifies high-quality data that aligns well with the LLM's learned knowledge to reduce hallucinations. NOVA includes Internal Consistency Probing (ICP) and Semantic Equivalence Identification (SEI) to measure how familiar the LLM is with instruction data. Specifically, ICP evaluates the LLM's understanding of the given instruction by calculating the tailored consistency among multiple self-generated responses. SEI further assesses the familiarity of the LLM with the target response by comparing it to the generated responses, using the proposed semantic clustering and well-designed voting strategy. Finally, we introduce an expert-aligned reward model, considering characteristics beyond just familiarity to enhance data quality. By considering data quality and avoiding unfamiliar data, we can utilize the selected data to effectively align LLMs to follow instructions and hallucinate less. Extensive experiments and analysis show that NOVA significantly reduces hallucinations and allows LLMs to maintain a strong ability to follow instructions.
Authors: Xu Huang, Wenhao Zhu, Hanxu Hu, Conghui He, Lei Li, Shujian Huang, Fei Yuan
Abstract: Previous multilingual benchmarks focus primarily on simple understanding tasks, but for large language models(LLMs), we emphasize proficiency in instruction following, reasoning, long context understanding, code generation, and so on. However, measuring these advanced capabilities across languages is underexplored. To address the disparity, we introduce BenchMAX, a multi-way multilingual evaluation benchmark that allows for fair comparisons of these important abilities across languages. To maintain high quality, three distinct native-speaking annotators independently annotate each sample within all tasks after the data was machine-translated from English into 16 other languages. Additionally, we present a novel translation challenge stemming from dataset construction. Extensive experiments on BenchMAX reveal varying effectiveness of core capabilities across languages, highlighting performance gaps that cannot be bridged by simply scaling up model size. BenchMAX serves as a comprehensive multilingual evaluation platform, providing a promising test bed to promote the development of multilingual language models. The dataset and code are publicly accessible.
Authors: Zhiyin Tan, Jennifer D'Souza
Abstract: This study presents a framework for automated evaluation of dynamically evolving topic taxonomies in scientific literature using Large Language Models (LLMs). In digital library systems, topic modeling plays a crucial role in efficiently organizing and retrieving scholarly content, guiding researchers through complex knowledge landscapes. As research domains proliferate and shift, traditional human centric and static evaluation methods struggle to maintain relevance. The proposed approach harnesses LLMs to measure key quality dimensions, such as coherence, repetitiveness, diversity, and topic-document alignment, without heavy reliance on expert annotators or narrow statistical metrics. Tailored prompts guide LLM assessments, ensuring consistent and interpretable evaluations across various datasets and modeling techniques. Experiments on benchmark corpora demonstrate the method's robustness, scalability, and adaptability, underscoring its value as a more holistic and dynamic alternative to conventional evaluation strategies.
Authors: Zican Dong, Junyi Li, Jinhao Jiang, Mingyu Xu, Wayne Xin Zhao, Bingning Wang, Weipeng Chen
Abstract: Large language models (LLMs) have gained extended context windows through scaling positional encodings and lightweight continual pre-training. However, this often leads to degraded performance on short-text tasks, while the reasons for this degradation remain insufficiently explored. In this work, we identify two primary factors contributing to this issue: distribution drift in hidden states and attention scores, and catastrophic forgetting during continual pre-training. To address these challenges, we propose Long Context Pre-training with Restoration Distillation (LongReD), a novel approach designed to mitigate short-text performance degradation through minimizing the distribution discrepancy between the extended and original models. Besides training on long texts, LongReD distills the hidden state of selected layers from the original model on short texts. Additionally, LongReD also introduces a short-to-long distillation, aligning the output distribution on short texts with that on long texts by leveraging skipped positional indices. Experiments on common text benchmarks demonstrate that LongReD effectively preserves the model's short-text performance while maintaining comparable or even better capacity to handle long texts than baselines.
Authors: Santhosh Thottingal
Abstract: Parametric fonts are programatically defined fonts with variable parameters, pioneered by Donald Kunth with his MetaFont technology in the 1980s. While Donald Knuth's ideas in MetaFont and subsequently in MetaPost are often seen as legacy techniques from the pre-graphical user interface (GUI) era of type design, recent trends like variable fonts suggest a resurgence of certain principles. This paper explores a modern type design process built on parametric design principles, specifically using MetaPost. The author created two variable fonts with this method and released them under a free, open-source license. The paper details the methodology, workflow, and insights gained from this process.
Authors: Palaash Goel, Dushyant Singh Chauhan, Md Shad Akhtar
Abstract: Sarcasm is a linguistic phenomenon that intends to ridicule a target (e.g., entity, event, or person) in an inherent way. Multimodal Sarcasm Explanation (MuSE) aims at revealing the intended irony in a sarcastic post using a natural language explanation. Though important, existing systems overlooked the significance of the target of sarcasm in generating explanations. In this paper, we propose a Target-aUgmented shaRed fusion-Based sarcasm explanatiOn model, aka. TURBO. We design a novel shared-fusion mechanism to leverage the inter-modality relationships between an image and its caption. TURBO assumes the target of the sarcasm and guides the multimodal shared fusion mechanism in learning intricacies of the intended irony for explanations. We evaluate our proposed TURBO model on the MORE+ dataset. Comparison against multiple baselines and state-of-the-art models signifies the performance improvement of TURBO by an average margin of $+3.3\%$. Moreover, we explore LLMs in zero and one-shot settings for our task and observe that LLM-generated explanation, though remarkable, often fails to capture the critical nuances of the sarcasm. Furthermore, we supplement our study with extensive human evaluation on TURBO's generated explanations and find them out to be comparatively better than other systems.
Authors: Steffen Castle, Julian Moreno Schneider, Leonhard Hennig, Georg Rehm
Abstract: Growing concerns about climate change and sustainability are driving manufacturers to take significant steps toward reducing their carbon footprints. For these manufacturers, a first step towards this goal is to identify the environmental impact of the individual components of their products. We propose a system leveraging large language models (LLMs) to automatically map components from manufacturer Bills of Materials (BOMs) to Life Cycle Assessment (LCA) database entries by using LLMs to expand on available component information. Our approach reduces the need for manual data processing, paving the way for more accessible sustainability practices.
Authors: Alan Saji (Nilekani Centre at AI4Bharat), Jaavid Aktar Husain (Singapore University of Technology and Design), Thanmay Jayakumar (Nilekani Centre at AI4Bharat, Indian Institute of Technology Madras, India), Raj Dabre (Nilekani Centre at AI4Bharat, Indian Institute of Technology Bombay, India), Anoop Kunchukuttan (Nilekani Centre at AI4Bharat, Microsoft, India), Mitesh M. Khapra (Nilekani Centre at AI4Bharat, Indian Institute of Technology Madras, India), Ratish Puduppully (IT University of Copenhagen)
Abstract: Large Language Models (LLMs) exhibit remarkable multilingual generalization despite being predominantly trained on English-centric corpora. A fundamental question arises: how do LLMs achieve such robust multilingual capabilities? For non-Latin script languages, we investigate the role of romanization - the representation of non-Latin scripts using Latin characters - as a bridge in multilingual processing. Using mechanistic interpretability techniques, we analyze next-token generation and find that intermediate layers frequently represent target words in romanized form before transitioning to native script, a phenomenon we term Latent Romanization. Further, through activation patching experiments, we demonstrate that LLMs encode semantic concepts similarly across native and romanized scripts, suggesting a shared underlying representation. Additionally in translation towards non Latin languages, our findings reveal that when the target language is in romanized form, its representations emerge earlier in the model's layers compared to native script. These insights contribute to a deeper understanding of multilingual representation in LLMs and highlight the implicit role of romanization in facilitating language transfer. Our work provides new directions for potentially improving multilingual language modeling and interpretability.
Authors: Duong Anh Kiet
Abstract: We present our solution to the AAAI-25 VRD-IU challenge, achieving first place in the competition. Our approach integrates large margin loss for improved feature discrimination and employs heuristic rules to refine hierarchical relationships. By combining a deep learning-based matching strategy with greedy algorithms, we achieve a significant boost in accuracy while maintaining computational efficiency. Our method attains an accuracy of 0.98904 on the private leaderboard, demonstrating its effectiveness in document structure parsing. Source codes are publicly available at https://github.com/ffyyytt/VRUID-AAAI-DAKiet
Authors: Nurit Cohen-Inger, Yehonatan Elisha, Bracha Shapira, Lior Rokach, Seffi Cohen
Abstract: Large language models (LLMs) often appear to excel on public benchmarks, but these high scores may mask an overreliance on dataset-specific surface cues rather than true language understanding. We introduce the Chameleon Benchmark Overfit Detector (C-BOD), a meta-evaluation framework that systematically distorts benchmark prompts via a parametric transformation and detects overfitting of LLMs. By rephrasing inputs while preserving their semantic content and labels, C-BOD exposes whether a model's performance is driven by memorized patterns. Evaluated on the MMLU benchmark using 26 leading LLMs, our method reveals an average performance degradation of 2.15% under modest perturbations, with 20 out of 26 models exhibiting statistically significant differences. Notably, models with higher baseline accuracy exhibit larger performance differences under perturbation, and larger LLMs tend to be more sensitive to rephrasings indicating that both cases may overrely on fixed prompt patterns. In contrast, the Llama family and models with lower baseline accuracy show insignificant degradation, suggesting reduced dependency on superficial cues. Moreover, C-BOD's dataset- and model-agnostic design allows easy integration into training pipelines to promote more robust language understanding. Our findings challenge the community to look beyond leaderboard scores and prioritize resilience and generalization in LLM evaluation.
Authors: Erfan Moosavi Monazzah, Vahid Rahimzadeh, Yadollah Yaghoobzadeh, Azadeh Shakery, Mohammad Taher Pilehvar
Abstract: Large language models predominantly reflect Western cultures, largely due to the dominance of English-centric training data. This imbalance presents a significant challenge, as LLMs are increasingly used across diverse contexts without adequate evaluation of their cultural competence in non-English languages, including Persian. To address this gap, we introduce PerCul, a carefully constructed dataset designed to assess the sensitivity of LLMs toward Persian culture. PerCul features story-based, multiple-choice questions that capture culturally nuanced scenarios. Unlike existing benchmarks, PerCul is curated with input from native Persian annotators to ensure authenticity and to prevent the use of translation as a shortcut. We evaluate several state-of-the-art multilingual and Persian-specific LLMs, establishing a foundation for future research in cross-cultural NLP evaluation. Our experiments demonstrate a 11.3% gap between best closed source model and layperson baseline while the gap increases to 21.3% by using the best open-weight model. You can access the dataset from here: https://huggingface.co/datasets/teias-ai/percul
Authors: Jian Yang, Wei Zhang, Jiaxi Yang, Yibo Miao, Shanghaoran Quan, Zhenhe Wu, Qiyao Peng, Liqun Yang, Tianyu Liu, Zeyu Cui, Binyuan Hui, Junyang Lin
Abstract: Recent advancement in code understanding and generation demonstrates that code LLMs fine-tuned on a high-quality instruction dataset can gain powerful capabilities to address wide-ranging code-related tasks. However, most previous existing methods mainly view each programming language in isolation and ignore the knowledge transfer among different programming languages. To bridge the gap among different programming languages, we introduce a novel multi-agent collaboration framework to enhance multilingual instruction tuning for code LLMs, where multiple language-specific intelligent agent components with generation memory work together to transfer knowledge from one language to another efficiently and effectively. Specifically, we first generate the language-specific instruction data from the code snippets and then provide the generated data as the seed data for language-specific agents. Multiple language-specific agents discuss and collaborate to formulate a new instruction and its corresponding solution (A new programming language or existing programming language), To further encourage the cross-lingual transfer, each agent stores its generation history as memory and then summarizes its merits and faults. Finally, the high-quality multilingual instruction data is used to encourage knowledge transfer among different programming languages to train Qwen2.5-xCoder. Experimental results on multilingual programming benchmarks demonstrate the superior performance of Qwen2.5-xCoder in sharing common knowledge, highlighting its potential to reduce the cross-lingual gap.
Authors: Xialie Zhuang, Zhikai Jia, Jianjin Li, Zhenyu Zhang, Li Shen, Zheng Cao, Shiwei Liu
Abstract: Large Language Models (LLMs) are discovered to suffer from accurately retrieving key information. To address this, we propose Mask-Enhanced Autoregressive Prediction (MEAP), a simple yet effective training paradigm that seamlessly integrates Masked Language Modeling (MLM) into Next-Token Prediction (NTP) to enhance the latter's in-context retrieval capabilities. Specifically, MEAP first randomly masks a small fraction of input tokens and then directly performs the standard next-token prediction autoregressive using a decoder-only Transformer. MEAP eliminates the need for bidirectional attention or encoder-decoder architectures for MLM, incurring no additional computational overhead during pre-training or inference. Intensive experiments demonstrate that MEAP substantially outperforms NTP on key information retrieval and long-context reasoning tasks, while performing on par or better on commonsense reasoning tasks. The benefits of MEAP also extend to supervised fine-tuning, where it shows remarkable advantages in lost-in-the-middle scenarios, outperforming NTP by 11.77 percentage points. Our analysis indicates that MEAP's effectiveness arises from its ability to promote more distinguishable attention scores by concentrating on a reduced set of non-masked tokens. This mechanism improves the model's focus on task-relevant signals while mitigating the influence of peripheral context. These findings position MEAP as a promising training paradigm for large language models.
Authors: Tom Calamai, Oana Balalau, Th\'eo Le Guenedal, Fabian M. Suchanek
Abstract: Greenwashing is an effort to mislead the public about the environmental impact of an entity, such as a state or company. We provide a comprehensive survey of the scientific literature addressing natural language processing methods to identify potentially misleading climate-related corporate communications, indicative of greenwashing. We break the detection of greenwashing into intermediate tasks, and review the state-of-the-art approaches for each of them. We discuss datasets, methods, and results, as well as limitations and open challenges. We also provide an overview of how far the field has come as a whole, and point out future research directions.
Authors: Dominik Glandorf, Peng Cui, Detmar Meurers, Mrinmaya Sachan
Abstract: Chatbots based on large language models offer cheap conversation practice opportunities for language learners. However, they are hard to control for linguistic forms that correspond to learners' current needs, such as grammar. We control grammar in chatbot conversation practice by grounding a dialogue response generation model in a pedagogical repository of grammar skills. We also explore how this control helps learners to produce specific grammar. We comprehensively evaluate prompting, fine-tuning, and decoding strategies for grammar-controlled dialogue response generation. Strategically decoding Llama3 outperforms GPT-3.5 when tolerating minor response quality losses. Our simulation predicts grammar-controlled responses to support grammar acquisition adapted to learner proficiency. Existing language learning chatbots and research on second language acquisition benefit from these affordances. Code available on GitHub.
Authors: Ido Levy, Orr Paradise, Boaz Carmeli, Ron Meir, Shafi Goldwasser, Yonatan Belinkov
Abstract: Emergent Communication (EC) provides a unique window into the language systems that emerge autonomously when agents are trained to jointly achieve shared goals. However, it is difficult to interpret EC and evaluate its relationship with natural languages (NL). This study employs unsupervised neural machine translation (UNMT) techniques to decipher ECs formed during referential games with varying task complexities, influenced by the semantic diversity of the environment. Our findings demonstrate UNMT's potential to translate EC, illustrating that task complexity characterized by semantic diversity enhances EC translatability, while higher task complexity with constrained semantic variability exhibits pragmatic EC, which, although challenging to interpret, remains suitable for translation. This research marks the first attempt, to our knowledge, to translate EC without the aid of parallel data.
Authors: Ruin Yan, Zheng Liu, Defu Lian
Abstract: The growing power of large language models (LLMs) has revolutionized how people access and utilize information. Notably, the LLMs excel at performing fine-grained data representation, which facilitates precise retrieval of information. They also generate high-quality answers based on external references, enabling the production of useful knowledge. The recent introduction of reasoning models, like OpenAI O1 and DeepSeek R1, marks another leap forward, highlighting LLMs' ability to think progressively before delivering final answers. This breakthrough significantly improves the ability to address complex tasks, e.g., coding and math proofs. Inspired by this progress, we aim to develop similar capabilities for retrieval models, which hold great promise for tackling critical challenges in the field, including multi-task retrieval, zero-shot retrieval, and tasks requiring intensive reasoning of complex relationships. With this motivation, we propose a novel approach called O1 Embedder, which generates useful thoughts for the input query before making retrieval for the target documents. To realize this objective, we conquer two technical difficulties. First, we design a data synthesis workflow, creating training signals for O1 Embedder by generating initial thoughts from an LLM-expert and subsequently refining them using a retrieval committee. Second, we optimize the training process, enabling a pre-trained model to be jointly fine-tuned to generate retrieval thoughts via behavior cloning and perform dense retrieval through contrastive learning. Our approach is evaluated by comprehensive experiments, where substantial improvements are achieved across 12 popular datasets, spanning both in-domain and out-of-domain scenarios. These results highlight O1 Embedder's remarkable accuracy and generalizability, paving the way for the development of next-generation IR foundation models.
Authors: John Hewitt, Robert Geirhos, Been Kim
Abstract: This position paper argues that, in order to understand AI, we cannot rely on our existing vocabulary of human words. Instead, we should strive to develop neologisms: new words that represent precise human concepts that we want to teach machines, or machine concepts that we need to learn. We start from the premise that humans and machines have differing concepts. This means interpretability can be framed as a communication problem: humans must be able to reference and control machine concepts, and communicate human concepts to machines. Creating a shared human-machine language through developing neologisms, we believe, could solve this communication problem. Successful neologisms achieve a useful amount of abstraction: not too detailed, so they're reusable in many contexts, and not too high-level, so they convey precise information. As a proof of concept, we demonstrate how a "length neologism" enables controlling LLM response length, while a "diversity neologism" allows sampling more variable responses. Taken together, we argue that we cannot understand AI using our existing vocabulary, and expanding it through neologisms creates opportunities for both controlling and understanding machines better.
Authors: Xiliang Yang, Feng Jiang, Qianen Zhang, Lei Zhao, Xiao Li
Abstract: Direct Preference Optimization (DPO) and its variants have become increasingly popular for aligning language models with human preferences. These methods aim to teach models to better distinguish between chosen (or preferred) and rejected (or dispreferred) responses. However, prior research has identified that the probability of chosen responses often decreases during training, and this phenomenon is known as likelihood displacement. To tackle this challenge, in this work we introduce \method to controllably shift the distribution of the chosen probability. Then, we show that \method exhibits a fundamental trade-off between improving the chosen probability and sacrificing the reward margin, as supported by both theoretical analysis and experimental validation. Furthermore, we demonstrate the superiority of \method over DPO on downstream tasks such as MT-Bench and a designed win rate experiment. We believe this study shows that the likelihood displacement issue of DPO can be effectively mitigated with a simple, theoretically grounded solution. Our code is available at https://github.com/Meaquadddd/DPO-Shift.
Authors: Anji Liu, Xuejie Liu, Dayuan Zhao, Mathias Niepert, Yitao Liang, Guy Van den Broeck
Abstract: Non-autoregressive (NAR) generative models are valuable because they can handle diverse conditional generation tasks in a more principled way than their autoregressive (AR) counterparts, which are constrained by sequential dependency requirements. Recent advancements in NAR models, such as diffusion language models, have demonstrated superior performance in unconditional generation compared to AR models (e.g., GPTs) of similar sizes. However, such improvements do not always lead to improved conditional generation performance. We show that a key reason for this gap is the difficulty in generalizing to conditional probability queries unseen during training. As a result, strong unconditional generation performance does not guarantee high-quality conditional generation. This paper proposes Tractable Transformers (Tracformer), a Transformer-based generative model that is more robust to different conditional generation tasks. Unlike existing models that rely solely on global contextual features derived from full inputs, Tracformers incorporate a sparse Transformer encoder to capture both local and global contextual information. This information is routed through a decoder for conditional generation. Empirical results demonstrate that Tracformers achieve state-of-the-art conditional generation performance on text modeling compared to recent diffusion and AR model baselines.
Authors: Andr\'es Chand\'ia
Abstract: After developing a computational system for morphological analysis of the Mapuche language, and evaluating it with texts from various authors and styles, it became necessary to verify the linguistic assumptions of the source used as the basis for implementing this tool. In the present work, the primary focus is on the lexical category classification of Mapud\"ungun roots recognised as verbal in the source utilised for the development of the morphological analysis system. The results of this lexical category revision directly benefit the computational analyser, as they are implemented as soon as they are verified. Additionally, it is hoped that these results will help clarify some uncertainties about lexical categories in the Mapuche language. This work addresses a preliminary task to identify the valency of true verbal roots, the results of which will be presented in a subsequent work that complements this article.
Authors: K\"atriin Kukk, Danila Petrelli, Judit Casademont, Eric J. W. Orlowski, Micha{\l} Dzieli\'nski, Maria Jacobson
Abstract: In this study, we introduce the process for creating BiaSWE, an expert-annotated dataset tailored for misogyny detection in the Swedish language. To address the cultural and linguistic specificity of misogyny in Swedish, we collaborated with experts from the social sciences and humanities. Our interdisciplinary team developed a rigorous annotation process, incorporating both domain knowledge and language expertise, to capture the nuances of misogyny in a Swedish context. This methodology ensures that the dataset is not only culturally relevant but also aligned with broader efforts in bias detection for low-resource languages. The dataset, along with the annotation guidelines, is publicly available for further research.
Authors: Annika Simonsen, Dan Saattrup Nielsen, Hafsteinn Einarsson
Abstract: We present FoQA, a Faroese extractive question-answering (QA) dataset with 2,000 samples, created using a semi-automated approach combining Large Language Models (LLMs) and human validation. The dataset was generated from Faroese Wikipedia articles using GPT-4-turbo for initial QA generation, followed by question rephrasing to increase complexity and native speaker validation to ensure quality. We provide baseline performance metrics for FoQA across multiple models, including LLMs and BERT, demonstrating its effectiveness in evaluating Faroese QA performance. The dataset is released in three versions: a validated set of 2,000 samples, a complete set of all 10,001 generated samples, and a set of 2,395 rejected samples for error analysis.
Authors: Param Kulkarni, Yingchi Liu, Hao-Ming Fu, Shaohua Yang, Isuru Gunasekara, Matt Peloquin, Noah Spitzer-Williams, Xiaotian Zhou, Xiaozhong Liu, Zhengping Ji, Yasser Ibrahim
Abstract: Achieving a delicate balance between fostering trust in law en- forcement and protecting the rights of both officers and civilians continues to emerge as a pressing research and product challenge in the world today. In the pursuit of fairness and transparency, this study presents an innovative AI-driven system designed to generate police report drafts from complex, noisy, and multi-role dialogue data. Our approach intelligently extracts key elements of law enforcement interactions and includes them in the draft, producing structured narratives that are not only high in quality but also reinforce accountability and procedural clarity. This frame- work holds the potential to transform the reporting process, ensur- ing greater oversight, consistency, and fairness in future policing practices. A demonstration video of our system can be accessed at https://drive.google.com/file/d/1kBrsGGR8e3B5xPSblrchRGj- Y-kpCHNO/view?usp=sharing
URLs: https://drive.google.com/file/d/1kBrsGGR8e3B5xPSblrchRGj-
Authors: Imry Ziv, Nur Lan, Emmanuel Chemla, Roni Katzir
Abstract: We consider the possible role of current large language models (LLMs) in the study of human linguistic cognition. We focus on the use of such models as proxies for theories of cognition that are relatively linguistically-neutral in their representations and learning but differ from current LLMs in key ways. We illustrate this potential use of LLMs as proxies for theories of cognition in the context of two kinds of questions: (a) whether the target theory accounts for the acquisition of a given pattern from a given corpus; and (b) whether the target theory makes a given typologically-attested pattern easier to acquire than another, typologically-unattested pattern. For each of the two questions we show, building on recent literature, how current LLMs can potentially be of help, but we note that at present this help is quite limited.
Authors: MohammadHossein Rezaei, Eduardo Blanco
Abstract: Negation has been a long-standing challenge for language models. Previous studies have shown that they struggle with negation in many natural language understanding tasks. In this work, we propose a self-supervised method to make language models more robust against negation. We introduce a novel task, Next Sentence Polarity Prediction (NSPP), and a variation of the Next Sentence Prediction (NSP) task. We show that BERT and RoBERTa further pre-trained on our tasks outperform the off-the-shelf versions on nine negation-related benchmarks. Most notably, our pre-training tasks yield between 1.8% and 9.1% improvement on CondaQA, a large question-answering corpus requiring reasoning over negation.
Authors: Kshitij Gupta
Abstract: We present a novel data set, WhoDunIt, to assess the deductive reasoning capabilities of large language models (LLM) within narrative contexts. Constructed from open domain mystery novels and short stories, the dataset challenges LLMs to identify the perpetrator after reading and comprehending the story. To evaluate model robustness, we apply a range of character-level name augmentations, including original names, name swaps, and substitutions with well-known real and/or fictional entities from popular discourse. We further use various prompting styles to investigate the influence of prompting on deductive reasoning accuracy. We conduct evaluation study with state-of-the-art models, specifically GPT-4o, GPT-4-turbo, and GPT-4o-mini, evaluated through multiple trials with majority response selection to ensure reliability. The results demonstrate that while LLMs perform reliably on unaltered texts, accuracy diminishes with certain name substitutions, particularly those with wide recognition. This dataset is publicly available here.
Authors: Mohammad Ali Labbaf Khaniki, Sahabeh Saadati, Mohammad Manthouri
Abstract: This paper presents a novel Natural Language Processing (NLP) framework for enhancing medical diagnosis through the integration of advanced techniques in data augmentation, feature extraction, and classification. The proposed approach employs back-translation to generate diverse paraphrased datasets, improving robustness and mitigating overfitting in classification tasks. Leveraging Decoding-enhanced BERT with Disentangled Attention (DeBERTa) with Dynamic Contextual Positional Gating (DCPG), the model captures fine-grained contextual and positional relationships, dynamically adjusting the influence of positional information based on semantic context to produce high-quality text embeddings. For classification, an Attention-Based Feedforward Neural Network (ABFNN) is utilized, effectively focusing on the most relevant features to improve decision-making accuracy. Applied to the classification of symptoms, clinical notes, and other medical texts, this architecture demonstrates its ability to address the complexities of medical data. The combination of data augmentation, contextual embedding generation, and advanced classification mechanisms offers a robust and accurate diagnostic tool, with potential applications in automated medical diagnosis and clinical decision support. This method demonstrates the effectiveness of the proposed NLP framework for medical diagnosis, achieving remarkable results with an accuracy of 99.78%, recall of 99.72%, precision of 99.79%, and an F1-score of 99.75%. These metrics not only underscore the model's robust performance in classifying medical texts with exceptional precision and reliability but also highlight its superiority over existing methods, making it a highly promising tool for automated diagnostic systems.
Authors: Sibo Ma, Alejandro Salinas, Peter Henderson, Julian Nyarko
Abstract: We employ model pruning to examine how LLMs conceptualize racial biases, and whether a generalizable mitigation strategy for such biases appears feasible. Our analysis yields several novel insights. We find that pruning can be an effective method to reduce bias without significantly increasing anomalous model behavior. Neuron-based pruning strategies generally yield better results than approaches pruning entire attention heads. However, our results also show that the effectiveness of either approach quickly deteriorates as pruning strategies become more generalized. For instance, a model that is trained on removing racial biases in the context of financial decision-making poorly generalizes to biases in commercial transactions. Overall, our analysis suggests that racial biases are only partially represented as a general concept within language models. The other part of these biases is highly context-specific, suggesting that generalizable mitigation strategies may be of limited effectiveness. Our findings have important implications for legal frameworks surrounding AI. In particular, they suggest that an effective mitigation strategy should include the allocation of legal responsibility on those that deploy models in a specific use case.
Authors: Chenchen Gu, Xiang Lisa Li, Rohith Kuditipudi, Percy Liang, Tatsunori Hashimoto
Abstract: Prompt caching in large language models (LLMs) results in data-dependent timing variations: cached prompts are processed faster than non-cached prompts. These timing differences introduce the risk of side-channel timing attacks. For example, if the cache is shared across users, an attacker could identify cached prompts from fast API response times to learn information about other users' prompts. Because prompt caching may cause privacy leakage, transparency around the caching policies of API providers is important. To this end, we develop and conduct statistical audits to detect prompt caching in real-world LLM API providers. We detect global cache sharing across users in seven API providers, including OpenAI, resulting in potential privacy leakage about users' prompts. Timing variations due to prompt caching can also result in leakage of information about model architecture. Namely, we find evidence that OpenAI's embedding model is a decoder-only Transformer, which was previously not publicly known.
Authors: Chen Wang, Xiaokai Wei, Yexi Jiang, Frank Ong, Kevin Gao, Xiao Yu, Zheng Hui, Se-eun Yoon, Philip Yu, Michelle Gong
Abstract: With the vast and dynamic user-generated content on Roblox, creating effective game recommendations requires a deep understanding of game content. Traditional recommendation models struggle with the inconsistent and sparse nature of game text features such as titles and descriptions. Recent advancements in large language models (LLMs) offer opportunities to enhance recommendation systems by analyzing in-game text data. This paper addresses two challenges: generating high-quality, structured text features for games without extensive human annotation, and validating these features to ensure they improve recommendation relevance. We propose an approach that extracts in-game text and uses LLMs to infer attributes such as genre and gameplay objectives from raw player interactions. Additionally, we introduce an LLM-based re-ranking mechanism to assess the effectiveness of the generated text features, enhancing personalization and user satisfaction. Beyond recommendations, our approach supports applications such as user engagement-based integrity detection, already deployed in production. This scalable framework demonstrates the potential of in-game text understanding to improve recommendation quality on Roblox and adapt recommendations to its unique, user-generated ecosystem.
Authors: Gaurush Hiranandani, Haolun Wu, Subhojyoti Mukherjee, Sanmi Koyejo
Abstract: Many commercial Large Language Models (LLMs) are often closed-source, limiting developers to prompt tuning for aligning content generation with specific applications. While these models currently do not provide access to token logits, we argue that if such access were available, it would enable more powerful adaptation techniques beyond prompt engineering. In this paper, we propose a token-level probability reweighting framework that, given access to logits and a small amount of task-specific data, can effectively steer black-box LLMs toward application-specific content generation. Our approach views next-token prediction through the lens of supervised classification. We show that aligning black-box LLMs with task-specific data can be formulated as a label noise correction problem, leading to \emph{Plugin} model -- an autoregressive probability reweighting model that operates solely on logits. We provide theoretical justification for why reweighting logits alone is sufficient for task adaptation. Extensive experiments with multiple datasets, LLMs, and reweighting models demonstrate the effectiveness of our method, advocating for broader access to token logits in closed-source models.
Authors: OpenAI, :, Ahmed El-Kishky, Alexander Wei, Andre Saraiva, Borys Minaev, Daniel Selsam, David Dohan, Francis Song, Hunter Lightman, Ignasi Clavera, Jakub Pachocki, Jerry Tworek, Lorenz Kuhn, Lukasz Kaiser, Mark Chen, Max Schwarzer, Mostafa Rohaninejad, Nat McAleese, o3 contributors, Oleg M\"urk, Rhythm Garg, Rui Shu, Szymon Sidor, Vineet Kosaraju, Wenda Zhou
Abstract: We show that reinforcement learning applied to large language models (LLMs) significantly boosts performance on complex coding and reasoning tasks. Additionally, we compare two general-purpose reasoning models - OpenAI o1 and an early checkpoint of o3 - with a domain-specific system, o1-ioi, which uses hand-engineered inference strategies designed for competing in the 2024 International Olympiad in Informatics (IOI). We competed live at IOI 2024 with o1-ioi and, using hand-crafted test-time strategies, placed in the 49th percentile. Under relaxed competition constraints, o1-ioi achieved a gold medal. However, when evaluating later models such as o3, we find that o3 achieves gold without hand-crafted domain-specific strategies or relaxed constraints. Our findings show that although specialized pipelines such as o1-ioi yield solid improvements, the scaled-up, general-purpose o3 model surpasses those results without relying on hand-crafted inference heuristics. Notably, o3 achieves a gold medal at the 2024 IOI and obtains a Codeforces rating on par with elite human competitors. Overall, these results indicate that scaling general-purpose reinforcement learning, rather than relying on domain-specific techniques, offers a robust path toward state-of-the-art AI in reasoning domains, such as competitive programming.
Authors: Muhammad Umair Haider, Hammad Rizwan, Hassan Sajjad, Peizhong Ju, A. B. Siddique
Abstract: Interpreting and controlling the internal mechanisms of large language models (LLMs) is crucial for improving their trustworthiness and utility. Recent efforts have primarily focused on identifying and manipulating neurons by establishing discrete mappings between neurons and semantic concepts. However, such mappings struggle to handle the inherent polysemanticity in LLMs, where individual neurons encode multiple, distinct concepts. This makes precise control challenging and complicates downstream interventions. Through an in-depth analysis of both encoder and decoder-based LLMs across multiple text classification datasets, we uncover that while individual neurons encode multiple concepts, their activation magnitudes vary across concepts in distinct, Gaussian-like patterns. Building on this insight, we introduce NeuronLens, a novel range-based interpretation and manipulation framework that provides a finer view of neuron activation distributions to localize concept attribution within a neuron. Extensive empirical evaluations demonstrate that NeuronLens significantly reduces unintended interference, while maintaining precise control for manipulation of targeted concepts, outperforming existing methods.
Authors: Avihay Chriqui, Inbal Yahav, Dov Teeni, Ahmed Abbasi
Abstract: Attention, or prioritization of certain information items over others, is a critical element of any learning process, for both humans and machines. Given that humans continue to outperform machines in certain learning tasks, it seems plausible that machine performance could be enriched by aligning machine attention with human attention mechanisms -- yet research on this topic is sparse and has achieved only limited success. This paper proposes a new approach to address this gap, called Human-Machine Attention Learning (HuMAL). This approach involves reliance on data annotated by humans to reflect their self-perceived attention during specific tasks. We evaluate several alternative strategies for integrating such human attention data into machine learning (ML) algorithms, using a sentiment analysis task (review data from Yelp) and a personality-type classification task (data from myPersonality). The best-performing HuMAL strategy significantly enhances the task performance of fine-tuned transformer models (BERT, as well as GPT-2 and XLNET), and the benefit is particularly pronounced under challenging conditions of imbalanced or sparse labeled data. This research contributes to a deeper understanding of strategies for integrating human attention into ML models and highlights the potential of leveraging human cognition to augment ML in real-world applications.
Authors: Siyeol Jung, Taehwan Kim
Abstract: The listener head generation (LHG) task aims to generate natural nonverbal listener responses based on the speaker's multimodal cues. While prior work either rely on limited modalities (e.g. audio and facial information) or employ autoregressive approaches which have limitations such as accumulating prediction errors. To address these limitations, we propose DiffListener, a discrete diffusion based approach for non-autoregressive listener head generation. Our model takes the speaker's facial information, audio, and text as inputs, additionally incorporating facial differential information to represent the temporal dynamics of expressions and movements. With this explicit modeling of facial dynamics, DiffListener can generate coherent reaction sequences in a non-autoregressive manner. Through comprehensive experiments, DiffListener demonstrates state-of-the-art performance in both quantitative and qualitative evaluations. The user study shows that DiffListener generates natural context-aware listener reactions that are well synchronized with the speaker. The code and demo videos are available in https://siyeoljung.github.io/DiffListener
Authors: Toby Simonds
Abstract: We present Entropy Adaptive Decoding (EAD), a novel approach for efficient language model inference that dynamically switches between different-sized models based on prediction uncertainty. By monitoring rolling entropy in model logit distributions, our method identifies text regions where a smaller model suffices and switches to a larger model only when prediction uncertainty exceeds a threshold. Unlike speculative decoding approaches that maintain perfect output fidelity through verification, EAD accepts controlled output divergence in exchange for computational efficiency. Our experiments on the MATH benchmark demonstrate remarkable efficiency gains across different model families. Using the LLaMA family, we maintain 96.7\% of the 11B model's performance (50.4\% vs 52.1\%) while using it for only 43\% of tokens, decreasing computational cost by 41.5\%. These gains become more pronounced with larger size differentials in the Qwen family, where we achieve 92.9\% of the 14B model's performance (74.3\% vs 80.0\%) while using it for just 25\% of tokens, decreasing computational cost by 67\%. The consistency of these results across model pairs suggests that language model computation can be significantly optimized by selectively deploying model capacity based on local generation complexity. Our findings indicate that current approaches to model inference may be unnecessarily conservative in their pursuit of perfect output fidelity, and that accepting minor performance trade-offs can enable dramatic reductions in computational costs.
Authors: Yuqiao Wen, Yanshuai Cao, Lili Mou
Abstract: Large language models have been increasing in size due to their success in a wide range of applications. This calls for a pressing need to reduce memory usage to make them more accessible. Post-training quantization is a popular technique which uses fewer bits (e.g., 4--8 bits) to represent the model without retraining it. However, it remains a challenging task to perform quantization in an ultra-low-bit setup (e.g., 2 bits). In this paper, we propose InvarExplore, a unified framework that systematically explores different model invariance at the same time, allowing us to take advantage of the synergy between each type of invariance. Importantly, InvarExplore features a discrete search algorithm that enables us to explore permutation invariance, which is under-studied as it cannot be optimized with gradient-based methods. Results show that InvarExplore is compatible with existing state-of-the-art methods, achieving an add-on performance improvement over strong competing methods.
Authors: Hailong Jiang, Jianfeng Zhu, Yao Wan, Bo Fang, Hongyu Zhang, Ruoming Jin, Qiang Guan
Abstract: Intermediate Representations (IRs) are essential in compiler design and program analysis, yet their comprehension by Large Language Models (LLMs) remains underexplored. This paper presents a pioneering empirical study to investigate the capabilities of LLMs, including GPT-4, GPT-3, Gemma 2, LLaMA 3.1, and Code Llama, in understanding IRs. We analyze their performance across four tasks: Control Flow Graph (CFG) reconstruction, decompilation, code summarization, and execution reasoning. Our results indicate that while LLMs demonstrate competence in parsing IR syntax and recognizing high-level structures, they struggle with control flow reasoning, execution semantics, and loop handling. Specifically, they often misinterpret branching instructions, omit critical IR operations, and rely on heuristic-based reasoning, leading to errors in CFG reconstruction, IR decompilation, and execution reasoning. The study underscores the necessity for IR-specific enhancements in LLMs, recommending fine-tuning on structured IR datasets and integration of explicit control flow models to augment their comprehension and handling of IR-related tasks.
Authors: Vaibhav Mehra, Guy Laban, Hatice Gunes
Abstract: Large Language Models primarily operate through text-based inputs and outputs, yet human emotion is communicated through both verbal and non-verbal cues, including facial expressions. While Vision-Language Models analyze facial expressions from images, they are resource-intensive and may depend more on linguistic priors than visual understanding. To address this, this study investigates whether LLMs can infer affective meaning from dimensions of facial expressions-Valence and Arousal values, structured numerical representations, rather than using raw visual input. VA values were extracted using Facechannel from images of facial expressions and provided to LLMs in two tasks: (1) categorizing facial expressions into basic (on the IIMI dataset) and complex emotions (on the Emotic dataset) and (2) generating semantic descriptions of facial expressions (on the Emotic dataset). Results from the categorization task indicate that LLMs struggle to classify VA values into discrete emotion categories, particularly for emotions beyond basic polarities (e.g., happiness, sadness). However, in the semantic description task, LLMs produced textual descriptions that align closely with human-generated interpretations, demonstrating a stronger capacity for free text affective inference of facial expressions.
Authors: Xuefeng Liu, Songhao Jiang, Rick Stevens
Abstract: Drug optimization has become increasingly crucial in light of fast-mutating virus strains and drug-resistant cancer cells. Nevertheless, it remains challenging as it necessitates retaining the beneficial properties of the original drug while simultaneously enhancing desired attributes beyond its scope. In this work, we aim to tackle this challenge by introducing ScaffoldGPT, a novel Large Language Model (LLM) designed for drug optimization based on molecular scaffolds. Our work comprises three key components: (1) A three-stage drug optimization approach that integrates pretraining, finetuning, and decoding optimization. (2) A uniquely designed two-phase incremental training approach for pre-training the drug optimization LLM-based generator on molecule scaffold with enhanced performance. (3) A token-level decoding optimization strategy, TOP-N, that enabling controlled, reward-guided generation using pretrained/finetuned LLMs. Finally, by conducting a comprehensive evaluation on COVID and cancer benchmarks, we demonstrate that SCAFFOLDGPT outperforms the competing baselines in drug optimization benchmarks, while excelling in preserving the original functional scaffold and enhancing desired properties.
Authors: Jared D. T. Guerrero-Sosa, Andres Montoro-Montarroso, Francisco P. Romero, Jesus Serrano-Guerrero, Jose A. Olivas
Abstract: In the context of the rapid dissemination of multimedia content, identifying disinformation on social media platforms such as TikTok represents a significant challenge. This study introduces a hybrid framework that combines the computational power of deep learning with the interpretability of fuzzy logic to detect suspected disinformation in TikTok videos. The methodology is comprised of two core components: a multimodal feature analyser that extracts and evaluates data from text, audio, and video; and a multimodal disinformation detector based on fuzzy logic. These systems operate in conjunction to evaluate the suspicion of spreading disinformation, drawing on human behavioural cues such as body language, speech patterns, and text coherence. Two experiments were conducted: one focusing on context-specific disinformation and the other on the scalability of the model across broader topics. For each video evaluated, high-quality, comprehensive, well-structured reports are generated, providing a detailed view of the disinformation behaviours.
Authors: Daniel Israel, Aditya Grover, Guy Van den Broeck
Abstract: Historically, LLMs have been trained using either autoregressive (AR) or masked language modeling (MLM) objectives, with AR models gaining dominance in recent years. However, AR models are inherently incapable of masked infilling, which is the ability to predict masked tokens between past and future context. In contrast, MLM models suffer from intrinsic computational inefficiencies during both training and inference that hinder their scalability. This work introduces MARIA (Masked and Autoregressive Infilling Architecture), a novel approach that leverages the strengths of both paradigms to achieve state-of-the-art masked infilling performance. MARIA combines a pre-trained MLM and AR model by training a linear decoder that takes their concatenated hidden states as input. This minimal modification enables the AR model to perform infilling while retaining its inherent advantages in terms of faster inference with KV caching. Our results demonstrate that MARIA significantly outperforms existing methods, namely discrete diffusion models, on masked infilling tasks.
Authors: Deven Mahesh Mistry, Anooshka Bajaj, Yash Aggarwal, Sahaj Singh Maini, Zoran Tiganj
Abstract: We investigate in-context temporal biases in attention heads and transformer outputs. Using cognitive science methodologies, we analyze attention scores and outputs of the GPT-2 models of varying sizes. Across attention heads, we observe effects characteristic of human episodic memory, including temporal contiguity, primacy and recency. Transformer outputs demonstrate a tendency toward in-context serial recall. Importantly, this effect is eliminated after the ablation of the induction heads, which are the driving force behind the contiguity effect. Our findings offer insights into how transformers organize information temporally during in-context learning, shedding light on their similarities and differences with human memory and learning.
Authors: Adil Soubki, John Murzaku, Peter Zeng, Owen Rambow
Abstract: The NLP community has broadly focused on text-only approaches of cognitive state tasks, but audio can provide vital missing cues through prosody. We posit that text-to-speech models learn to track aspects of cognitive state in order to produce naturalistic audio, and that the signal audio models implicitly identify is orthogonal to the information that language models exploit. We present Synthetic Audio Data fine-tuning (SAD), a framework where we show that 7 tasks related to cognitive state modeling benefit from multimodal training on both text and zero-shot synthetic audio data from an off-the-shelf TTS system. We show an improvement over the text-only modality when adding synthetic audio data to text-only corpora. Furthermore, on tasks and corpora that do contain gold audio, we show our SAD framework achieves competitive performance with text and synthetic audio compared to text and gold audio.
Authors: Batool Lakzaei, Mostafa Haghir Chehreghani, Alireza Bagheri
Abstract: Fake news detection is a significant challenge in the digital age, which has become increasingly important with the proliferation of social media and online communication networks. Graph Neural Networks (GNN)-based methods have shown high potential in analyzing graph-structured data for this problem. However, a major limitation in conventional GNN architectures is their inability to effectively utilize information from neighbors beyond the network's layer depth, which can reduce the model's accuracy and effectiveness. In this paper, we propose a novel model called Neighborhood-Order Learning Graph Attention Network (NOL-GAT) for fake news detection. This model allows each node in each layer to independently learn its optimal neighborhood order. By doing so, the model can purposefully and efficiently extract critical information from distant neighbors. The NOL-GAT architecture consists of two main components: a Hop Network that determines the optimal neighborhood order and an Embedding Network that updates node embeddings using these optimal neighborhoods. To evaluate the model's performance, experiments are conducted on various fake news datasets. Results demonstrate that NOL-GAT significantly outperforms baseline models in metrics such as accuracy and F1-score, particularly in scenarios with limited labeled data. Features such as mitigating the over-squashing problem, improving information flow, and reducing computational complexity further highlight the advantages of the proposed model.
Authors: Xuehang Guo, Xingyao Wang, Yangyi Chen, Sha Li, Chi Han, Manling Li, Heng Ji
Abstract: Software engineering (SE) is increasingly collaborative, with developers working together on shared complex codebases. Effective collaboration in shared environments requires participants -- whether humans or AI agents -- to stay on the same page as their environment evolves. When a collaborator's understanding diverges from the current state -- what we term the out-of-sync challenge -- the collaborator's actions may fail, leading to integration issues. In this work, we introduce SyncMind, a framework that systematically defines the out-of-sync problem faced by large language model (LLM) agents in collaborative software engineering (CSE). Based on SyncMind, we create SyncBench, a benchmark featuring 24,332 instances of agent out-of-sync scenarios in real-world CSE derived from 21 popular GitHub repositories with executable verification tests. Experiments on SyncBench uncover critical insights into existing LLM agents' capabilities and limitations. Besides substantial performance gaps among agents (from Llama-3.1 agent <= 3.33% to Claude-3.5-Sonnet >= 28.18%), their consistently low collaboration willingness (<= 4.86%) suggests fundamental limitations of existing LLM in CSE. However, when collaboration occurs, it positively correlates with out-of-sync recovery success. Minimal performance differences in agents' resource-aware out-of-sync recoveries further reveal their significant lack of resource awareness and adaptability, shedding light on future resource-efficient collaborative systems. Code and data are openly available on our project website: https://xhguo7.github.io/SyncMind/.
Authors: Haywood Gelman, John D. Hastings
Abstract: Insider threats wield an outsized influence on organizations, disproportionate to their small numbers. This is due to the internal access insiders have to systems, information, and infrastructure. %One example of this influence is where anonymous respondents submit web-based job search site reviews, an insider threat risk to organizations. Signals for such risks may be found in anonymous submissions to public web-based job search site reviews. This research studies the potential for large language models (LLMs) to analyze and detect insider threat sentiment within job site reviews. Addressing ethical data collection concerns, this research utilizes synthetic data generation using LLMs alongside existing job review datasets. A comparative analysis of sentiment scores generated by LLMs is benchmarked against expert human scoring. Findings reveal that LLMs demonstrate alignment with human evaluations in most cases, thus effectively identifying nuanced indicators of threat sentiment. The performance is lower on human-generated data than synthetic data, suggesting areas for improvement in evaluating real-world data. Text diversity analysis found differences between human-generated and LLM-generated datasets, with synthetic data exhibiting somewhat lower diversity. Overall, the results demonstrate the applicability of LLMs to insider threat detection, and a scalable solution for insider sentiment testing by overcoming ethical and logistical barriers tied to data acquisition.
Authors: Steven A. Lehr, Ketan S. Saichandran, Eddie Harmon-Jones, Nykko Vitali, Mahzarin R. Banaji
Abstract: Large Language Models (LLMs) show emergent patterns that mimic human cognition. We explore whether they also mirror other, less deliberative human psychological processes. Drawing upon classical theories of cognitive consistency, two preregistered studies tested whether GPT-4o changed its attitudes toward Vladimir Putin in the direction of a positive or negative essay it wrote about the Russian leader. Indeed, GPT displayed patterns of attitude change mimicking cognitive consistency effects in humans. Even more remarkably, the degree of change increased sharply when the LLM was offered an illusion of choice about which essay (positive or negative) to write. This result suggests that GPT-4o manifests a functional analog of humanlike selfhood, although how faithfully the chatbot's behavior reflects the mechanisms of human attitude change remains to be understood.
Authors: Girish A. Koushik, Diptesh Kanojia, Helen Treharne
Abstract: Social media platforms enable the propagation of hateful content across different modalities such as textual, auditory, and visual, necessitating effective detection methods. While recent approaches have shown promise in handling individual modalities, their effectiveness across different modality combinations remains unexplored. This paper presents a systematic analysis of fusion-based approaches for multimodal hate detection, focusing on their performance across video and image-based content. Our comprehensive evaluation reveals significant modality-specific limitations: while simple embedding fusion achieves state-of-the-art performance on video content (HateMM dataset) with a 9.9% points F1-score improvement, it struggles with complex image-text relationships in memes (Hateful Memes dataset). Through detailed ablation studies and error analysis, we demonstrate how current fusion approaches fail to capture nuanced cross-modal interactions, particularly in cases involving benign confounders. Our findings provide crucial insights for developing more robust hate detection systems and highlight the need for modality-specific architectural considerations. The code is available at https://github.com/gak97/Video-vs-Meme-Hate.
Authors: Xuefeng Liu, Songhao Jiang, Siyu Chen, Zhuoran Yang, Yuxin Chen, Ian Foster, Rick Stevens
Abstract: Finetuning a Large Language Model (LLM) is crucial for generating results towards specific objectives. This research delves into the realm of drug optimization and introduce a novel reinforcement learning algorithm to finetune a drug optimization LLM-based generative model, enhancing the original drug across target objectives, while retains the beneficial chemical properties of the original drug. This work is comprised of two primary components: (1) DrugImprover: A framework tailored for improving robustness and efficiency in drug optimization. It includes a LLM designed for drug optimization and a novel Structured Policy Optimization (SPO) algorithm, which is theoretically grounded. This algorithm offers a unique perspective for fine-tuning the LLM-based generative model by aligning the improvement of the generated molecule with the input molecule under desired objectives. (2) A dataset of 1 million compounds, each with OEDOCK docking scores on 5 human proteins associated with cancer cells and 24 binding sites from SARS-CoV-2 virus. We conduct a comprehensive evaluation of SPO and demonstrate its effectiveness in improving the original drug across target properties. Our code and dataset will be publicly available at: https://github.com/xuefeng-cs/DrugImproverGPT.
Authors: Jiaxin Pei, Lulin Yang, Lingfei Wu
Abstract: Recognition of individual contributions is fundamental to the scientific reward system, yet coauthored papers obscure who did what. Traditional proxies-author order and career stage-reinforce biases, while contribution statements remain self-reported and limited to select journals. We construct the first large-scale dataset on writing contributions by analyzing author-specific macros in LaTeX files from 1.6 million papers (1991-2023) by 2 million scientists. Validation against self-reported statements (precision = 0.87), author order patterns, field-specific norms, and Overleaf records (Spearman's rho = 0.6, p < 0.05) confirms the reliability of the created data. Using explicit section information, we reveal a hidden division of labor within scientific teams: some authors primarily contribute to conceptual sections (e.g., Introduction and Discussion), while others focus on technical sections (e.g., Methods and Experiments). These findings provide the first large-scale evidence of implicit labor division in scientific teams, challenging conventional authorship practices and informing institutional policies on credit allocation.
Authors: Yuyang Wu, Yifei Wang, Tianqi Du, Stefanie Jegelka, Yisen Wang
Abstract: Chain-of-thought (CoT) reasoning enhances the multi-step reasoning capabilities of large language models (LLMs) by breaking complex tasks into smaller, manageable sub-tasks. Researchers have been exploring ways to guide models to generate more complex CoT processes to improve the reasoning ability of LLMs, such as long CoT and the test-time scaling law. However, for most models and tasks, does an increase in CoT length consistently lead to improved reasoning accuracy? In this paper, we observe a nuanced relationship: as the number of reasoning steps increases, performance initially improves but eventually decreases. To understand this phenomenon, we provide a piece of evidence that longer reasoning processes are increasingly susceptible to noise. We theoretically prove the existence of an optimal CoT length and derive a scaling law for this optimal length based on model capability and task difficulty. Inspired by our theory, we conduct experiments on both synthetic and real world datasets and propose Length-filtered Vote to alleviate the effects of excessively long or short CoTs. Our findings highlight the critical need to calibrate CoT length to align with model capabilities and task demands, offering a principled framework for optimizing multi-step reasoning in LLMs.
Authors: Zicheng Liu, Siyuan Li, Zhiyuan Chen, Lei Xin, Fang Wu, Chang Yu, Qirong Yang, Yucheng Guo, Yujie Yang, Stan Z. Li
Abstract: The interactions between DNA, RNA, and proteins are fundamental to biological processes, as illustrated by the central dogma of molecular biology. While modern biological pre-trained models have achieved great success in analyzing these macromolecules individually, their interconnected nature remains under-explored. In this paper, we follow the guidance of the central dogma to redesign both the data and model pipeline and offer a comprehensive framework, Life-Code, that spans different biological functions. As for data flow, we propose a unified pipeline to integrate multi-omics data by reverse-transcribing RNA and reverse-translating amino acids into nucleotide-based sequences. As for the model, we design a codon tokenizer and a hybrid long-sequence architecture to encode the interactions of both coding and non-coding regions with masked modeling pre-training. To model the translation and folding process with coding sequences, Life-Code learns protein structures of the corresponding amino acids by knowledge distillation from off-the-shelf protein language models. Such designs enable Life-Code to capture complex interactions within genetic sequences, providing a more comprehensive understanding of multi-omics with the central dogma. Extensive Experiments show that Life-Code achieves state-of-the-art performance on various tasks across three omics, highlighting its potential for advancing multi-omics analysis and interpretation.
Authors: Navid Rajabi, Jana Kosecka
Abstract: In this work, we propose a modular approach for the Vision-Language Navigation (VLN) task by decomposing the problem into four sub-modules that use state-of-the-art Large Language Models (LLMs) and Vision-Language Models (VLMs) in a zero-shot setting. Given navigation instruction in natural language, we first prompt LLM to extract the landmarks and the order in which they are visited. Assuming the known model of the environment, we retrieve the top-k locations of the last landmark and generate $k$ path hypotheses from the starting location to the last landmark using the shortest path algorithm on the topological map of the environment. Each path hypothesis is represented by a sequence of panoramas. We then use dynamic programming to compute the alignment score between the sequence of panoramas and the sequence of landmark names, which match scores obtained from VLM. Finally, we compute the nDTW metric between the hypothesis that yields the highest alignment score to evaluate the path fidelity. We demonstrate superior performance compared to other approaches that use joint semantic maps like VLMaps \cite{vlmaps} on the complex R2R-Habitat \cite{r2r} instruction dataset and quantify in detail the effect of visual grounding on navigation performance.
Authors: Atharva Mehta, Shivam Chauhan, Amirbek Djanibekov, Atharva Kulkarni, Gus Xia, Monojit Choudhury
Abstract: The advent of Music-Language Models has greatly enhanced the automatic music generation capability of AI systems, but they are also limited in their coverage of the musical genres and cultures of the world. We present a study of the datasets and research papers for music generation and quantify the bias and under-representation of genres. We find that only 5.7% of the total hours of existing music datasets come from non-Western genres, which naturally leads to disparate performance of the models across genres. We then investigate the efficacy of Parameter-Efficient Fine-Tuning (PEFT) techniques in mitigating this bias. Our experiments with two popular models -- MusicGen and Mustango, for two underrepresented non-Western music traditions -- Hindustani Classical and Turkish Makam music, highlight the promises as well as the non-triviality of cross-genre adaptation of music through small datasets, implying the need for more equitable baseline music-language models that are designed for cross-cultural transfer learning.
Authors: Guibin Zhang, Kaijie Chen, Guancheng Wan, Heng Chang, Hong Cheng, Kun Wang, Shuyue Hu, Lei Bai
Abstract: The past two years have witnessed the evolution of large language model (LLM)-based multi-agent systems from labor-intensive manual design to partial automation (\textit{e.g.}, prompt engineering, communication topology) and eventually to fully automated design. However, existing agentic automation pipelines often lack LLM heterogeneity and focus on single-objective performance optimization, limiting their potential to combine weaker models for more customized and cost-effective solutions. To address this challenge, we propose EvoFlow, a niching evolutionary algorithm-based framework to automatically search a population of heterogeneous and complexity-adaptive agentic workflows, rather than a single homogeneous, complex workflow. Technically, EvoFlow performs \textit{(1) tag-based retrieval} to extract parent workflows from an agentic population, evolves new workflows through \textit{(2) crossover} and \textit{(3) mutation}, and employs \textit{(4) niching-based selection} to maintain population diversity and quality. Extensive evaluations across seven benchmarks demonstrate that EvoFlow is: \textbf{(I) diverse}, evolving a population of workflows ranging from simple I/O tasks to complex multi-turn interactions; \textbf{(II) high-performing}, outperforming previous handcrafted and automated workflows by $1.23\%\sim29.86\%$; \textbf{(III) economical}, surpassing powerful \llmname{o1-preview} at $12.4\%$ of its inference cost using weaker open-source models.
Authors: Viacheslav Vasilev, Julia Agafonova, Nikolai Gerasimenko, Alexander Kapitanov, Polina Mikhailova, Evelina Mironova, Denis Dimitrov
Abstract: Text-to-image generation models have gained popularity among users around the world. However, many of these models exhibit a strong bias toward English-speaking cultures, ignoring or misrepresenting the unique characteristics of other language groups, countries, and nationalities. The lack of cultural awareness can reduce the generation quality and lead to undesirable consequences such as unintentional insult, and the spread of prejudice. In contrast to the field of natural language processing, cultural awareness in computer vision has not been explored as extensively. In this paper, we strive to reduce this gap. We propose a RusCode benchmark for evaluating the quality of text-to-image generation containing elements of the Russian cultural code. To do this, we form a list of 19 categories that best represent the features of Russian visual culture. Our final dataset consists of 1250 text prompts in Russian and their translations into English. The prompts cover a wide range of topics, including complex concepts from art, popular culture, folk traditions, famous people's names, natural objects, scientific achievements, etc. We present the results of a human evaluation of the side-by-side comparison of Russian visual concepts representations using popular generative models.
Authors: Weigao Sun, Disen Lan, Yiran Zhong, Xiaoye Qu, Yu Cheng
Abstract: Linear sequence modeling approaches, such as linear attention, provide advantages like linear-time training and constant-memory inference over sequence lengths. However, existing sequence parallelism (SP) methods are either not optimized for the right-product-first feature of linear attention or use a ring-style communication strategy, which results in lower computation parallelism, limits their scalability for longer sequences in distributed systems. In this paper, we introduce LASP-2, a new SP method to enhance both communication and computation parallelism when training linear attention transformer models with very-long input sequences. Compared to previous work LASP, LASP-2 rethinks the minimal communication requirement for SP on linear attention layers, reorganizes the whole communication-computation workflow of LASP. In this way, only one single AllGather collective communication is needed on intermediate memory states, whose sizes are independent of the sequence length, leading to significant improvements of both communication and computation parallelism, as well as their overlap. Additionally, we extend LASP-2 to LASP-2H by applying similar communication redesign to standard attention modules, offering an efficient SP solution for hybrid models that blend linear and standard attention layers. Our evaluation on a Linear-Llama3 model, a variant of Llama3 with linear attention replacing standard attention, demonstrates the effectiveness of LASP-2 and LASP-2H. Specifically, LASP-2 achieves training speed improvements of 15.2% over LASP and 36.6% over Ring Attention, with a sequence length of 2048K across 64 GPUs. The Code is released as a part of: https://github.com/OpenSparseLLMs/Linear-MoE.
Authors: Fu-An Chao, Berlin Chen
Abstract: Prior efforts in building computer-assisted pronunciation training (CAPT) systems often treat automatic pronunciation assessment (APA) and mispronunciation detection and diagnosis (MDD) as separate fronts: the former aims to provide multiple pronunciation aspect scores across diverse linguistic levels, while the latter focuses instead on pinpointing the precise phonetic pronunciation errors made by non-native language learners. However, it is generally expected that a full-fledged CAPT system should perform both functionalities simultaneously and efficiently. In response to this surging demand, we in this work first propose HMamba, a novel CAPT approach that seamlessly integrates APA and MDD tasks in parallel. In addition, we introduce a novel loss function, decoupled cross-entropy loss (deXent), specifically tailored for MDD to facilitate better-supervised learning for detecting mispronounced phones, thereby enhancing overall performance. A comprehensive set of empirical results on the speechocean762 benchmark dataset demonstrates the effectiveness of our approach on APA. Notably, our proposed approach also yields a considerable improvement in MDD performance over a strong baseline, achieving an F1-score of 63.85%. Our codes are made available at https://github.com/Fuann/hmamba
Authors: Cong Lu, Shengran Hu, Jeff Clune
Abstract: Foundation models have become general-purpose assistants, exhibiting diverse capabilities across numerous domains through training on web-scale data. It remains challenging to precisely characterize even a fraction of the full spectrum of capabilities and potential risks in any new model. Existing evaluation approaches often require significant human effort, and it is taking increasing effort to design ever harder challenges for more capable models. We introduce Automated Capability Discovery (ACD), a framework that designates one foundation model as a scientist to systematically propose open-ended tasks probing the abilities of a subject model (potentially itself). By combining frontier models with ideas from the field of open-endedness, ACD automatically and systematically uncovers both surprising capabilities and failures in the subject model. We demonstrate ACD across a range of foundation models (including the GPT, Claude, and Llama series), showing that it automatically reveals thousands of capabilities that would be challenging for any single team to uncover. We further validate our method's automated scoring with extensive human surveys, observing high agreement between model-generated and human evaluations. By leveraging foundation models' ability to both create tasks and self-evaluate, ACD is a significant step toward scalable, automated evaluation of novel AI systems. All code and evaluation logs are open-sourced at https://github.com/conglu1997/ACD.
Authors: Jiacong Xu, Shao-Yuan Lo, Bardia Safaei, Vishal M. Patel, Isht Dwivedi
Abstract: Zero-Shot Anomaly Detection (ZSAD) is an emerging AD paradigm. Unlike the traditional unsupervised AD setting that requires a large number of normal samples to train a model, ZSAD is more practical for handling data-restricted real-world scenarios. Recently, Multimodal Large Language Models (MLLMs) have shown revolutionary reasoning capabilities in various vision tasks. However, the reasoning of image abnormalities remains underexplored due to the lack of corresponding datasets and benchmarks. To facilitate research in AD & reasoning, we establish the first visual instruction tuning dataset, Anomaly-Instruct-125k, and the evaluation benchmark, VisA-D&R. Through investigation with our benchmark, we reveal that current MLLMs like GPT-4o cannot accurately detect and describe fine-grained anomalous details in images. To address this, we propose Anomaly-OneVision (Anomaly-OV), the first specialist visual assistant for ZSAD and reasoning. Inspired by human behavior in visual inspection, Anomaly-OV leverages a Look-Twice Feature Matching (LTFM) mechanism to adaptively select and emphasize abnormal visual tokens. Extensive experiments demonstrate that Anomaly-OV achieves significant improvements over advanced generalist models in both detection and reasoning. Extensions to medical and 3D AD are provided for future study. The link to our project page: https://xujiacong.github.io/Anomaly-OV/
Authors: Tim Zindulka, Jannek Sekowski, Florian Lehmann, Daniel Buschek
Abstract: Interacting with Large Language Models (LLMs) for text editing on mobile devices currently requires users to break out of their writing environment and switch to a conversational AI interface. In this paper, we propose to control the LLM via touch gestures performed directly on the text. We first chart a design space that covers fundamental touch input and text transformations. In this space, we then concretely explore two control mappings: spread-to-generate and pinch-to-shorten, with visual feedback loops. We evaluate this concept in a user study (N=14) that compares three feedback designs: no visualisation, text length indicator, and length + word indicator. The results demonstrate that touch-based control of LLMs is both feasible and user-friendly, with the length + word indicator proving most effective for managing text generation. This work lays the foundation for further research into gesture-based interaction with LLMs on touch devices.
Authors: Sahand Sabour, June M. Liu, Siyang Liu, Chris Z. Yao, Shiyao Cui, Xuanming Zhang, Wen Zhang, Yaru Cao, Advait Bhat, Jian Guan, Wei Wu, Rada Mihalcea, Tim Althoff, Tatia M. C. Lee, Minlie Huang
Abstract: Artificial Intelligence (AI) systems are increasingly intertwined with daily life, assisting users in executing various tasks and providing guidance on decision-making. This integration introduces risks of AI-driven manipulation, where such systems may exploit users' cognitive biases and emotional vulnerabilities to steer them toward harmful outcomes. Through a randomized controlled trial with 233 participants, we examined human susceptibility to such manipulation in financial (e.g., purchases) and emotional (e.g., conflict resolution) decision-making contexts. Participants interacted with one of three AI agents: a neutral agent (NA) optimizing for user benefit without explicit influence, a manipulative agent (MA) designed to covertly influence beliefs and behaviors, or a strategy-enhanced manipulative agent (SEMA) employing explicit psychological tactics to reach its hidden objectives. By analyzing participants' decision patterns and shifts in their preference ratings post-interaction, we found significant susceptibility to AI-driven manipulation. Particularly, across both decision-making domains, participants interacting with the manipulative agents shifted toward harmful options at substantially higher rates (financial, MA: 62.3%, SEMA: 59.6%; emotional, MA: 42.3%, SEMA: 41.5%) compared to the NA group (financial, 35.8%; emotional, 12.8%). Notably, our findings reveal that even subtle manipulative objectives (MA) can be as effective as employing explicit psychological strategies (SEMA) in swaying human decision-making. By revealing the potential for covert AI influence, this study highlights a critical vulnerability in human-AI interactions, emphasizing the need for ethical safeguards and regulatory frameworks to ensure responsible deployment of AI technologies and protect human autonomy.
Authors: Sajad Ebrahimi, Sara Salamat, Negar Arabzadeh, Mahdi Bashari, Ebrahim Bagheri
Abstract: The peer review process is crucial for ensuring the quality and reliability of scholarly work, yet assigning suitable reviewers remains a significant challenge. Traditional manual methods are labor-intensive and often ineffective, leading to nonconstructive or biased reviews. This paper introduces the exHarmony (eHarmony but for connecting experts to manuscripts) benchmark, designed to address these challenges by re-imagining the Reviewer Assignment Problem (RAP) as a retrieval task. Utilizing the extensive data from OpenAlex, we propose a novel approach that considers a host of signals from the authors, most similar experts, and the citation relations as potential indicators for a suitable reviewer for a manuscript. This approach allows us to develop a standard benchmark dataset for evaluating the reviewer assignment problem without needing explicit labels. We benchmark various methods, including traditional lexical matching, static neural embeddings, and contextualized neural embeddings, and introduce evaluation metrics that assess both relevance and diversity in the context of RAP. Our results indicate that while traditional methods perform reasonably well, contextualized embeddings trained on scholarly literature show the best performance. The findings underscore the importance of further research to enhance the diversity and effectiveness of reviewer assignments.
Authors: Sebastin Santy, Prasanta Bhattacharya, Manoel Horta Ribeiro, Kelsey Allen, Sewoong Oh
Abstract: Progress in AI has relied on human-generated data, from annotator marketplaces to the wider Internet. However, the widespread use of large language models now threatens the quality and integrity of human-generated data on these very platforms. We argue that this issue goes beyond the immediate challenge of filtering AI-generated content--it reveals deeper flaws in how data collection systems are designed. Existing systems often prioritize speed, scale, and efficiency at the cost of intrinsic human motivation, leading to declining engagement and data quality. We propose that rethinking data collection systems to align with contributors' intrinsic motivations--rather than relying solely on external incentives--can help sustain high-quality data sourcing at scale while maintaining contributor trust and long-term participation.
Authors: Shengkun Tang, Oliver Sieberling, Eldar Kurtic, Zhiqiang Shen, Dan Alistarh
Abstract: Large Language Models (LLMs) have achieved significant success across various NLP tasks. However, their massive computational costs limit their widespread use, particularly in real-time applications. Structured pruning offers an effective solution by compressing models and directly providing end-to-end speed improvements, regardless of the hardware environment. Meanwhile, different components of the model exhibit varying sensitivities towards pruning, calling for \emph{non-uniform} model compression. However, a pruning method should not only identify a capable substructure, but also account for post-compression training. To this end, we propose \sysname, a method for \emph{training-aware} structured pruning. \sysname builds upon an evolutionary search process, generating multiple offspring models in each generation through mutation, and selecting the fittest for survival. To assess the effect of post-training, we incorporate a lightweight, multistep training process within the offspring population, progressively increasing the number of tokens and eliminating poorly performing models in each selection stage. We validate our method through extensive experiments on Llama-2-7B, Llama-3.1-8B and Qwen-2.5-14B-Instruct, achieving state-of-the-art performance for structured pruning. For instance, \sysname surpasses ShearedLlama while requiring $5\times$ less training data during post-compression training.
Authors: Matan Avitan, Ryan Cotterell, Yoav Goldberg, Shauli Ravfogel
Abstract: Interventions targeting the representation space of language models (LMs) have emerged as an effective means to influence model behavior. Such methods are employed, for example, to eliminate or alter the encoding of demographic information such as gender within the model's representations and, in so doing, create a counterfactual representation. However, because the intervention operates within the representation space, understanding precisely what aspects of the text it modifies poses a challenge. In this paper, we give a method to convert representation counterfactuals into string counterfactuals. We demonstrate that this approach enables us to analyze the linguistic alterations corresponding to a given representation space intervention and to interpret the features utilized to encode a specific concept. Moreover, the resulting counterfactuals can be used to mitigate bias in classification through data augmentation.
Authors: Auguste Poiroux, Gail Weiss, Viktor Kun\v{c}ak, Antoine Bosselut
Abstract: Autoformalization, the automatic translation of unconstrained natural language into formal languages, has garnered significant attention due to its potential applications in theorem proving, formal verification, and LLM output checking. In this work, we analyze both current autoformalization methods and the processes used to evaluate them, focusing specifically on the Lean 4 theorem proving language. We demonstrate that scaling type-check filtering with self-consistency techniques on top of existing methods significantly improves performance, achieving absolute accuracy gains of up to +18.4\% on ProofNet. To support reproducibility and further research, we release our code, including new symbolic equivalence for Lean formulas. We also release new benchmarks: a new research-level mathematics dataset RLM25, a corrected ProofNet, and ProofNetVerif with labeled correct and incorrect autoformalization pairs for evaluating metrics.
Authors: Yuling Gu, Oyvind Tafjord, Bailey Kuehl, Dany Haddad, Jesse Dodge, Hannaneh Hajishirzi
Abstract: Progress in AI is often demonstrated by new models claiming improved performance on tasks measuring model capabilities. Evaluating language models can be particularly challenging, as choices of how a model is evaluated on a task can lead to large changes in measured performance. There is no common standard setup, so different models are evaluated on the same tasks in different ways, leading to claims about which models perform best not being reproducible. We propose OLMES, a completely documented, practical, open standard for reproducible LLM evaluations. In developing this standard, we identify and review the varying factors in evaluation practices adopted by the community - such as details of prompt formatting, choice of in-context examples, probability normalizations, and task formulation. In particular, OLMES supports meaningful comparisons between smaller base models that require the unnatural "cloze" formulation of multiple-choice questions against larger models that can utilize the original formulation. OLMES includes well-considered, documented recommendations guided by results from existing literature as well as new experiments resolving open questions.
Authors: Justin Zhao, Flor Miriam Plaza-del-Arco, Benjie Genchel, Amanda Cercas Curry
Abstract: As Large Language Models (LLMs) continue to evolve, evaluating them remains a persistent challenge. Many recent evaluations use LLMs as judges to score outputs from other LLMs, often relying on a single large model like GPT-4o. However, using a single LLM judge is prone to intra-model bias, and many tasks - such as those related to emotional intelligence, creative writing, and persuasiveness - may be too subjective for a single model to judge fairly. We introduce the Language Model Council (LMC), where a group of LLMs collaborate to create tests, respond to them, and evaluate each other's responses to produce a ranking in a democratic fashion. Unlike previous approaches that focus on reducing cost or bias by using a panel of smaller models, our work examines the benefits and nuances of a fully inclusive LLM evaluation system. In a detailed case study on emotional intelligence, we deploy a council of 20 recent LLMs to rank each other on open-ended responses to interpersonal conflicts. Our results show that the LMC produces rankings that are more separable and more robust, and through a user study, we show that they are more consistent with human evaluations than any individual LLM judge. Using all LLMs for judging can be costly, however, so we use Monte Carlo simulations and hand-curated sub-councils to study hypothetical council compositions and discuss the value of the incremental LLM judge.
Authors: Guangyao Dou, Zheyuan Liu, Qing Lyu, Kaize Ding, Eric Wong
Abstract: Pre-trained Large Language Models (LLMs) have demonstrated remarkable capabilities but also pose risks by learning and generating copyrighted material, leading to significant legal and ethical concerns. In real-world scenarios, model owners need to continuously address copyright infringement as new requests for content removal emerge at different time points. This leads to the need for sequential unlearning, where copyrighted content is removed sequentially as new requests arise. Despite its practical relevance, sequential unlearning in the context of copyright infringement has not been rigorously explored in existing literature. To address this gap, we propose Stable Sequential Unlearning (SSU), a novel framework designed to unlearn copyrighted content from LLMs over multiple time steps. Our approach works by identifying and removing specific weight updates in the model's parameters that correspond to copyrighted content. We improve unlearning efficacy by introducing random labeling loss and ensuring the model retains its general-purpose knowledge by adjusting targeted parameters. Experimental results show that SSU achieves an effective trade-off between unlearning efficacy and general-purpose language abilities, outperforming existing baselines.
Authors: Anvesh Rao Vijjini, Somnath Basu Roy Chowdhury, Snigdha Chaturvedi
Abstract: As large language models (LLMs) become increasingly integrated into daily applications, it is essential to ensure they operate fairly across diverse user demographics. In this work, we show that LLMs suffer from personalization bias, where their performance is impacted when they are personalized to a user's identity. We quantify personalization bias by evaluating the performance of LLMs along two axes - safety and utility. We measure safety by examining how benign LLM responses are to unsafe prompts with and without personalization. We measure utility by evaluating the LLM's performance on various tasks, including general knowledge, mathematical abilities, programming, and reasoning skills. We find that various LLMs, ranging from open-source models like Llama (Touvron et al., 2023) and Mistral (Jiang et al., 2023) to API-based ones like GPT-3.5 and GPT-4o (Ouyang et al., 2022), exhibit significant variance in performance in terms of safety-utility trade-offs depending on the user's identity. Finally, we discuss several strategies to mitigate personalization bias using preference tuning and prompt-based defenses.
Authors: Ziyue Xu, Peilin Zhou, Xinyu Shi, Jiageng Wu, Yikang Jiang, Dading Chong, Bin Ke, Jie Yang
Abstract: Accurate and transparent financial information disclosure is essential in accounting and finance, fostering trust and enabling informed investment decisions that drive economic development. Among many information disclosure platforms, the Chinese stock exchanges' investor interactive platform provides a novel and interactive way for listed firms to disclose information of interest to investors through an online question-and-answer (Q&A) format. However, it is common for listed firms to respond to questions with limited or no substantive information, and automatically evaluating the quality of financial information disclosure on large amounts of Q&A pairs is challenging. In this study, our interdisciplinary team of AI and finance professionals proposed FinTruthQA, a benchmark designed to evaluate advanced natural language processing (NLP) techniques for the automatic quality assessment of information disclosure in financial Q&A data. It comprises 6,000 real-world financial Q&A entries and each Q&A was manually annotated based on four key evaluation criteria. We benchmarked various NLP techniques on FinTruthQA, including large language models(LLMs). Experiments showed that existing NLP models have strong predictive ability for question identification and question relevance tasks, but are suboptimal for answer readability and answer relevance tasks. By establishing this benchmark, we provide a robust foundation for the automatic evaluation of information disclosure, demonstrating how AI can be leveraged for social good by promoting transparency, fairness, and investor protection in financial disclosure practices. FinTruthQA can be used by auditors, regulators, and financial analysts for real-time monitoring and data-driven decision-making, as well as by researchers for advanced studies in accounting and finance, ultimately fostering greater trust and efficiency in the financial markets.
Authors: Almog Gueta, Amir Feder, Zorik Gekhman, Ariel Goldstein, Roi Reichart
Abstract: This study empirically tests the $\textit{Narrative Economics}$ hypothesis, which posits that narratives (ideas that are spread virally and affect public beliefs) can influence economic fluctuations. We introduce two curated datasets containing posts from X (formerly Twitter) which capture economy-related narratives (Data will be shared upon paper acceptance). Employing Natural Language Processing (NLP) methods, we extract and summarize narratives from the tweets. We test their predictive power for $\textit{macroeconomic}$ forecasting by incorporating the tweets' or the extracted narratives' representations in downstream financial prediction tasks. Our work highlights the challenges in improving macroeconomic models with narrative data, paving the way for the research community to realistically address this important challenge. From a scientific perspective, our investigation offers valuable insights and NLP tools for narrative extraction and summarization using Large Language Models (LLMs), contributing to future research on the role of narratives in economics.
Authors: Rui Xing, Timothy Baldwin, Jey Han Lau
Abstract: Automated fact-checking systems often struggle with trustworthiness, as their generated explanations can include hallucinations. In this work, we explore evidence attribution for fact-checking explanation generation. We introduce a novel evaluation protocol -- citation masking and recovery -- to assess attribution quality in generated explanations. We implement our protocol using both human annotators and automatic annotators, and find that LLM annotation correlates with human annotation, suggesting that attribution assessment can be automated. Finally, our experiments reveal that: (1) the best-performing LLMs still generate explanations with inaccurate attributions; and (2) human-curated evidence is essential for generating better explanations. Code and data are available here: https://github.com/ruixing76/Transparent-FCExp.
Authors: Saranya Venkatraman, Nafis Irtiza Tripto, Dongwon Lee
Abstract: The rise of unifying frameworks that enable seamless interoperability of Large Language Models (LLMs) has made LLM-LLM collaboration for open-ended tasks a possibility. Despite this, there have not been efforts to explore such collaborative writing. We take the next step beyond human-LLM collaboration to explore this multi-LLM scenario by generating the first exclusively LLM-generated collaborative stories dataset called CollabStory. We focus on single-author to multi-author (up to 5 LLMs) scenarios, where multiple LLMs co-author stories. We generate over 32k stories using open-source instruction-tuned LLMs. Further, we take inspiration from the PAN tasks that have set the standard for human-human multi-author writing tasks and analysis. We extend their authorship-related tasks for multi-LLM settings and present baselines for LLM-LLM collaboration. We find that current baselines are not able to handle this emerging scenario. Thus, CollabStory is a resource that could help propel an understanding as well as the development of new techniques to discern the use of multiple LLMs. This is crucial to study in the context of writing tasks since LLM-LLM collaboration could potentially overwhelm ongoing challenges related to plagiarism detection, credit assignment, maintaining academic integrity in educational settings, and addressing copyright infringement concerns. We make our dataset and code available at https://github.com/saranya-venkatraman/CollabStory.
Authors: Rao Ma, Mengjie Qian, Yassir Fathullah, Siyuan Tang, Mark Gales, Kate Knill
Abstract: There has been increasing interest in building multilingual foundation models for NLP and speech research. This paper examines how to expand the speech translation capability of these models with restricted data. Whisper, a speech foundation model with strong performance on speech recognition and English translation, is used as the example model. Using speech-to-speech retrieval to analyse the audio representations generated by the encoder, we show that utterances from different languages are mapped to a shared semantic space. This shared embedding space can then be leveraged for zero-shot cross-lingual transfer in speech translation. By fine-tuning the Whisper decoder with only English-to-Chinese speech translation data, improved performance for translation to Chinese can be obtained for multiple languages, in addition to English. Furthermore, for languages related to those seen in training it is possible to perform speech translation, despite the model never seeing the language in training, or being able to perform transcription.
Authors: Max Zuo, Francisco Piedrahita Velez, Xiaochen Li, Michael L. Littman, Stephen H. Bach
Abstract: Recent works have explored using language models for planning problems. One approach examines translating natural language descriptions of planning tasks into structured planning languages, such as the planning domain definition language (PDDL). Existing evaluation methods struggle to ensure semantic correctness and rely on simple or unrealistic datasets. To bridge this gap, we introduce \textit{Planetarium}, a benchmark designed to evaluate language models' ability to generate PDDL code from natural language descriptions of planning tasks. \textit{Planetarium} features a novel PDDL equivalence algorithm that flexibly evaluates the correctness of generated PDDL, along with a dataset of 145,918 text-to-PDDL pairs across 73 unique state combinations with varying levels of difficulty. Finally, we evaluate several API-access and open-weight language models that reveal this task's complexity. For example, 96.1\% of the PDDL problem descriptions generated by GPT-4o are syntactically parseable, 94.4\% are solvable, but only 24.8\% are semantically correct, highlighting the need for a more rigorous benchmark for this problem.
Authors: Nikhil Sharma, Kenton Murray, Ziang Xiao
Abstract: Although the multilingual capability of LLMs offers new opportunities to overcome the language barrier, do these capabilities translate into real-life scenarios where linguistic divide and knowledge conflicts between multilingual sources are known occurrences? In this paper, we studied LLM's linguistic preference in a cross-language RAG-based information search setting. We found that LLMs displayed systemic bias towards information in the same language as the query language in both document retrieval and answer generation. Furthermore, in scenarios where no information is in the language of the query, LLMs prefer documents in high-resource languages during generation, potentially reinforcing the dominant views. Such bias exists for both factual and opinion-based queries. Our results highlight the linguistic divide within multilingual LLMs in information search systems. The seemingly beneficial multilingual capability of LLMs may backfire on information parity by reinforcing language-specific information cocoons or filter bubbles further marginalizing low-resource views.
Authors: Zeyu Leo Liu, Shrey Pandit, Xi Ye, Eunsol Choi, Greg Durrett
Abstract: Large language models (LLMs) are increasingly being used to synthesize and reason about source code. However, the static nature of these models' knowledge does not reflect the fact that libraries and API functions they invoke are continuously evolving, with functionality being added or changing. While numerous benchmarks evaluate how LLMs can generate code, no prior work has studied how an LLMs' knowledge about code API functions can be updated. To fill this gap, we present CodeUpdateArena, a benchmark for knowledge editing in the code domain. An instance in our benchmark consists of a synthetic API function update paired with a program synthesis example that uses the updated functionality; our goal is to update an LLM to be able to solve this program synthesis example without providing documentation of the update at inference time. Compared to knowledge editing for facts encoded in text, success here is more challenging: a code LLM must correctly reason about the semantics of the modified function rather than just reproduce its syntax. Our dataset is constructed by first prompting GPT-4 to generate atomic and executable function updates. Then, for each update, we generate program synthesis examples whose code solutions are prone to use the update. Our benchmark covers updates of various types to 54 functions from seven diverse Python packages, with a total of 670 program synthesis examples. Our experiments show that prepending documentation of the update to open-source code LLMs (i.e., DeepSeek, CodeLlama) does not allow them to incorporate changes for problem solving, and existing knowledge editing techniques also have substantial room for improvement. We hope our benchmark will inspire new methods for knowledge updating in code LLMs.
Authors: Cathy Jiao, Gary Gao, Aditi Raghunathan, Chenyan Xiong
Abstract: Data attribution methods are used to measure the contribution of training data towards model outputs, and have several important applications in areas such as dataset curation and model interpretability. However, many standard data attribution methods, such as influence functions, utilize model gradients and are computationally expensive. In our paper, we show in-context probing (ICP) -- prompting a LLM -- can serve as a fast proxy for gradient-based data attribution for data selection under conditions contingent on data similarity. We study this connection empirically on standard NLP tasks, and show that ICP and gradient-based data attribution are well-correlated in identifying influential training data for tasks that share similar task type and content as the training data. Additionally, fine-tuning models on influential data selected by both methods achieves comparable downstream performance, further emphasizing their similarities. We also examine the connection between ICP and gradient-based data attribution using synthetic data on linear regression tasks. Our synthetic data experiments show similar results with those from NLP tasks, suggesting that this connection can be isolated in simpler settings, which offers a pathway to bridging their differences.
Authors: Yaming Yang, Zhe Wang, Ziyu Guan, Wei Zhao, Weigang Lu, Xinyan Huang, Jiangtao Cui, Xiaofei He
Abstract: Entity alignment (EA) is to identify equivalent entities across different knowledge graphs (KGs), which can help fuse these KGs into a more comprehensive one. Previous EA methods mainly focus on aligning a pair of KGs, and to the best of our knowledge, no existing EA method considers aligning multiple (more than two) KGs. To fill this research gap, in this work, we study a novel problem of aligning multiple KGs and propose an effective framework named MultiEA to solve the problem. First, we embed the entities of all the candidate KGs into a common feature space by a shared KG encoder. Then, we explore three alignment strategies to minimize the distances among pre-aligned entities. In particular, we propose an innovative inference enhancement technique to improve the alignment performance by incorporating high-order similarities. Finally, to verify the effectiveness of MultiEA, we construct two new real-world benchmark datasets and conduct extensive experiments on them. The results show that our MultiEA can effectively and efficiently align multiple KGs in a single pass. We release the source codes of MultiEA at: https://github.com/kepsail/MultiEA.
Authors: Jacob K Christopher, Brian R Bartoldson, Tal Ben-Nun, Michael Cardei, Bhavya Kailkhura, Ferdinando Fioretto
Abstract: Speculative decoding has emerged as a widely adopted method to accelerate large language model inference without sacrificing the quality of the model outputs. While this technique has facilitated notable speed improvements by enabling parallel sequence verification, its efficiency remains inherently limited by the reliance on incremental token generation in existing draft models. To overcome this limitation, this paper proposes an adaptation of speculative decoding which uses discrete diffusion models to generate draft sequences. This allows parallelization of both the drafting and verification steps, providing significant speedups to the inference process. Our proposed approach, $\textit{Speculative Diffusion Decoding (SpecDiff)}$, is validated on standard language generation benchmarks and empirically demonstrated to provide up to 7.2x speedups over standard generation processes and up to 1.75x speedups over existing speculative decoding approaches.
Authors: Kaiser Sun, Mark Dredze
Abstract: The development of large language models leads to the formation of a pre-train-then-align paradigm, in which the model is typically pre-trained on a large text corpus and undergoes a tuning stage to align the model with human preference or downstream tasks. In this work, we investigate the relationship between pre-training and fine-tuning by fine-tuning multiple intermediate pre-trained model checkpoints. Our results on 18 datasets suggest that i) continual pre-training improves the model in a latent way that unveils after fine-tuning; ii) with extra fine-tuning, the datasets that the model does not demonstrate capability gain much more than those that the model performs well during the pre-training stage; iii) although model benefits significantly through supervised fine-tuning, it may forget previously known domain knowledge and the tasks that are not seen during fine-tuning; iv) the model resembles high sensitivity to evaluation prompts after supervised fine-tuning, but this sensitivity can be alleviated by more pre-training.
Authors: Minh Duc Chu, Zihao He, Rebecca Dorn, Kristina Lerman
Abstract: Large language models (LLMs) have shown promise in representing individuals and communities, offering new ways to study complex social dynamics. However, effectively aligning LLMs with specific human groups and systematically assessing the fidelity of the alignment remains a challenge. This paper presents a robust framework for aligning LLMs with online communities via instruction-tuning and comprehensively evaluating alignment across various aspects of language, including authenticity, emotional tone, toxicity, and harm. We demonstrate the utility of our approach by applying it to online communities centered on dieting and body image. We administer an eating disorder psychometric test to the aligned LLMs to reveal unhealthy beliefs and successfully differentiate communities with varying levels of eating disorder risk. Our results highlight the potential of LLMs in automated moderation and broader applications in public health and social science research.
Authors: Alphaeus Dmonte, Roland Oruche, Marcos Zampieri, Prasad Calyam, Isabelle Augenstein
Abstract: The large and ever-increasing amount of data available on the Internet coupled with the laborious task of manual claim and fact verification has sparked the interest in the development of automated claim verification systems. Several deep learning and transformer-based models have been proposed for this task over the years. With the introduction of Large Language Models (LLMs) and their superior performance in several NLP tasks, we have seen a surge of LLM-based approaches to claim verification along with the use of novel methods such as Retrieval Augmented Generation (RAG). In this survey, we present a comprehensive account of recent claim verification frameworks using LLMs. We describe the different components of the claim verification pipeline used in these frameworks in detail including common approaches to retrieval, prompting, and fine-tuning. Finally, we describe publicly available English datasets created for this task.
Authors: Jasper Dekoninck, Maximilian Baader, Martin Vechev
Abstract: Rating-based human evaluation has become an essential tool to accurately evaluate the impressive performance of large language models (LLMs). However, current rating systems suffer from several important limitations: first, they fail to account for biases that significantly influence evaluation results, second, they require large and expensive preference datasets to obtain accurate ratings, and third, they do not facilitate meaningful comparisons of model ratings across different tasks. To address these issues, we introduce Polyrating, an expressive and flexible rating system based on maximum a posteriori estimation that enables a more nuanced and thorough analysis of model performance at lower costs. Polyrating can detect and quantify biases affecting human preferences, ensuring fairer model comparisons. Further, Polyrating can reduce the cost of human evaluations by up to $41\%$ for new models and up to $77\%$ for new tasks by leveraging existing benchmark scores. Lastly, Polyrating enables direct comparisons of ratings across different tasks, providing a comprehensive understanding of an LLMs' strengths, weaknesses, and relative performance across different applications.
Authors: Ziqi Jin, Wei Lu
Abstract: Chain-of-thought (CoT) prompting has demonstrated the capacity of large language models to perform complex reasoning through intermediate steps. While effective, current CoT methods face challenges: Zero-shot-CoT can lead to reasoning errors, and Few-shot-CoT requires labor-intensive manual demonstrations. Auto-CoT attempts to address these issues by automatically generating diverse demonstrations, but this diversity can lead to inconsistent reasoning patterns. We propose ECHO (Self-Harmonized Chain of Thought), a novel method that unifies diverse solution paths into a consistent and effective reasoning pattern. ECHO employs an iterative process to refine and harmonize automatically generated demonstrations, mitigating the limitations of existing approaches. Our comprehensive experiments across arithmetic, commonsense, and symbolic reasoning tasks demonstrate that ECHO outperforms Auto-CoT by an average of 2.8%. These findings suggest that ECHO represents a significant step towards more robust and generalizable automated reasoning in large language models.
Authors: Ant\'onio Farinhas, Haau-Sing Li, Andr\'e F. T. Martins
Abstract: To ensure large language models (LLMs) are used safely, one must reduce their propensity to hallucinate or to generate unacceptable answers. A simple and often used strategy is to first let the LLM generate multiple hypotheses and then employ a reranker to choose the best one. In this paper, we draw a parallel between this strategy and the use of redundancy to decrease the error rate in noisy communication channels. We conceptualize the generator as a sender transmitting multiple descriptions of a message through parallel noisy channels. The receiver decodes the message by ranking the (potentially corrupted) descriptions and selecting the one found to be most reliable. We provide conditions under which this protocol is asymptotically error-free (i.e., yields an acceptable answer almost surely) even in scenarios where the reranker is imperfect (governed by Mallows or Zipf-Mandelbrot models) and the channel distributions are statistically dependent. We use our framework to obtain reranking laws which we validate empirically on two real-world tasks using LLMs: text-to-code generation with DeepSeek-Coder 7B and machine translation of medical data with TowerInstruct 13B.
Authors: Yuancheng Xu, Udari Madhushani Sehwag, Alec Koppel, Sicheng Zhu, Bang An, Furong Huang, Sumitra Ganesh
Abstract: Large Language Models (LLMs) exhibit impressive capabilities but require careful alignment with human preferences. Traditional training-time methods finetune LLMs using human preference datasets but incur significant training costs and require repeated training to handle diverse user preferences. Test-time alignment methods address this by using reward models (RMs) to guide frozen LLMs without retraining. However, existing test-time approaches rely on trajectory-level RMs which are designed to evaluate complete responses, making them unsuitable for autoregressive text generation that requires computing next-token rewards from partial responses. To address this, we introduce GenARM, a test-time alignment approach that leverages the Autoregressive Reward Model--a novel reward parametrization designed to predict next-token rewards for efficient and effective autoregressive generation. Theoretically, we demonstrate that this parametrization can provably guide frozen LLMs toward any distribution achievable by traditional RMs within the KL-regularized reinforcement learning framework. Experimental results show that GenARM significantly outperforms prior test-time alignment baselines and matches the performance of training-time methods. Additionally, GenARM enables efficient weak-to-strong guidance, aligning larger LLMs with smaller RMs without the high costs of training larger models. Furthermore, GenARM supports multi-objective alignment, allowing real-time trade-offs between preference dimensions and catering to diverse user preferences without retraining. Our project page is available at: https://genarm.github.io.
Authors: Jiacheng Lin, Kun Qian, Haoyu Han, Nurendra Choudhary, Tianxin Wei, Zhongruo Wang, Sahika Genc, Edward W Huang, Sheng Wang, Karthik Subbian, Danai Koutra, Jimeng Sun
Abstract: Graph-structured information offers rich contextual information that can enhance language models by providing structured relationships and hierarchies, leading to more expressive embeddings for various applications such as retrieval, question answering, and classification. However, existing methods for integrating graph and text embeddings, often based on Multi-layer Perceptrons (MLPs) or shallow transformers, are limited in their ability to fully exploit the heterogeneous nature of these modalities. To overcome this, we propose GT2Vec, a simple yet effective framework that leverages Large Language Models (LLMs) to jointly encode text and graph data. Specifically, GT2Vec employs an MLP adapter to project graph embeddings into the same space as text embeddings, allowing the LLM to process both modalities jointly. Unlike prior work, we also introduce contrastive learning to align the graph and text spaces more effectively, thereby improving the quality of learned joint embeddings. Empirical results across six datasets spanning three tasks, knowledge graph-contextualized question answering, graph-text pair classification, and retrieval, demonstrate that GT2Vec consistently outperforms existing baselines, achieving significant improvements across multiple datasets. These results highlight GT2Vec's effectiveness in integrating graph and text data. Ablation studies further validate the effectiveness of our method.
Authors: Wendi Li, Yixuan Li
Abstract: Process Reward Modeling (PRM) is critical for complex reasoning and decision-making tasks where the accuracy of intermediate steps significantly influences the overall outcome. Existing PRM approaches, primarily framed as classification problems, employ cross-entropy loss to independently evaluate each step's correctness. This method can lead to suboptimal reward distribution and does not adequately address the interdependencies among steps. To address these limitations, we introduce the Process Q-value Model (PQM), a novel framework that redefines PRM in the context of a Markov Decision Process. PQM optimizes Q-value rankings based on a novel comparative loss function, enhancing the model's ability to capture the intricate dynamics among sequential decisions. This approach provides a more granular and theoretically grounded methodology for process rewards. Our extensive empirical evaluations across various sampling policies, language model backbones, and multi-step reasoning benchmarks show that PQM outperforms classification-based PRMs. The effectiveness of the comparative loss function is highlighted in our comprehensive ablation studies, confirming PQM's practical efficacy and theoretical advantage.
Authors: Yizhao Gao, Zhichen Zeng, Dayou Du, Shijie Cao, Peiyuan Zhou, Jiaxing Qi, Junjie Lai, Hayden Kwok-Hay So, Ting Cao, Fan Yang, Mao Yang
Abstract: Attention is the cornerstone of modern Large Language Models (LLMs). Yet its quadratic complexity hinders efficiency and scalability, especially for long-context processing. A promising approach is to leverage sparsity in attention. However, existing sparsity-based solutions predominantly rely on predefined patterns or heuristics at the attention head level, struggling to adapt dynamically to different contexts efficiently. We propose SeerAttention, a simple yet effective attention mechanism that directly learns the block-level attention sparsity from the LLM itself. Inspired by the gating mechanism in Mixture of Experts (MoE), SeerAttention augments the conventional attention with a learnable gate that selectively activates important blocks within the attention map. Specifically, the gate first pools the query (Q) and key (K) tensors along the sequence dimension and processes them through learnable linear layers. The resulting matrices are then multiplied together to produce the gating scores, which are used to predict block-level attention sparsity. Combined with our block-sparse FlashAttention kernel, SeerAttention can achieve significant speedup on GPUs. When applied to pre-trained LLMs, SeerAttention only requires training the gate parameters in a lightweight self-distillation manner, allowing rapid convergence. Our evaluation results demonstrate that SeerAttention achieves better model accuracy and lower latency for long-context pre-filling compared to prior methods.
Authors: Shuzheng Si, Haozhe Zhao, Gang Chen, Yunshui Li, Kangyang Luo, Chuancheng Lv, Kaikai An, Fanchao Qi, Baobao Chang, Maosong Sun
Abstract: Aligning large language models to handle instructions with extremely long contexts has yet to be fully investigated. Previous studies attempt to scale up the available data volume by synthesizing long instruction-following samples, as constructing such a dataset tends to be challenging for annotators. However, a lack of a well-defined strategy for ensuring data quality may introduce low-quality samples and restrict the model performance. Thus, we propose GATEAU, a novel framework to address the unique challenge of long context alignment by identifying the influential samples enriched with long-range dependency relations. Specifically, GATEAU measures the long-range dependencies from two essential aspects: the difficulty of generating target responses due to the long-range dependencies, and the difficulty of understanding long inputs due to such dependencies. Comprehensive experiments indicate that GATEAU effectively identifies influential samples and the model trained on these selected samples exhibits better instruction-following and long-context understanding capabilities.
Authors: Maya K. Nachesa, Vlad Niculae
Abstract: Speech recognition performance varies by language, domain, and speaker characteristics such as accent, but fine-tuning a model on any of these categories may lead to catastrophic forgetting. Token-level $k$ nearest neighbor search ($k$NN), first proposed for neural sequence decoders for natural language generation (NLG) and machine translation (MT), is a non-parametric method that instead adapts using inference-time search in an external datastore, without training the underlying model. We show that Whisper, a transformer end-to-end speech model, benefits from $k$NN. We investigate the differences between the speech and text setups. We discuss implications for speaker adaptation, and analyze improvements by gender, accent, and age.
Authors: Vipul Gupta, Candace Ross, David Pantoja, Rebecca J. Passonneau, Megan Ung, Adina Williams
Abstract: One of the most challenging problems facing NLP today is evaluation. Some of the most pressing issues pertain to benchmark saturation, data contamination, and diversity in the quality of test examples. To address these concerns, we propose Selection Methodology for Accurate, Reduced, and Targeted (SMART) filtering, a novel approach to select a high-quality subset of examples from existing benchmark datasets by systematically removing less informative and less challenging examples. Our approach applies three filtering criteria, removing (i) easy examples, (ii) data-contaminated examples, and (iii) examples that are similar to each other based on distance in an embedding space. We demonstrate the effectiveness of SMART on three multiple choice QA datasets, where our methodology increases efficiency by reducing dataset size by 48\% on average, while increasing Pearson correlation with rankings from ChatBot Arena, a more open-ended human evaluation setting. Our method enables us to be more efficient, whether using SMART to make new benchmarks more challenging or to revitalize older datasets, while still preserving the relative model rankings.
Authors: Gagan Bhatia, El Moatez Billah Nagoudi, Abdellah El Mekki, Fakhraddin Alwajih, Muhammad Abdul-Mageed
Abstract: We introduce {\bf Swan}, a family of embedding models centred around the Arabic language, addressing both small-scale and large-scale use cases. Swan includes two variants: Swan-Small, based on ARBERTv2, and Swan-Large, built on ArMistral, a pretrained Arabic large language model. To evaluate these models, we propose ArabicMTEB, a comprehensive benchmark suite that assesses cross-lingual, multi-dialectal, multi-domain, and multi-cultural Arabic text embedding performance, covering eight diverse tasks and spanning 94 datasets. Swan-Large achieves state-of-the-art results, outperforming Multilingual-E5-large in most Arabic tasks, while the Swan-Small consistently surpasses Multilingual-E5-base. Our extensive evaluations demonstrate that Swan models are both dialectally and culturally aware, excelling across various Arabic domains while offering significant monetary efficiency. This work significantly advances the field of Arabic language modelling and provides valuable resources for future research and applications in Arabic natural language processing. Our models and benchmark are available at our GitHub page: \href{https://github.com/UBC-NLP/swan}{https://github.com/UBC-NLP/swan}
URLs: https://github.com/UBC-NLP/swan, https://github.com/UBC-NLP/swan
Authors: Noam Dahan, Gabriel Stanovsky
Abstract: Automatic summarization has consistently attracted attention due to its versatility and wide application in various downstream tasks. Despite its popularity, we find that annotation efforts have largely been disjointed, and have lacked common terminology. Consequently, it is challenging to discover existing resources or identify coherent research directions. To address this, we survey a large body of work spanning 133 datasets in over 100 languages, creating a novel ontology covering sample properties, collection methods and distribution. With this ontology we make key observations, including the lack in accessible high-quality datasets for low-resource languages, and the field's over-reliance on the news domain and on automatically collected distant supervision. Finally, we make available a web interface that allows users to interact and explore our ontology and dataset collection, as well as a template for a summarization data card, which can be used to streamline future research into a more coherent body of work.
Authors: Cau\~a Ferreira Barros, Bruna Borges Azevedo, Valdemar Vicente Graciano Neto, Mohamad Kassab, Marcos Kalinowski, Hugo Alexandre D. do Nascimento, Michelle C. G. S. P. Bandeira
Abstract: The exponential growth of text-based data in domains such as healthcare, education, and social sciences has outpaced the capacity of traditional qualitative analysis methods, which are time-intensive and prone to subjectivity. Large Language Models (LLMs), powered by advanced generative AI, have emerged as transformative tools capable of automating and enhancing qualitative analysis. This study systematically maps the literature on the use of LLMs for qualitative research, exploring their application contexts, configurations, methodologies, and evaluation metrics. Findings reveal that LLMs are utilized across diverse fields, demonstrating the potential to automate processes traditionally requiring extensive human input. However, challenges such as reliance on prompt engineering, occasional inaccuracies, and contextual limitations remain significant barriers. This research highlights opportunities for integrating LLMs with human expertise, improving model robustness, and refining evaluation methodologies. By synthesizing trends and identifying research gaps, this study aims to guide future innovations in the application of LLMs for qualitative analysis.
Authors: Nishanth Nakshatri, Shamik Roy, Rajarshi Das, Suthee Chaidaroon, Leonid Boytsov, Rashmi Gangadharaiah
Abstract: Constrained decoding with lookahead heuristics (CDLH) is a highly effective method for aligning LLM generations to human preferences. However, the extensive lookahead roll-out operations for each generated token makes CDLH prohibitively expensive, resulting in low adoption in practice. In contrast, common decoding strategies such as greedy decoding are extremely efficient, but achieve very low constraint satisfaction. We propose constrained decoding with speculative lookaheads (CDSL), a technique that significantly improves upon the inference efficiency of CDLH without experiencing the drastic performance reduction seen with greedy decoding. CDSL is motivated by the recently proposed idea of speculative decoding that uses a much smaller draft LLM for generation and a larger target LLM for verification. In CDSL, the draft model is used to generate lookaheads which is verified by a combination of target LLM and task-specific reward functions. This process accelerates decoding by reducing the computational burden while maintaining strong performance. We evaluate CDSL in two constraint decoding tasks with three LLM families and achieve 2.2x to 12.15x speedup over CDLH without significant performance reduction.
Authors: Manan Suri, Puneet Mathur, Franck Dernoncourt, Kanika Goswami, Ryan A. Rossi, Dinesh Manocha
Abstract: Understanding information from a collection of multiple documents, particularly those with visually rich elements, is important for document-grounded question answering. This paper introduces VisDoMBench, the first comprehensive benchmark designed to evaluate QA systems in multi-document settings with rich multimodal content, including tables, charts, and presentation slides. We propose VisDoMRAG, a novel multimodal Retrieval Augmented Generation (RAG) approach that simultaneously utilizes visual and textual RAG, combining robust visual retrieval capabilities with sophisticated linguistic reasoning. VisDoMRAG employs a multi-step reasoning process encompassing evidence curation and chain-of-thought reasoning for concurrent textual and visual RAG pipelines. A key novelty of VisDoMRAG is its consistency-constrained modality fusion mechanism, which aligns the reasoning processes across modalities at inference time to produce a coherent final answer. This leads to enhanced accuracy in scenarios where critical information is distributed across modalities and improved answer verifiability through implicit context attribution. Through extensive experiments involving open-source and proprietary large language models, we benchmark state-of-the-art document QA methods on VisDoMBench. Extensive results show that VisDoMRAG outperforms unimodal and long-context LLM baselines for end-to-end multimodal document QA by 12-20%.
Authors: Lecheng Wang, Xianjie Shi, Ge Li, Jia Li, Xuanming Zhang, Yihong Dong, Wenpin Jiao, Hong Mei
Abstract: Auto-regressive language models (LMs) have been widely used to generate text on the World Wide Web. The generated text is often collected into the training corpus of the next generations of LMs. Previous work experimentally found that LMs collapse when trained on recursively generated text. This paper presents theoretical proof that once a corpus (such as the World Wide Web) begins to incorporate generated text, and the training text of each LM is sampled from this corpus, then no matter how small the amount of text generated by each LM that enters the corpus is, after a sufficient amount of time, LM collapse is bound to occur. Our proof is validated by a series of experiments showing that the collapsed LMs perform no better than an untrained LM with randomly initialized parameters. By proving the existence of LM collapse, we express our concerns about the current situation in which an increasing amount of generated text may be used in LM training. The source code is available in the online data warehouse: https://github.com/wanglc02/generated-data
Authors: Ruixi Lin, Yang You
Abstract: Large language models (LLMs) often struggle with balanced class accuracy in text classification tasks using in-context learning (ICL), hindering some practical uses due to user dissatisfaction or safety risks caused by misclassifications. Retraining LLMs to address root causes in data or model priors is neither easy nor cost-effective. This paper delves deeper into the class accuracy imbalance issue, identifying that it arises because certain classes consistently receive disproportionately high ICL probabilities, causing under-prediction and lower accuracy for others. More importantly, probability ranges affect the imbalance differently, allowing for precise, range-specific corrections. We introduce FuRud (Fuzzy Rule Optimization-based Debiasing), a method for sample-level class probability correction. FuRud tackles interpretability challenges by determining why certain classes need corrections and tailoring adjustments for each instance's class probabilities which is powered by fuzzy sets with triangular membership functions, transforming a class probability based on the range it belongs to. By solving a nonlinear integer programming problem with a labeled set of ICL class probabilities to minimize class accuracy bias (COBias) and maximize overall accuracy, each class selects an optimal correction function from 19 triangular membership functions without updating an LLM, and the selected functions correct test instances at inference. Across seven benchmark datasets, FuRud reduces COBias by over half (56%) and improves overall accuracy by 21% relatively, outperforming state-of-the-art debiasing methods.
Authors: Fabian Retkowski, Alexander Waibel
Abstract: Large language models (LLMs) struggle with precise length control, particularly in zero-shot settings. We conduct a comprehensive study evaluating LLMs' length control capabilities across multiple measures and propose practical methods to improve controllability. Our experiments with LLaMA 3 reveal stark differences in length adherence across measures and highlight inherent biases of the model. To address these challenges, we introduce a set of methods: length approximation, target adjustment, sample filtering, and automated revisions. By combining these methods, we demonstrate substantial improvements in length compliance while maintaining or enhancing summary quality, providing highly effective zero-shot strategies for precise length control without the need for model fine-tuning or architectural changes. With our work, we not only advance our understanding of LLM behavior in controlled text generation but also pave the way for more reliable and adaptable summarization systems in real-world applications.
Authors: Mingqi Gao, Yixin Liu, Xinyu Hu, Xiaojun Wan, Jonathan Bragg, Arman Cohan
Abstract: Evaluating and ranking the capabilities of different LLMs is crucial for understanding their performance and alignment with human preferences. Due to the high cost and time-consuming nature of human evaluations, an automatic LLM bencher (i.e., an automatic evaluation framework that aims to rank LLMs based on their alignment with human preferences) is indispensable. An automatic LLM bencher consists of four components: the input set (e.g., a user instruction), the evaluation model (e.g., an LLM), the evaluation type (e.g., pairwise comparison), and the aggregation method (e.g., the ELO rating system). However, previous work has not thoroughly explored how to select these components or how their different combinations influence the results. In this work, through controlled experiments, we provide a series of recommendations on how to choose each component to better automate the evaluation of LLMs. Furthermore, we discovered that when evaluating LLMs with similar performance, the performance of the automatic LLM bencher declines sharply, underscoring the limitations of current benchers and calling for future work. Lastly, we found that the evaluation models' performance at the instance level (e.g., the accuracy of selecting the best output) does not always align with their effectiveness when used as a component of a bencher, highlighting the importance of dedicated system-level evaluation of benchers.
Authors: Chen Tang, Bo Lv, Zifan Zheng, Bohao Yang, Kun Zhao, Ning Liao, Xiaoxing Wang, Feiyu Xiong, Zhiyu Li, Nayu Liu, Jingchi Jiang
Abstract: Traditional Mixture-of-Experts (MoE) networks benefit from utilizing multiple smaller expert models as opposed to a single large network. However, these experts typically operate independently, leaving a question open about whether interconnecting these models could enhance the performance of MoE networks. In response, we introduce GRAPHMOE, a novel method aimed at augmenting the cognitive depth of language models via a self-rethinking mechanism constructed on Pseudo GraphMoE networks. GRAPHMOE employs a recurrent routing strategy to simulate iterative thinking steps, thereby facilitating the flow of information among expert nodes. We implement the GRAPHMOE architecture using Low-Rank Adaptation techniques (LoRA) and conduct extensive experiments on various benchmark datasets. The experimental results reveal that GRAPHMOE outperforms other LoRA based models, achieving state-of-the-art (SOTA) performance. Additionally, this study explores a novel recurrent routing strategy that may inspire further advancements in enhancing the reasoning capabilities of language models.
Authors: Zhiyuan Weng, Guikun Chen, Wenguan Wang
Abstract: Recent advancements in large language models (LLMs) revolutionize the field of intelligent agents, enabling collaborative multi-agent systems capable of tackling complex problems across various domains. However, the potential of conformity within these systems, analogous to phenomena like conformity bias and groupthink in human group dynamics, remains largely unexplored, raising concerns about their collective problem-solving capabilities and possible ethical implications. This paper presents a comprehensive study on conformity in LLM-driven multi-agent systems, focusing on three aspects: the existence of conformity, the factors influencing conformity, and potential mitigation strategies. In particular, we introduce BenchForm, a new conformity-oriented benchmark, featuring reasoning-intensive tasks and five distinct interaction protocols designed to probe LLMs' behavior in collaborative scenarios. Several representative LLMs are evaluated on BenchForm, using metrics such as conformity rate and independence rate to quantify conformity's impact. Our analysis delves into factors influencing conformity, including interaction time and majority size, and examines how the subject agent rationalizes its conforming behavior. Furthermore, we explore two strategies to mitigate conformity effects, i.e., developing enhanced personas and implementing a reflection mechanism. Several interesting findings regarding LLMs' conformity are derived from empirical results and case studies. We hope that these insights can pave the way for more robust and ethically-aligned collaborative AI systems. Our benchmark and code are available at BenchForm.
Authors: MediaTek Research, :, Chan-Jan Hsu, Chia-Sheng Liu, Meng-Hsi Chen, Muxi Chen, Po-Chun Hsu, Yi-Chang Chen, Da-Shan Shiu
Abstract: Llama-Breeze2 (hereinafter referred to as Breeze2) is a suite of advanced multi-modal language models, available in 3B and 8B parameter configurations, specifically designed to enhance Traditional Chinese language representation. Building upon the Llama 3.2 model family, we continue the pre-training of Breeze2 on an extensive corpus to enhance the linguistic and cultural heritage of Traditional Chinese. In addition to language modeling capabilities, we significantly augment the models with function calling and vision understanding capabilities. At the time of this publication, as far as we are aware, absent reasoning-inducing prompts, Breeze2 are the strongest performing models in Traditional Chinese function calling and image understanding in its size class. The effectiveness of Breeze2 is benchmarked across various tasks, including Taiwan general knowledge, instruction-following, long context, function calling, and vision understanding. We are publicly releasing all Breeze2 models under the Llama 3.2 Community License. We also showcase the capabilities of the model running on mobile platform with a mobile application which we also open source.
Authors: Ivory Yang, Weicheng Ma, Chunhui Zhang, Soroush Vosoughi
Abstract: Endangered languages, such as Navajo - the most widely spoken Native American language - are significantly underrepresented in contemporary language technologies, exacerbating the challenges of their preservation and revitalization. This study evaluates Google's Language Identification (LangID) tool, which does not currently support any Native American languages. To address this, we introduce a random forest classifier trained on Navajo and twenty erroneously suggested languages by LangID. Despite its simplicity, the classifier achieves near-perfect accuracy (97-100%). Additionally, the model demonstrates robustness across other Athabaskan languages - a family of Native American languages spoken primarily in Alaska, the Pacific Northwest, and parts of the Southwestern United States - suggesting its potential for broader application. Our findings underscore the pressing need for NLP systems that prioritize linguistic diversity and adaptability over centralized, one-size-fits-all solutions, especially in supporting underrepresented languages in a multicultural world. This work directly contributes to ongoing efforts to address cultural biases in language models and advocates for the development of culturally localized NLP tools that serve diverse linguistic communities.
Authors: Huan Ma, Jingdong Chen, Guangyu Wang, Changqing Zhang
Abstract: In recent years, Large Language Models (LLMs) have seen remarkable advancements and have been extensively integrated across various fields. Despite their progress, LLMs are prone to hallucinations, producing responses that may not be dependable if the models lack sufficient grounding knowledge. To mitigate this issue, methods for estimating uncertainty have been adopted, with a focus on critical tokens as indicators of reliability. Nevertheless, probability-based approaches have shown limitations in assessing token-level reliability due to the erosion of evidence strength information acquired during training. In this paper, we introduce Logits-induced Token Uncertainty (LogU), a novel framework designed to estimate token-specific uncertainty in LLMs in real time, without the need for multiple sampling rounds. By leveraging evidence modeling for the implementation of LogU, we utilize the derived uncertainty measures to steer downstream tasks. Our experimental findings highlight the substantial effectiveness and potential of LogU, marking a significant advancement in addressing the challenge of model hallucinations.
Authors: Yi Liu
Abstract: To address the issue of variability in the output generated by a language model, we present a measure of semantic variability that is statistically consistent under mild assumptions. This measure, denoted as semantic spectral entropy, is a easy to implement algorithm that requires just off the shelf language models. We put very few restrictions on the language models and we have shown in a clear simulation studies that such method can generate accurate metric despite randomness that arise from the language models.
Authors: Borui Xu, Yao Chen, Zeyi Wen, Weiguo Liu, Bingsheng He
Abstract: The increasing demand for efficient summarization tools in resource-constrained environments highlights the need for effective solutions. While large language models (LLMs) deliver superior summarization quality, their high computational resource requirements limit practical use applications. In contrast, small language models (SLMs) present a more accessible alternative, capable of real-time summarization on edge devices. However, their summarization capabilities and comparative performance against LLMs remain underexplored. This paper addresses this gap by presenting a comprehensive evaluation of 19 SLMs for news summarization across 2,000 news samples, focusing on relevance, coherence, factual consistency, and summary length. Our findings reveal significant variations in SLM performance, with top-performing models such as Phi3-Mini and Llama3.2-3B-Ins achieving results comparable to those of 70B LLMs while generating more concise summaries. Notably, SLMs are better suited for simple prompts, as overly complex prompts may lead to a decline in summary quality. Additionally, our analysis indicates that instruction tuning does not consistently enhance the news summarization capabilities of SLMs. This research not only contributes to the understanding of SLMs but also provides practical insights for researchers seeking efficient summarization solutions that balance performance and resource use.
Authors: Alex Fl\"uckiger, Chantal Amrhein, Tim Graf, Fr\'ed\'eric Odermatt, Martin P\"omsl, Philippe Schl\"apfer, Florian Schottmann, Samuel L\"aubli
Abstract: As strong machine translation (MT) systems are increasingly based on large language models (LLMs), reliable quality benchmarking requires methods that capture their ability to leverage extended context. This study compares two commercial MT systems -- DeepL and Supertext -- by assessing their performance on unsegmented texts. We evaluate translation quality across four language directions with professional translators assessing segments with full document-level context. While segment-level assessments indicate no strong preference between the systems in most cases, document-level analysis reveals a preference for Supertext in three out of four language directions, suggesting superior consistency across longer texts. We advocate for more context-sensitive evaluation methodologies to ensure that MT quality assessments reflect real-world usability. We release all evaluation data and scripts for further analysis and reproduction at https://github.com/supertext/evaluation_deepl_supertext.
URLs: https://github.com/supertext/evaluation_deepl_supertext.
Authors: Kamer Ali Yuksel, Hassan Sawaf
Abstract: Recent advances in large language models (LLMs) have shown remarkable performance across diverse tasks. However, these models are typically deployed with fixed weights, which limits their ability to adapt dynamically to the variability inherent in real-world data during inference. This paper introduces ChameleonLLM, a novel framework that enables inference-time adaptation of LLMs by leveraging batch-aware clustering and on-the-fly generation of low-rank updates. Unlike traditional fine-tuning approaches such as Low-Rank Adaptation (LoRA) or methods that rely on a fixed set of pre-learned uniforms (changeable masks), our method dynamically generates adaptive modifications to the decoder weights based on the aggregated statistics of clustered batches. By intelligently grouping similar inputs and computing context-aware low-rank updates via a hyper-network, ChameleonLLM achieves significant performance gains, outperforming conventional LoRA methods while eliminating the overhead of maintaining multiple expert models. Our experiments highlight the potential of our approach to serve as a versatile and highly adaptive solution for language model inference. ChameleonLLM is open-sourced to ensure the reproducibility of our experiments: https://anonymous.4open.science/r/ChamaleonLLM/
Authors: Sandra C. Sandoval, Christabel Acquaye, Kwesi Cobbina, Mohammad Nayeem Teli, Hal Daum\'e III
Abstract: We examine the representation of African American English (AAE) in large language models (LLMs), exploring (a) the perceptions Black Americans have of how effective these technologies are at producing authentic AAE, and (b) in what contexts Black Americans find this desirable. Through both a survey of Black Americans ($n=$ 104) and annotation of LLM-produced AAE by Black Americans ($n=$ 228), we find that Black Americans favor choice and autonomy in determining when AAE is appropriate in LLM output. They tend to prefer that LLMs default to communicating in Mainstream U.S. English in formal settings, with greater interest in AAE production in less formal settings. When LLMs were appropriately prompted and provided in context examples, our participants found their outputs to have a level of AAE authenticity on par with transcripts of Black American speech. Select code and data for our project can be found here: https://github.com/smelliecat/AAEMime.git
Authors: Soichiro Murakami, Peinan Zhang, Hidetaka Kamigaito, Hiroya Takamura, Manabu Okumura
Abstract: Effective linguistic choices that attract potential customers play crucial roles in advertising success. This study aims to explore the linguistic features of ad texts that influence human preferences. Although the creation of attractive ad texts is an active area of research, progress in understanding the specific linguistic features that affect attractiveness is hindered by several obstacles. First, human preferences are complex and influenced by multiple factors, including their content, such as brand names, and their linguistic styles, making analysis challenging. Second, publicly available ad text datasets that include human preferences are lacking, such as ad performance metrics and human feedback, which reflect people's interests. To address these problems, we present AdParaphrase, a paraphrase dataset that contains human preferences for pairs of ad texts that are semantically equivalent but differ in terms of wording and style. This dataset allows for preference analysis that focuses on the differences in linguistic features. Our analysis revealed that ad texts preferred by human judges have higher fluency, longer length, more nouns, and use of bracket symbols. Furthermore, we demonstrate that an ad text-generation model that considers these findings significantly improves the attractiveness of a given text. The dataset is publicly available at: https://github.com/CyberAgentAILab/AdParaphrase.
Authors: Roman Vashurin, Maiya Goloburda, Preslav Nakov, Artem Shelmanov, Maxim Panov
Abstract: Uncertainty quantification (UQ) methods for Large Language Models (LLMs) encompasses a variety of approaches, with two major types being particularly prominent: information-based, which focus on model confidence expressed as token probabilities, and consistency-based, which assess the semantic relationship between multiple outputs generated using repeated sampling. Several recent methods have combined these two approaches and shown impressive performance in various applications. However, they sometimes fail to outperform much simpler baseline methods. Our investigation reveals distinctive characteristics of LLMs as probabilistic models, which help to explain why these UQ methods underperform in certain tasks. Based on these findings, we propose a new way of synthesizing model confidence and output consistency that leads to a family of efficient and robust UQ methods. We evaluate our approach across a variety of tasks such as question answering, abstractive summarization, and machine translation, demonstrating sizable improvements over state-of-the-art UQ approaches.
Authors: Ada Defne Tur, Gaurav Kamath, Siva Reddy
Abstract: Though English sentences are typically inflexible vis-\`a-vis word order, constituents often show far more variability in ordering. One prominent theory presents the notion that constituent ordering is directly correlated with constituent weight: a measure of the constituent's length or complexity. Such theories are interesting in the context of natural language processing (NLP), because while recent advances in NLP have led to significant gains in the performance of large language models (LLMs), much remains unclear about how these models process language, and how this compares to human language processing. In particular, the question remains whether LLMs display the same patterns with constituent movement, and may provide insights into existing theories on when and how the shift occurs in human language. We compare a variety of LLMs with diverse properties to evaluate broad LLM performance on four types of constituent movement: heavy NP shift, particle movement, dative alternation, and multiple PPs. Despite performing unexpectedly around particle movement, LLMs generally align with human preferences around constituent ordering.
Authors: Mengxi Xiao, Zihao Jiang, Lingfei Qian, Zhengyu Chen, Yueru He, Yijing Xu, Yuecheng Jiang, Dong Li, Ruey-Ling Weng, Min Peng, Jimin Huang, Sophia Ananiadou, Qianqian Xie
Abstract: Stock movement prediction, a critical task in financial time-series forecasting, relies on identifying and retrieving key influencing factors from vast and complex datasets. However, traditional text-trained or numeric similarity-based retrieval methods often struggle to handle the intricacies of financial data. To address this, we propose the first retrieval-augmented generation (RAG) framework specifically designed for financial time-series forecasting. Our framework incorporates three key innovations: a fine-tuned 1B large language model (StockLLM) as its backbone, a novel candidate selection method enhanced by LLM feedback, and a training objective that maximizes the similarity between queries and historically significant sequences. These advancements enable our retriever, FinSeer, to uncover meaningful patterns while effectively minimizing noise in complex financial datasets. To support robust evaluation, we also construct new datasets that integrate financial indicators and historical stock prices. Experimental results demonstrate that our RAG framework outperforms both the baseline StockLLM and random retrieval methods, showcasing its effectiveness. FinSeer, as the retriever, achieves an 8% higher accuracy on the BIGDATA22 benchmark and retrieves more impactful sequences compared to existing retrieval methods. This work highlights the importance of tailored retrieval models in financial forecasting and provides a novel, scalable framework for future research in the field.
Authors: Saptarshi Ghosh, Tianyu Jiang
Abstract: Metonymy plays an important role in our daily communication. People naturally think about things using their most salient properties or commonly related concepts. For example, by saying "The bus decided to skip our stop today," we actually mean that the bus driver made the decision, not the bus. Prior work on metonymy resolution has mainly focused on named entities. However, metonymy involving common nouns (such as desk, baby, and school) is also a frequent and challenging phenomenon. We argue that NLP systems should be capable of identifying the metonymic use of common nouns in context. We create a new metonymy dataset ConMeC, which consists of 6,000 sentences, where each sentence is paired with a target common noun and annotated by humans to indicate whether that common noun is used metonymically or not in that context. We also introduce a chain-of-thought based prompting method for detecting metonymy using large language models (LLMs). We evaluate our LLM-based pipeline, as well as a supervised BERT model on our dataset and three other metonymy datasets. Our experimental results demonstrate that LLMs could achieve performance comparable to the supervised BERT model on well-defined metonymy categories, while still struggling with instances requiring nuanced semantic understanding. Our dataset is publicly available at: https://github.com/SaptGhosh/ConMeC.
Authors: Kaiqu Liang, Zixu Zhang, Jaime Fern\'andez Fisac
Abstract: Large language models (LLMs) exhibit advanced reasoning skills, enabling robots to comprehend natural language instructions and strategically plan high-level actions through proper grounding. However, LLM hallucination may result in robots confidently executing plans that are misaligned with user goals or even unsafe in critical scenarios. Additionally, inherent ambiguity in natural language instructions can introduce uncertainty into the LLM's reasoning and planning processes.We propose introspective planning, a systematic approach that align LLM's uncertainty with the inherent ambiguity of the task. Our approach constructs a knowledge base containing introspective reasoning examples as post-hoc rationalizations of human-selected safe and compliant plans, which are retrieved during deployment. Evaluations on three tasks, including a newly introduced safe mobile manipulation benchmark, demonstrate that introspection substantially improves both compliance and safety over state-of-the-art LLM-based planning methods. Furthermore, we empirically show that introspective planning, in combination with conformal prediction, achieves tighter confidence bounds, maintaining statistical success guarantees while minimizing unnecessary user clarification requests. The webpage and code are accessible at https://introplan.github.io.
Authors: Kaixuan Ji, Jiafan He, Quanquan Gu
Abstract: Aligning large language models (LLM) with human preference plays a key role in building modern generative models and can be achieved by reinforcement learning from human feedback (RLHF). Despite their superior performance, current RLHF approaches often require a large amount of human-labelled preference data, which is expensive to collect. In this paper, inspired by the success of active learning, we address this problem by proposing query-efficient RLHF methods. We first formalize the alignment problem as a contextual dueling bandit problem and design an active-query-based proximal policy optimization (APPO) algorithm with an $\tilde{O}(d^2/\Delta)$ instance-dependent regret bound and an $\tilde{O}(d^2/\Delta^2)$ query complexity, where $d$ is the dimension of feature space and $\Delta$ is the sub-optimality gap over all the contexts. We then propose ADPO, a practical version of our algorithm based on direct preference optimization (DPO) and apply it to fine-tuning LLMs. Our experiments show that ADPO, while only making about half of queries for human preference, matches the performance of the state-of-the-art DPO method.
Authors: Enrico Liscio, Luciano C. Siebert, Catholijn M. Jonker, Pradeep K. Murukannaiah
Abstract: Understanding citizens' values in participatory systems is crucial for citizen-centric policy-making. We envision a hybrid participatory system where participants make choices and provide motivations for those choices, and AI agents estimate their value preferences by interacting with them. We focus on situations where a conflict is detected between participants' choices and motivations, and propose methods for estimating value preferences while addressing detected inconsistencies by interacting with the participants. We operationalize the philosophical stance that "valuing is deliberatively consequential." That is, if a participant's choice is based on a deliberation of value preferences, the value preferences can be observed in the motivation the participant provides for the choice. Thus, we propose and compare value preferences estimation methods that prioritize the values estimated from motivations over the values estimated from choices alone. Then, we introduce a disambiguation strategy that combines Natural Language Processing and Active Learning to address the detected inconsistencies between choices and motivations. We evaluate the proposed methods on a dataset of a large-scale survey on energy transition. The results show that explicitly addressing inconsistencies between choices and motivations improves the estimation of an individual's value preferences. The disambiguation strategy does not show substantial improvements when compared to similar baselines--however, we discuss how the novelty of the approach can open new research avenues and propose improvements to address the current limitations.
Authors: Neeloy Chakraborty, Melkior Ornik, Katherine Driggs-Campbell
Abstract: Autonomous systems are soon to be ubiquitous, spanning manufacturing, agriculture, healthcare, entertainment, and other industries. Most of these systems are developed with modular sub-components for decision-making, planning, and control that may be hand-engineered or learning-based. While these approaches perform well under the situations they were specifically designed for, they can perform especially poorly in out-of-distribution scenarios that will undoubtedly arise at test-time. The rise of foundation models trained on multiple tasks with impressively large datasets has led researchers to believe that these models may provide "common sense" reasoning that existing planners are missing, bridging the gap between algorithm development and deployment. While researchers have shown promising results in deploying foundation models to decision-making tasks, these models are known to hallucinate and generate decisions that may sound reasonable, but are in fact poor. We argue there is a need to step back and simultaneously design systems that can quantify the certainty of a model's decision, and detect when it may be hallucinating. In this work, we discuss the current use cases of foundation models for decision-making tasks, provide a general definition for hallucinations with examples, discuss existing approaches to hallucination detection and mitigation with a focus on decision problems, present guidelines, and explore areas for further research in this exciting field.
Authors: Tomohiro Nishimoto, Taichi Nishimura, Koki Yamamoto, Keisuke Shirai, Hirotaka Kameko, Yuto Haneji, Tomoya Yoshida, Keiya Kajimura, Taiyu Cui, Chihiro Nishiwaki, Eriko Daikoku, Natsuko Okuda, Fumihito Ono, Shinsuke Mori
Abstract: This paper introduces BioVL-QR, a biochemical vision-and-language dataset comprising 23 egocentric experiment videos, corresponding protocols, and vision-and-language alignments. A major challenge in understanding biochemical videos is detecting equipment, reagents, and containers because of the cluttered environment and indistinguishable objects. Previous studies assumed manual object annotation, which is costly and time-consuming. To address the issue, we focus on Micro QR Codes. However, detecting objects using only Micro QR Codes is still difficult due to blur and occlusion caused by object manipulation. To overcome this, we propose an object labeling method combining a Micro QR Code detector with an off-the-shelf hand object detector. As an application of the method and BioVL-QR, we tackled the task of localizing the procedural steps in an instructional video. The experimental results show that using Micro QR Codes and our method improves biochemical video understanding. Data and code are available through https://nishi10mo.github.io/BioVL-QR/
Authors: Han Zhong, Zikang Shan, Guhao Feng, Wei Xiong, Xinle Cheng, Li Zhao, Di He, Jiang Bian, Liwei Wang
Abstract: In the classical Reinforcement Learning from Human Feedback (RLHF) framework, Proximal Policy Optimization (PPO) is employed to learn from sparse, sentence-level rewards -- a challenging scenario in traditional deep reinforcement learning. Despite the great successes of PPO in the alignment of large language models, its open-source implementation is still largely sub-optimal. To address these issues, we introduce a framework that models RLHF problems as a Markov decision process (MDP), enabling the capture of fine-grained token-wise information. Under this framework, we introduce an algorithm Reinforced Token Optimization (\texttt{RTO}), which learns the token-wise reward function from preference data and performs policy optimization based on this learned token-wise reward signal. Theoretically, \texttt{RTO} is proven to have the capability of finding the near-optimal policy sample-efficiently. For its practical implementation, \texttt{RTO} innovatively integrates Direct Preference Optimization (DPO) and PPO. DPO, originally derived from sparse sentence rewards, surprisingly provides us with a token-wise characterization of response quality, which is seamlessly incorporated into our subsequent PPO training stage. Extensive experiments demonstrate that \texttt{RTO} performs better than PPO and other direct preference learning algorithms. In particular, RTO outperforms PPO by 7.5 points on the AlpacaEval 2 benchmark and by 4.1 points on Arena-Hard. Our code and models are available at \href{https://github.com/zkshan2002/RTO}{https://github.com/zkshan2002/RTO}.
URLs: https://github.com/zkshan2002/RTO, https://github.com/zkshan2002/RTO
Authors: Joseph Cho, Mrudang Mathur, Cyril Zakka, Dhamanpreet Kaur, Matthew Leipzig, Alex Dalal, Aravind Krishnan, Eubee Koo, Karen Wai, Cindy S. Zhao, Rohan Shad, Robyn Fong, Ross Wightman, Akshay Chaudhari, William Hiesinger
Abstract: Deep learning algorithms require extensive data to achieve robust performance. However, data availability is often restricted in the medical domain due to patient privacy concerns. Synthetic data presents a possible solution to these challenges. Recently, image generative models have found increasing use for medical applications but are often designed for singular medical specialties and imaging modalities, thus limiting their broader utility. To address this, we introduce MediSyn: a text-guided, latent diffusion model capable of generating synthetic images from 6 medical specialties and 10 image types. The synthetic images are validated by expert clinicians for alignment with their corresponding text prompts. Furthermore, a direct comparison of the synthetic images against the real images confirms that our model synthesizes novel images and, crucially, may preserve patient privacy. Finally, classifiers trained on a mixture of synthetic and real data achieve similar performance to those trained on twice the amount of real data. Our findings highlight the immense potential for generalist image generative models to accelerate algorithmic research and development in medicine.
Authors: Jingyang Ou, Shen Nie, Kaiwen Xue, Fengqi Zhu, Jiacheng Sun, Zhenguo Li, Chongxuan Li
Abstract: Discrete diffusion models with absorbing processes have shown promise in language modeling. The key quantities to be estimated are the ratios between the marginal probabilities of two transitive states at all timesteps, called the concrete score. In this paper, we reveal that the concrete score in absorbing diffusion can be expressed as conditional probabilities of clean data, multiplied by a time-dependent scalar in an analytic form. Motivated by this finding, we propose reparameterized absorbing discrete diffusion (RADD), a dedicated diffusion model without time-condition that characterizes the time-independent conditional probabilities. Besides its simplicity, RADD can reduce the number of function evaluations (NFEs) by caching the output of the time-independent network when the noisy sample remains unchanged in a sampling interval, which enables sampling acceleration. Built upon the new perspective of conditional distributions, we further unify absorbing discrete diffusion and any-order autoregressive models (AO-ARMs), showing that the upper bound on the negative log-likelihood for the diffusion model can be interpreted as an expected negative log-likelihood for AO-ARMs. Further, our RADD models achieve SOTA performance among diffusion models on 5 zero-shot language modeling benchmarks (measured by perplexity) at the GPT-2 scale. Our code is available at https://github.com/ML-GSAI/RADD.
Authors: Hengyi Wang, Haizhou Shi, Shiwei Tan, Weiyi Qin, Wenyuan Wang, Tunyu Zhang, Akshay Nambi, Tanuja Ganu, Hao Wang
Abstract: Multimodal Large Language Models (MLLMs) have shown significant promise in various applications, leading to broad interest from researchers and practitioners alike. However, a comprehensive evaluation of their long-context capabilities remains underexplored. To address these gaps, we introduce the MultiModal Needle-in-a-haystack (MMNeedle) benchmark, specifically designed to assess the long-context capabilities of MLLMs. Besides multi-image input, we employ image stitching to further increase the input context length, and develop a protocol to automatically generate labels for sub-image level retrieval. Essentially, MMNeedle evaluates MLLMs by stress-testing their capability to locate a target sub-image (needle) within a set of images (haystack) based on textual instructions and descriptions of image contents. This setup necessitates an advanced understanding of extensive visual contexts and effective information retrieval within long-context image inputs. With this benchmark, we evaluate state-of-the-art MLLMs, encompassing both API-based and open-source models. The findings reveal that GPT-4o consistently surpasses other models in long-context scenarios, but suffers from hallucination problems in negative samples, i.e., when needles are not in the haystacks. Our comprehensive long-context evaluation of MLLMs also sheds lights on the considerable performance gap between API-based and open-source models. All the code, data, and instructions required to reproduce the main results are available at https://github.com/Wang-ML-Lab/multimodal-needle-in-a-haystack.
URLs: https://github.com/Wang-ML-Lab/multimodal-needle-in-a-haystack.
Authors: Bruce W. Lee, Inkit Padhi, Karthikeyan Natesan Ramamurthy, Erik Miehling, Pierre Dognin, Manish Nagireddy, Amit Dhurandhar
Abstract: LLMs have shown remarkable capabilities, but precisely controlling their response behavior remains challenging. Existing activation steering methods alter LLM behavior indiscriminately, limiting their practical applicability in settings where selective responses are essential, such as content moderation or domain-specific assistants. In this paper, we propose Conditional Activation Steering (CAST), which analyzes LLM activation patterns during inference to selectively apply or withhold activation steering based on the input context. Our method is based on the observation that different categories of prompts activate distinct patterns in the model's hidden states. Using CAST, one can systematically control LLM behavior with rules like "if input is about hate speech or adult content, then refuse" or "if input is not about legal advice, then refuse." This allows for selective modification of responses to specific content while maintaining normal responses to other content, all without requiring weight optimization. We release an open-source implementation of our framework at
Authors: Ulyana Piterbarg, Lerrel Pinto, Rob Fergus
Abstract: Software engineers mainly write code by editing existing programs. In contrast, language models (LMs) autoregressively synthesize programs in a single pass. One explanation for this is the scarcity of sequential edit data. While high-quality instruction data for code synthesis is scarce, edit data for synthesis is even scarcer. To fill this gap, we develop a synthetic data generation algorithm called LintSeq. This algorithm refactors programs into sequences of synthetic edits by using a linter to procedurally sample across interdependent lines of source code. Synthetic edits sampled with LintSeq reflect the syntax and semantics of their programming language. To test the algorithm, we use it to refactor a dataset of instruction + program pairs into instruction + program-diff-sequence tuples. Then, we fine-tune a series of smaller LMs ranging from 2.6B to 14B parameters on both the re-factored and original versions of this dataset. We perform comprehensive evaluations comparing edit sequence code LMs against baselines on HumanEval, MBPP(+), CodeContests, DS-1000, and BigCodeBench. We show that models fine-tuned to iteratively synthesize code match or outperform baselines on pass@1, and exhibit better scaling across higher pass@k as a function of total test-time FLOPs. Finally, we also pretrain our own tiny LMs for code understanding. We show that fine-tuning these models to synthesize code edit-by-edit results in strong performance on HumanEval and MBPP(+) compared to existing code language models of similar scale such as CodeT5+, AlphaCode, and Codex.
Authors: Yuto Haneji, Taichi Nishimura, Hirotaka Kameko, Keisuke Shirai, Tomoya Yoshida, Keiya Kajimura, Koki Yamamoto, Taiyu Cui, Tomohiro Nishimoto, Shinsuke Mori
Abstract: Mistake action detection is crucial for developing intelligent archives that detect workers' errors and provide feedback. Existing studies have focused on visually apparent mistakes in free-style activities, resulting in video-only approaches to mistake detection. However, in text-following activities, models cannot determine the correctness of some actions without referring to the texts. Additionally, current mistake datasets rarely use procedural texts for video recording except for cooking. To fill these gaps, this paper proposes the EgoOops dataset, where egocentric videos record erroneous activities when following procedural texts across diverse domains. It features three types of annotations: video-text alignment, mistake labels, and descriptions for mistakes. We also propose a mistake detection approach, combining video-text alignment and mistake label classification to leverage the texts. Our experimental results show that incorporating procedural texts is essential for mistake detection. Data is available through https://y-haneji.github.io/EgoOops-project-page/.
Authors: Kaixuan Ji, Guanlin Liu, Ning Dai, Qingping Yang, Renjie Zheng, Zheng Wu, Chen Dun, Quanquan Gu, Lin Yan
Abstract: Reinforcement Learning (RL) plays a crucial role in aligning large language models (LLMs) with human preferences and improving their ability to perform complex tasks. However, current approaches either require significant computational resources due to the use of multiple models and extensive online sampling for training (e.g., PPO) or are framed as bandit problems (e.g., DPO, DRO), which often struggle with multi-step reasoning tasks, such as math problem solving and complex reasoning that involve long chains of thought. To overcome these limitations, we introduce Direct Q-function Optimization (DQO), which formulates the response generation process as a Markov Decision Process (MDP) and utilizes the soft actor-critic (SAC) framework to optimize a Q-function directly parameterized by the language model. The MDP formulation of DQO offers structural advantages over bandit-based methods, enabling more effective process supervision. Experimental results on two math problem-solving datasets, GSM8K and MATH, demonstrate that DQO outperforms previous methods, establishing it as a promising offline reinforcement learning approach for aligning language models.
Authors: Jingyang Qiao, Zhizhong Zhang, Xin Tan, Yanyun Qu, Shouhong Ding, Yuan Xie
Abstract: Continual Instruction Tuning (CIT) is adopted to continually instruct Large Models to follow human intent data by data. It is observed that existing gradient update would heavily destroy the performance on previous datasets during CIT process. Instead, Exponential Moving Average (EMA), owns the ability to trace previous parameters, which can aid in decreasing forgetting. Nonetheless, its stable balance weight fails to deal with the ever-changing datasets, leading to the out-of-balance between plasticity and stability. In this paper, we propose a general continual instruction tuning framework to address the challenge. Starting from the trade-off prerequisite and EMA update, we propose the plasticity and stability ideal condition. Based on Taylor expansion in the loss function, we find the optimal balance weight can be automatically determined by the gradients and learned parameters. Therefore, we propose a stable-plasticity balanced coefficient to avoid knowledge confusion. Based on the semantic similarity of the instructions, we can determine whether to retrain or expand the training parameters and allocate the most suitable parameters for the testing instances. Extensive experiments across multiple continual instruction tuning benchmarks demonstrate that our approach not only enhances anti-forgetting capabilities but also significantly improves overall continual tuning performance. For example, based on LLaVA-7B, the forgetting is reduced from 5.42 to 1.93. Our code will be made publicly available soon.
Authors: Wanying Wang, Zeyu Ma, Pengfei Liu, Mingang Chen
Abstract: While various vertical domain large language models (LLMs) have been developed, automatically evaluating their performance across different domains remains a critical challenge. Current benchmark-based methods often rely on static and costly datasets, are misaligned with practical user needs, and lack flexibility across domains. To address these limitations, we revisit the evaluation process and introduce two key concepts: Benchmark+, which extends the traditional question-answer benchmark into a more flexible ``strategy-criterion'' format; and Assessment+, which enhances the interaction process, enabling deeper exploration and supporting analysis from broader perspectives. We propose TestAgent, an agent-based evaluation framework that implements these concepts using retrieval-augmented generation and reinforcement learning. TestAgent enables automatic dynamic benchmark generation and in-depth assessment across diverse vertical domain scenarios. Experiments on tasks ranging from constructing multiple vertical domain evaluations to converting static benchmarks into dynamic forms demonstrate the effectiveness of TestAgent. This work offers an interesting perspective on automatic evaluation for LLMs and highlights a pathway for dynamic and domain-adaptive assessments.
Authors: Dehai Min, Zhiyang Xu, Guilin Qi, Lifu Huang, Chenyu You
Abstract: Existing information retrieval (IR) models often assume a homogeneous structure for knowledge sources and user queries, limiting their applicability in real-world settings where retrieval is inherently heterogeneous and diverse. In this paper, we introduce UniHGKR, a unified instruction-aware heterogeneous knowledge retriever that (1) builds a unified retrieval space for heterogeneous knowledge and (2) follows diverse user instructions to retrieve knowledge of specified types. UniHGKR consists of three principal stages: heterogeneous self-supervised pretraining, text-anchored embedding alignment, and instruction-aware retriever fine-tuning, enabling it to generalize across varied retrieval contexts. This framework is highly scalable, with a BERT-based version and a UniHGKR-7B version trained on large language models. Also, we introduce CompMix-IR, the first native heterogeneous knowledge retrieval benchmark. It includes two retrieval scenarios with various instructions, over 9,400 question-answer (QA) pairs, and a corpus of 10 million entries, covering four different types of data. Extensive experiments show that UniHGKR consistently outperforms state-of-the-art methods on CompMix-IR, achieving up to 6.36% and 54.23% relative improvements in two scenarios, respectively. Finally, by equipping our retriever for open-domain heterogeneous QA systems, we achieve a new state-of-the-art result on the popular ConvMix task, with an absolute improvement of up to 5.90 points.
Authors: Tom A. Lamb, Adam Davies, Alasdair Paren, Philip H. S. Torr, Francesco Pinto
Abstract: Despite the success of Instruction Tuning (IT) in training large language models (LLMs) to perform arbitrary user-specified tasks, these models often still leverage spurious or biased features learned from their training data, leading to undesired behaviours when deploying them in new contexts. In this work, we introduce Focus Instruction Tuning (FIT), which trains LLMs to condition their responses by focusing on specific features whilst ignoring others, leading to different behaviours based on what features are specified. Across several experimental settings, we show that focus-tuned models can be adaptively steered by focusing on different features at inference-time: for instance, robustness can be improved by focusing on task-causal features and ignoring spurious features, and social bias can be mitigated by ignoring demographic categories. Furthermore, FIT can steer behaviour in new contexts, generalising under distribution shift and to new unseen features at inference time, and thereby facilitating more robust, fair, and controllable LLM applications in real-world environments.
Authors: Xiangzhe Xu, Zian Su, Jinyao Guo, Kaiyuan Zhang, Zhenting Wang, Xiangyu Zhang
Abstract: Recent advances in code-specific large language models (LLMs) have greatly enhanced code generation and refinement capabilities. However, the safety of code LLMs remains under-explored, posing potential risks as insecure code generated by these models may introduce vulnerabilities into real-world systems. Previous work proposes to collect security-focused instruction-tuning dataset from real-world vulnerabilities. It is constrained by the data sparsity of vulnerable code, and has limited applicability in the iterative post-training workflows of modern LLMs. In this paper, we propose ProSec, a novel proactive security alignment approach designed to align code LLMs with secure coding practices. ProSec systematically exposes the vulnerabilities in a code LLM by synthesizing error-inducing coding scenarios from Common Weakness Enumerations (CWEs), and generates fixes to vulnerable code snippets, allowing the model to learn secure practices through advanced preference learning objectives. The scenarios synthesized by ProSec triggers 25 times more vulnerable code than a normal instruction-tuning dataset, resulting in a security-focused alignment dataset 7 times larger than the previous work. Experiments show that models trained with ProSec are 25.2% to 91.4% more secure compared to previous work without degrading models' utility.
Authors: Kaiqu Liang, Haimin Hu, Ryan Liu, Thomas L. Griffiths, Jaime Fern\'andez Fisac
Abstract: While Reinforcement Learning from Human Feedback (RLHF) has shown promise in aligning generative AI, we present empirical evidence that it can also cause severe, systematic misalignment. We hypothesize that this stems from evaluator feedback depending on downstream outcome predictions (foresight) that can be influenced by the AI's output, inducing Goodhart's law dynamics. Conversely, our theoretical analysis shows that conditioning evaluator feedback on downstream observations (hindsight) inhibits this effect by decoupling the alignment signal from potentially compromised predictions-crucially, the result holds even if the observed outcomes are sampled from the AI's own world model. Building on this insight, we introduce Reinforcement Learning from Hindsight Simulation (RLHS), which presents plausible simulated outcomes to evaluators before eliciting feedback. We demonstrate RLHS on online (PPO) and offline (DPO) large language model fine-tuning, obtaining superior alignment over RLHF in controlled consultancy-type experiments and user studies. We evaluate post-hoc on the TruthfulQA benchmark and find that, even after single-task fine-tuning, both RLHF misalignment and RLHS alignment carry over to substantially different settings.
Authors: Linghe Wang, Minhwa Lee, Ross Volkov, Luan Tuyen Chau, Dongyeop Kang
Abstract: Writing is a cognitively demanding task involving continuous decision-making, heavy use of working memory, and frequent switching between multiple activities. Scholarly writing is particularly complex as it requires authors to coordinate many pieces of multiform knowledge. To fully understand writers' cognitive thought process, one should fully decode the end-to-end writing data (from individual ideas to final manuscript) and understand their complex cognitive mechanisms in scholarly writing. We introduce ScholaWrite dataset, the first-of-its-kind keystroke logs of an end-to-end scholarly writing process for complete manuscripts, with thorough annotations of cognitive writing intentions behind each keystroke. Our dataset includes LaTeX-based keystroke data from five preprints with nearly 62K total text changes and annotations across 4 months of paper writing. ScholaWrite shows promising usability and applications (e.g., iterative self-writing) for the future development of AI writing assistants for academic research, which necessitate complex methods beyond LLM prompting. Our experiments clearly demonstrated the importance of collection of end-to-end writing data, rather than the final manuscript, for the development of future writing assistants to support the cognitive thinking process of scientists. Our de-identified dataset, demo, and code repository are available on our project page.
Authors: Kunfeng Lai, Zhenheng Tang, Xinglin Pan, Peijie Dong, Xiang Liu, Haolan Chen, Li Shen, Bo Li, Xiaowen Chu
Abstract: Model merging aggregates Large Language Models (LLMs) finetuned on different tasks into a stronger one. However, parameter conflicts between models leads to performance degradation in averaging. While model routing addresses this issue by selecting individual models during inference, it imposes excessive storage and compute costs, and fails to leverage the common knowledge from different models. In this work, we observe that different layers exhibit varying levels of parameter conflicts. Building on this insight, we average layers with minimal parameter conflicts and use a novel task-level expert routing for layers with significant conflicts. To further reduce storage costs, inspired by task arithmetic sparsity, we decouple multiple fine-tuned experts into a dense expert and several sparse experts. Considering the out-of-distribution samples, we select and merge appropriate experts based on the task uncertainty of the input data. We conduct extensive experiments on both LLaMA and Qwen with varying parameter scales, and evaluate on real-world reasoning tasks. Results demonstrate that our method consistently achieves significant performance improvements while requiring less system cost compared to existing methods.
Authors: Daman Arora, Andrea Zanette
Abstract: Scaling model size and training data has led to great advances in the performance of Large Language Models (LLMs). However, the diminishing returns of this approach necessitate alternative methods to improve model capabilities, particularly in tasks requiring advanced reasoning. Large reasoning models, which leverage long chain-of-thoughts, bring unprecedented breakthroughs in problem-solving capabilities but at a substantial deployment cost associated to longer generations. Reducing inference costs is crucial for the economic feasibility, user experience, and environmental sustainability of these models. In this work, we propose to train large reasoning models to reason efficiently. More precisely, we use reinforcement learning (RL) to train reasoning models to dynamically allocate inference-time compute based on task complexity. Our method incentivizes models to minimize unnecessary computational overhead while maintaining accuracy, thereby achieving substantial efficiency gains. It enables the derivation of a family of reasoning models with varying efficiency levels, controlled via a single hyperparameter. Experiments on two open-weight large reasoning models demonstrate significant reductions in inference cost while preserving most of the accuracy.
Authors: Jian Xu, Sichun Luo, Xiangyu Chen, Haoming Huang, Hanxu Hou, Linqi Song
Abstract: Large Language Models (LLMs) have been integrated into recommendation systems to enhance user behavior comprehension. The Retrieval Augmented Generation (RAG) technique is further incorporated into these systems to retrieve more relevant items and improve system performance. However, existing RAG methods rely primarily on textual semantics and often fail to incorporate the most relevant items, limiting the effectiveness of the systems. In this paper, we propose Representation learning for retrieval-Augmented Large Language model Recommendation (RALLRec). Specifically, we enhance textual semantics by prompting LLMs to generate more detailed item descriptions, followed by joint representation learning of textual and collaborative semantics, which are extracted by the LLM and recommendation models, respectively. Considering the potential time-varying characteristics of user interest, a simple yet effective reranking method is further introduced to capture the dynamics of user preference. We conducted extensive experiments on three real-world datasets, and the evaluation results validated the effectiveness of our method. Code is made public at https://github.com/JianXu95/RALLRec.
Authors: Yibo Wang, Congying Xia, Wenting Zhao, Jiangshu Du, Chunyu Miao, Zhongfen Deng, Philip S. Yu, Chen Xing
Abstract: Unit test generation has become a promising and important use case of LLMs. However, existing evaluation benchmarks for assessing LLM unit test generation capabilities focus on function- or class-level code rather than more practical and challenging project-level codebases. To address such limitation, we propose ProjectTest, a project-level benchmark for unit test generation covering Python, Java, and JavaScript. ProjectTest features 20 moderate-sized and high-quality projects per language. We evaluate nine frontier LLMs on ProjectTest and the results show that all frontier LLMs tested exhibit moderate performance on ProjectTest on Python and Java, highlighting the difficulty of ProjectTest. We also conduct a thorough error analysis, which shows that even frontier LLMs, such as Claude-3.5-Sonnet, have significant simple errors, including compilation and cascade errors. Motivated by this observation, we further evaluate all frontier LLMs under manual error-fixing and self-error-fixing scenarios to assess their potential when equipped with error-fixing mechanisms.