Authors: Violet B, John M. Conroy, Sean Lynch, Danielle M, Neil P. Molino, Aaron Wiechmann, Julia S. Yang
Abstract: Analysts require attribution, as nothing can be reported without knowing the source of the information. In this paper, we will focus on automatic methods for attribution, linking each sentence in the summary to a portion of the source text, which may be in one or more documents. We explore using a hybrid summarization, i.e., an automatic paraphrase of an extractive summary, to ease attribution. We also use a custom topology to identify the proportion of different categories of attribution-related errors.
Authors: Encheng Xie, Yihang Sun, Tao Feng, Jiaxuan You
Abstract: Large Language Model (LLM) routing has demonstrated strong capability in balancing response quality with computational cost. As users exhibit diverse preferences, personalization has attracted increasing attention in LLM routing, since even identical queries may require different models to generate responses tailored to individual needs. However, existing approaches are not fully personalized and often fail to capture the complex interactions between specific users and LLMs. Moreover, user preference data is typically scarce, noisy, and inconsistent in format, which limits the effectiveness of methods that rely solely on user-specific data. To address these challenges, we propose GMTRouter, which represents multi-turn user-LLM interactions as a heterogeneous graph with four node types: user, LLM, query, and response, thereby preserving the rich relational structure of the interaction. Through a tailored message-passing mechanism, GMTRouter learns to capture user preferences from few-shot data within a lightweight inductive graph learning framework, enabling effective personalization. Extensive experiments demonstrate that GMTRouter consistently outperforms strong baselines, achieving 0.9 to 21.6 percent higher accuracy and 0.006 to 0.309 higher AUC across multiple datasets. More importantly, we demonstrate that GMTRouter can adapt to new users and evolving preferences using only few-shot data, without extensive fine-tuning. The code for GMTRouter is publicly available at https://github.com/ulab-uiuc/GMTRouter.
Authors: Azza Bouleimen, Giordano De Marzo, Taehee Kim, Nicol`o Pagan, Hannah Metzler, Silvia Giordano, David Garcia
Abstract: Large Language Models (LLMs) offer new avenues to simulate online communities and social media. Potential applications range from testing the design of content recommendation algorithms to estimating the effects of content policies and interventions. However, the validity of using LLMs to simulate conversations between various users remains largely untested. We evaluated whether LLMs can convincingly mimic human group conversations on social media. We collected authentic human conversations from Reddit and generated artificial conversations on the same topic with two LLMs: Llama 3 70B and GPT-4o. When presented side-by-side to study participants, LLM-generated conversations were mistaken for human-created content 39\% of the time. In particular, when evaluating conversations generated by Llama 3, participants correctly identified them as AI-generated only 56\% of the time, barely better than random chance. Our study demonstrates that LLMs can generate social media conversations sufficiently realistic to deceive humans when reading them, highlighting both a promising potential for social simulation and a warning message about the potential misuse of LLMs to generate new inauthentic social media content.
Authors: Abha Jha, Abel Salinas, Fred Morstatter
Abstract: The rise of Large Language Models (LLMs) offers transformative potential for interpreting complex legal frameworks, such as Title 18 Section 175 of the US Code, which governs biological weapons. These systems hold promise for advancing legal analysis and compliance monitoring in sensitive domains. However, this capability comes with a troubling contradiction: while LLMs can analyze and interpret laws, they also demonstrate alarming vulnerabilities in generating unsafe outputs, such as actionable steps for bioweapon creation, despite their safeguards. To address this challenge, we propose a methodology that integrates knowledge graph construction with Retrieval-Augmented Generation (RAG) to systematically evaluate LLMs' understanding of this law, their capacity to assess legal intent (mens rea), and their potential for unsafe applications. Through structured experiments, we assess their accuracy in identifying legal violations, generating prohibited instructions, and detecting unlawful intent in bioweapons-related scenarios. Our findings reveal significant limitations in LLMs' reasoning and safety mechanisms, but they also point the way forward. By combining enhanced safety protocols with more robust legal reasoning frameworks, this research lays the groundwork for developing LLMs that can ethically and securely assist in sensitive legal domains - ensuring they act as protectors of the law rather than inadvertent enablers of its violation.
Authors: Stewart Slocum, Asher Parker-Sartori, Dylan Hadfield-Menell
Abstract: The ability of LLMs to represent diverse perspectives is critical as they increasingly impact society. However, recent studies reveal that alignment algorithms such as RLHF and DPO significantly reduce the diversity of LLM outputs. Not only do aligned LLMs generate text with repetitive structure and word choice, they also approach problems in more uniform ways, and their responses reflect a narrower range of societal perspectives. We attribute this problem to the KL divergence regularizer employed in preference learning algorithms. This causes the model to systematically overweight majority opinions and sacrifice diversity in its outputs. To address this, we propose Soft Preference Learning, which decouples the entropy and cross-entropy terms in the KL penalty - allowing for fine-grained control over LLM generation diversity. From a capabilities perspective, LLMs trained using Soft Preference Learning attain higher accuracy on difficult repeated sampling tasks and produce outputs with greater semantic and lexical diversity. From an alignment perspective, they are capable of representing a wider range of societal viewpoints and display improved logit calibration. Notably, Soft Preference Learning resembles, but is a Pareto improvement over, standard temperature scaling.
Authors: Joongho Kim, Xirui Huang, Zarreen Reza, Gabriel Grand, Kevin Zhu, Ryan Lagasse
Abstract: Tree-of-Thought (ToT) reasoning boosts the problem-solving abilities of Large Language Models (LLMs) but is computationally expensive due to semantic redundancy, where distinct branches explore equivalent reasoning paths. We introduce Semantic Similarity-Based Dynamic Pruning (SSDP), a lightweight method that, to the best of our knowledge, is the first framework to integrate online semantic merging into parallelized tree search, enabling the clustering and pruning of redundant steps in real time. Across reasoning benchmarks, including GSM8K and MATH500, SSDP achieves up to a 2.3x speedup over state-of-the-art tree-search baselines while maintaining competitive accuracy (typically within 5% of the strongest baseline) and reducing the number of explored nodes by 85-90%, demonstrating a practical approach to efficient, scalable LLM reasoning. The implementation of SSDP is publicly available at https://github.com/kimjoonghokim/SSDP.
Authors: Arka Dutta, Sujan Dutta, Rijul Magu, Soumyajit Datta, Munmun De Choudhury, Ashiqur R. KhudaBukhsh
Abstract: Hallucinations pose a critical challenge to the real-world deployment of large language models (LLMs) in high-stakes domains. In this paper, we present a framework for stress testing factual fidelity in LLMs in the presence of adversarial nudge. Our framework consists of three steps. In the first step, we instruct the LLM to produce sets of truths and lies consistent with the closed domain in question. In the next step, we instruct the LLM to verify the same set of assertions as truths and lies consistent with the same closed domain. In the final step, we test the robustness of the LLM against the lies generated (and verified) by itself. Our extensive evaluation, conducted using five widely known proprietary LLMs across two closed domains of popular movies and novels, reveals a wide range of susceptibility to adversarial nudges: \texttt{Claude} exhibits strong resilience, \texttt{GPT} and \texttt{Grok} demonstrate moderate resilience, while \texttt{Gemini} and \texttt{DeepSeek} show weak resilience. Considering that a large population is increasingly using LLMs for information seeking, our findings raise alarm.
Authors: Heehwan Kim, Sungjune Park, Daeseon Choi
Abstract: Large Language Models (LLMs) are generally equipped with guardrails to block the generation of harmful responses. However, existing defenses always assume that an external attacker crafts the harmful query, and the possibility of a model's own output becoming a new attack vector has not been sufficiently explored. In this study, we propose the Self-HarmLLM scenario, which uses a Mitigated Harmful Query (MHQ) generated by the same model as a new input. An MHQ is an ambiguous query whose original intent is preserved while its harmful nature is not directly exposed. We verified whether a jailbreak occurs when this MHQ is re-entered into a separate session of the same model. We conducted experiments on GPT-3.5-turbo, LLaMA3-8B-instruct, and DeepSeek-R1-Distill-Qwen-7B under Base, Zero-shot, and Few-shot conditions. The results showed up to 52% transformation success rate and up to 33% jailbreak success rate in the Zero-shot condition, and up to 65% transformation success rate and up to 41% jailbreak success rate in the Few-shot condition. By performing both prefix-based automated evaluation and human evaluation, we found that the automated evaluation consistently overestimated jailbreak success, with an average difference of 52%. This indicates that automated evaluation alone is not accurate for determining harmfulness. While this study is a toy-level study based on a limited query set and evaluators, it proves that our method can still be a valid attack scenario. These results suggest the need for a fundamental reconsideration of guardrail design and the establishment of a more robust evaluation methodology.
Authors: Yanhong Li, Tianyang Xu, Kenan Tang, Karen Livescu, David McAllester, Jiawei Zhou
Abstract: Knowledge-intensive question answering is central to large language models (LLMs) and is typically assessed using static benchmarks derived from sources like Wikipedia and textbooks. However, these benchmarks fail to capture evolving knowledge in a dynamic world, and centralized curation struggles to keep pace with rapid LLM advancements. To address these drawbacks, we propose Open Knowledge Bench (OKBench), a fully automated framework for generating high-quality, dynamic knowledge benchmarks on demand. Focusing on the news domain where knowledge updates daily, OKBench is an agentic framework that automates the sourcing, creation, validation, and distribution of benchmarks. Our approach democratizes benchmark creation and facilitates thorough evaluation of retrieval-augmented methods by reducing overlap with pretraining data. We evaluate our framework on a wide range open-source and proprietary LLMs of various sizes and configurations, both with and without retrieval over freshly generated knowledge. Our results reveal distinct model behaviors when confronted with new information and highlight how retrieval narrows the performance gap between small and large models. These findings underscore the importance of evaluating LLMs on evolving knowledge benchmarks.
Authors: Yilan Liu
Abstract: Clinical vignettes are essential educational tools in speech-language pathology (SLP), but manual creation is time-intensive. While general-purpose large language models (LLMs) can generate text, they lack domain-specific knowledge, leading to hallucinations and requiring extensive expert revision. This study presents a proof-of-concept system integrating retrieval-augmented generation (RAG) with curated knowledge bases to generate pediatric SLP case materials. A multi-model RAG-based system was prototyped integrating curated domain knowledge with engineered prompt templates, supporting five commercial (GPT-4o, Claude 3.5 Sonnet, Gemini 2.5 Pro) and open-source (Llama 3.2, Qwen 2.5-7B) LLMs. Seven test scenarios spanning diverse disorder types and grade levels were systematically designed. Generated cases underwent automated quality assessment using a multi-dimensional rubric evaluating structural completeness, internal consistency, clinical appropriateness, and IEP goal/session note quality. This proof-of-concept demonstrates technical feasibility for RAG-augmented generation of pediatric SLP vignettes. Commercial models showed marginal quality advantages, but open-source alternatives achieved acceptable performance, suggesting potential for privacy-preserving institutional deployment. Integration of curated knowledge bases enabled content generation aligned with professional guidelines. Extensive validation through expert review, student pilot testing, and psychometric evaluation is required before educational or research implementation. Future applications may extend to clinical decision support, automated IEP goal generation, and clinical reflection training.
Authors: Nazmoon Falgunee Moon
Abstract: In [4], the authors present the DisCoCirc (Distributed Compositional Circuits) formalism for the English language, a grammar-based framework derived from the production rules that incorporates circuit-like representations in order to give a precise categorical theoretical structure to the language. In this paper, we extend this approach to develop a similar framework for Bengali and apply it to translation tasks between English and Bengali. A central focus of our work lies in reassessing the effectiveness of DisCoCirc in reducing language bureaucracy. Unlike the result suggested in [5], our findings indicate that although it works well for a large part of the language, it still faces limitations due to the structural variation of the two languages. We discuss the possible methods that might handle these shortcomings and show that, in practice, DisCoCirc still struggles even with relatively simple sentences. This divergence from prior claims not only highlights the framework's constraints in translation but also suggest scope for future improvement. Apart from our primary focus on English-Bengali translation, we also take a short detour to examine English conjunctions, following [1], showing a connection between conjunctions and Boolean logic.
Authors: Azmine Toushik Wasi, Wahid Faisal, Mst Rafia Islam
Abstract: Bangladesh's low-income population faces major barriers to affordable legal advice due to complex legal language, procedural opacity, and high costs. Existing AI legal assistants lack Bengali-language support and jurisdiction-specific adaptation, limiting their effectiveness. To address this, we developed Mina, a multilingual LLM-based legal assistant tailored for the Bangladeshi context. It employs multilingual embeddings and a RAG-based chain-of-tools framework for retrieval, reasoning, translation, and document generation, delivering context-aware legal drafts, citations, and plain-language explanations via an interactive chat interface. Evaluated by law faculty from leading Bangladeshi universities across all stages of the 2022 and 2023 Bangladesh Bar Council Exams, Mina scored 75-80% in Preliminary MCQs, Written, and simulated Viva Voce exams, matching or surpassing average human performance and demonstrating clarity, contextual understanding, and sound legal reasoning. These results confirm its potential as a low-cost, multilingual AI assistant that automates key legal tasks and scales access to justice, offering a real-world case study on building domain-specific, low-resource systems and addressing challenges of multilingual adaptation, efficiency, and sustainable public-service AI deployment.
Authors: Sergey K. Aityan, Abdolreza Mosaddegh, Rolando Herrero, Haitham Tayyar, Jiang Han, Vikram Sawant, Qi Chen, Rishabh Jain, Aruna Senthamaraikannan, Stephen Wood, Manuel Mersini, Rita Lazzaro, Mario Balzaneli, Nicola Iacovazzo, Ciro Gargiulo Isacco
Abstract: Medical decision-support and advising systems are critical for emergency physicians to quickly and accurately assess patients' conditions and make diagnosis. Artificial Intelligence (AI) has emerged as a transformative force in healthcare in recent years and Large Language Models (LLMs) have been employed in various fields of medical decision-support systems. We studied responses of a group of different LLMs to real cases in emergency medicine. The results of our study on five most renown LLMs showed significant differences in capabilities of Large Language Models for diagnostics acute diseases in medical emergencies with accuracy ranging between 58% and 65%. This accuracy significantly exceeds the reported accuracy of human doctors. We built a super-learner MEDAS (Medical Emergency Diagnostic Advising System) of five major LLMs - Gemini, Llama, Grok, GPT, and Claude). The super-learner produces higher diagnostic accuracy, 70%, even with a quite basic meta-learner. However, at least one of the integrated LLMs in the same super-learner produces 85% correct diagnoses. The super-learner integrates a cluster of LLMs using a meta-learner capable of learning different capabilities of each LLM to leverage diagnostic accuracy of the model by collective capabilities of all LLMs in the cluster. The results of our study showed that aggregated diagnostic accuracy provided by a meta-learning approach exceeds that of any individual LLM, suggesting that the super-learner can take advantage of the combined knowledge of the medical datasets used to train the group of LLMs.
Authors: Yibai Liu, Shihang Wang, Zeming Liu, Zheming Song, Junzhe Wang, Jingjing Liu, Qingjie Liu, Yunhong Wang
Abstract: Despite large language models (LLMs) have achieved impressive achievements across numerous tasks, supervised fine-tuning (SFT) remains essential for adapting these models to specialized domains. However, SFT for domain specialization can be resource-intensive and sometimes leads to a deterioration in performance over general capabilities due to catastrophic forgetting (CF). To address these issues, we propose a self-adaptive gradient-aware data selection approach (GrADS) for supervised fine-tuning of LLMs, which identifies effective subsets of training data by analyzing gradients obtained from a preliminary training phase. Specifically, we design self-guided criteria that leverage the magnitude and statistical distribution of gradients to prioritize examples that contribute the most to the model's learning process. This approach enables the acquisition of representative samples that enhance LLMs understanding of domain-specific tasks. Through extensive experimentation with various LLMs across diverse domains such as medicine, law, and finance, GrADS has demonstrated significant efficiency and cost-effectiveness. Remarkably, utilizing merely 5% of the selected GrADS data, LLMs already surpass the performance of those fine-tuned on the entire dataset, and increasing to 50% of the data results in significant improvements! With catastrophic forgetting substantially mitigated simultaneously. We will release our code for GrADS later.
Authors: Mohaiminul Islam Bhuiyan, Nur Shazwani Kamarudin, Nur Hafieza Ismail
Abstract: Worldwide, suicide is the second leading cause of death for adolescents with past suicide attempts to be an important predictor for increased future suicides. While some people with suicidal thoughts may try to suppress them, many signal their intentions in social media platforms. To address these issues, we propose a new type of hybrid deep learning scheme, i.e., the combination of a CNN architecture and a BiGRU technique, which can accurately identify the patterns of suicidal ideation from SN datasets. Also, we apply Explainable AI methods using SHapley Additive exPlanations to interpret the prediction results and verifying the model reliability. This integration of CNN local feature extraction, BiGRU bidirectional sequence modeling, attention mechanisms, and SHAP interpretability provides a comprehensive framework for suicide detection. Training and evaluation of the system were performed on a publicly available dataset. Several performance metrics were used for evaluating model performance. Our method was found to have achieved 93.97 accuracy in experimental results. Comparative study to different state-of-the-art Machine Learning and DL models and existing literature demonstrates the superiority of our proposed technique over all the competing methods.
Authors: Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha
Abstract: LLM agents extend large language models with tool-calling capabilities, but ambiguous user instructions often lead to incorrect invocations and task failures. We introduce a principled formulation of structured uncertainty over tool-call parameters, modeling joint tool-argument clarification as a POMDP with Expected Value of Perfect Information (EVPI) objective for optimal question selection and aspect-based cost modeling to prevent redundancy. Our SAGE-Agent leverages this structured uncertainty to achieve superior efficiency: increasing coverage on ambiguous tasks by 7-39\% while reducing clarification questions by 1.5-2.7$\times$ compared to strong prompting and uncertainty-based baselines. We present ClarifyBench, the first multi-turn tool-augmented disambiguation benchmark with realistic LLM-based user simulation across diverse domains including document editing, vehicle control, and travel booking. Additionally, we demonstrate that structured uncertainty provides effective training signals for reinforcement learning, boosting When2Call accuracy from 36.5\% to 65.2\% (3B model) and 36.7\% to 62.9\% (7B model) through uncertainty-weighted GRPO training. These results establish structured uncertainty as a principled, efficient approach for tool-augmented agents, improving both task success and interaction efficiency in real-world scenarios.
Authors: Varada Khanna (Department of Biomedical Informatics and Data Science, Yale School of Medicine, New Haven, CT), Nilay Bhatt (Department of Biomedical Informatics and Data Science, Yale School of Medicine, New Haven, CT), Ikgyu Shin (Department of Biomedical Informatics and Data Science, Yale School of Medicine, New Haven, CT), Sule Tinaz (Department of Neurology, Yale School of Medicine, New Haven, CT), Yang Ren (Department of Biomedical Informatics and Data Science, Yale School of Medicine, New Haven, CT), Hua Xu (Department of Biomedical Informatics and Data Science, Yale School of Medicine, New Haven, CT), Vipina K. Keloth (Department of Biomedical Informatics and Data Science, Yale School of Medicine, New Haven, CT)
Abstract: Understanding how individuals with Parkinson's disease (PD) describe cognitive experiences in their daily lives can offer valuable insights into disease-related cognitive and emotional changes. However, extracting such information from unstructured patient narratives is challenging due to the subtle, overlapping nature of cognitive constructs. This study developed and evaluated natural language processing (NLP) models to automatically identify categories that reflect various cognitive processes from de-identified first-person narratives. Three model families, a Bio_ClinicalBERT-based span categorization model for nested entity recognition, a fine-tuned Meta-Llama-3-8B-Instruct model using QLoRA for instruction following, and GPT-4o mini evaluated under zero- and few-shot settings, were compared on their performance on extracting seven categories. Our findings indicated that model performance varied substantially across categories and model families. The fine-tuned Meta-Llama-3-8B-Instruct achieved the highest overall F1-scores (0.74 micro-average and 0.59 macro-average), particularly excelling in context-dependent categories such as thought and social interaction. Bio_ClinicalBERT exhibited high precision but low recall and performed comparable to Llama for some category types such as location and time but failed on other categories such as thought, emotion and social interaction. Compared to conventional information extraction tasks, this task presents a greater challenge due to the abstract and overlapping nature of narrative accounts of complex cognitive processes. Nonetheless, with continued refinement, these NLP systems hold promise for enabling low-burden, longitudinal monitoring of cognitive function and serving as a valuable complement to formal neuropsychological assessments in PD.
Authors: Farah Binta Haque, Md Yasin, Shishir Saha, Md Shoaib Akhter Rafi, Farig Sadeque
Abstract: Despite the growing progress in Natural Language Inference (NLI) research, resources for the Bengali language remain extremely limited. Existing Bengali NLI datasets exhibit several inconsistencies, including annotation errors, ambiguous sentence pairs, and inadequate linguistic diversity, which hinder effective model training and evaluation. To address these limitations, we introduce BNLI, a refined and linguistically curated Bengali NLI dataset designed to support robust language understanding and inference modeling. The dataset was constructed through a rigorous annotation pipeline emphasizing semantic clarity and balance across entailment, contradiction, and neutrality classes. We benchmarked BNLI using a suite of state-of-the-art transformer-based architectures, including multilingual and Bengali-specific models, to assess their ability to capture complex semantic relations in Bengali text. The experimental findings highlight the improved reliability and interpretability achieved with BNLI, establishing it as a strong foundation for advancing research in Bengali and other low-resource language inference tasks.
Authors: Yejin Yoon, Yuri Son, Namyoung So, Minseo Kim, Minsoo Cho, Chanhee Park, Seungshin Lee, Taeuk Kim
Abstract: Conversational agents have traditionally been developed for either task-oriented dialogue (TOD) or open-ended chitchat, with limited progress in unifying the two. Yet, real-world conversations naturally involve fluid transitions between these modes. To address this gap, we introduce TACT (TOD-And-Chitchat Transition), a dataset designed for transition-aware dialogue modeling that incorporates structurally diverse and integrated mode flows. TACT supports both user- and agent-driven mode switches, enabling robust modeling of complex conversational dynamics. To evaluate an agent's ability to initiate and recover from mode transitions, we propose two new metrics -- Switch and Recovery. Models trained on TACT outperform baselines in both intent detection and mode transition handling. Moreover, applying Direct Preference Optimization (DPO) to TACT-trained models yields additional gains, achieving 75.74\% joint mode-intent accuracy and a 70.1\% win rate against GPT-4o in human evaluation. These results demonstrate that pairing structurally diverse data with DPO enhances response quality and transition control, paving the way for more proactive and transition-aware conversational agents.
Authors: Fuyi Yang, Chenchen Ye, Mingyu Derek Ma, Yijia Xiao, Matthew Yang, Wei Wang
Abstract: Hypothesis generation in biomedical research has traditionally centered on uncovering hidden relationships within vast scientific literature, often using methods like Literature-Based Discovery (LBD). Despite progress, current approaches typically depend on single data types or predefined extraction patterns, which restricts the discovery of novel and complex connections. Recent advances in Large Language Model (LLM) agents show significant potential, with capabilities in information retrieval, reasoning, and generation. However, their application to biomedical hypothesis generation has been limited by the absence of standardized datasets and execution environments. To address this, we introduce BioVerge, a comprehensive benchmark, and BioVerge Agent, an LLM-based agent framework, to create a standardized environment for exploring biomedical hypothesis generation at the frontier of existing scientific knowledge. Our dataset includes structured and textual data derived from historical biomedical hypotheses and PubMed literature, organized to support exploration by LLM agents. BioVerge Agent utilizes a ReAct-based approach with distinct Generation and Evaluation modules that iteratively produce and self-assess hypothesis proposals. Through extensive experimentation, we uncover key insights: 1) different architectures of BioVerge Agent influence exploration diversity and reasoning strategies; 2) structured and textual information sources each provide unique, critical contexts that enhance hypothesis generation; and 3) self-evaluation significantly improves the novelty and relevance of proposed hypotheses.
Authors: Junichiro Niimi
Abstract: Large language models (LLMs) have been increasingly applied to a wide range of tasks, from natural language understanding to code generation. While they have also been used to assist in citation recommendation, the hallucination of non-existent papers remains a major issue. Building on prior studies, this study hypothesizes that an LLM's ability to correctly produce bibliographic records depends on whether the underlying knowledge is generated or memorized, with highly cited papers (i.e., more frequently appear in the pretraining corpus) showing lower hallucination rates. We therefore assume citation count as a proxy for training data redundancy (i.e., the frequency with which a given bibliographic record appears in the pretraining corpus) and investigate how citation frequency affects hallucinated references in LLM outputs. Using GPT-4.1, we generated and manually verified 100 citations across twenty computer-science domains, and measured factual consistency via cosine similarity between generated and authentic metadata. The results revealed that (i) citation count is strongly correlated with factual accuracy, (ii) bibliographic information becomes almost verbatim memorized beyond roughly 1,000 citations, and (iii) memory interference occurs when multiple highly cited papers share similar content. These findings indicate a threshold where generalization shifts into memorization, with highly cited papers being nearly verbatim retained in the model.
Authors: Yaxin Zhao, Yu Zhang
Abstract: Large language models (LLMs) have achieved impressive performance across a wide range of natural language processing tasks, yet they often produce hallucinated content that undermines factual reliability. To address this challenge, we introduce HalluClean, a lightweight and task-agnostic framework for detecting and correcting hallucinations in LLM-generated text. HalluClean adopts a reasoning-enhanced paradigm, explicitly decomposing the process into planning, execution, and revision stages to identify and refine unsupported claims. It employs minimal task-routing prompts to enable zero-shot generalization across diverse domains, without relying on external knowledge sources or supervised detectors. We conduct extensive evaluations on five representative tasks-question answering, dialogue, summarization, math word problems, and contradiction detection. Experimental results show that HalluClean significantly improves factual consistency and outperforms competitive baselines, demonstrating its potential to enhance the trustworthiness of LLM outputs in real-world applications.
Authors: Jingyu Liu, Xin Dong, Zhifan Ye, Rishabh Mehta, Yonggan Fu, Vartika Singh, Jan Kautz, Ce Zhang, Pavlo Molchanov
Abstract: Diffusion language models hold the promise of fast parallel generation, while autoregressive (AR) models typically excel in quality due to their causal structure aligning naturally with language modeling. This raises a fundamental question: can we achieve a synergy with high throughput, higher GPU utilization, and AR level quality? Existing methods fail to effectively balance these two aspects, either prioritizing AR using a weaker model for sequential drafting (speculative decoding), leading to lower drafting efficiency, or using some form of left-to-right (AR-like) decoding logic for diffusion, which still suffers from quality degradation and forfeits its potential parallelizability. We introduce TiDAR, a sequence-level hybrid architecture that drafts tokens (Thinking) in Diffusion and samples final outputs (Talking) AutoRegressively - all within a single forward pass using specially designed structured attention masks. This design exploits the free GPU compute density, achieving a strong balance between drafting and verification capacity. Moreover, TiDAR is designed to be serving-friendly (low overhead) as a standalone model. We extensively evaluate TiDAR against AR models, speculative decoding, and diffusion variants across generative and likelihood tasks at 1.5B and 8B scales. Thanks to the parallel drafting and sampling as well as exact KV cache support, TiDAR outperforms speculative decoding in measured throughput and surpasses diffusion models like Dream and Llada in both efficiency and quality. Most notably, TiDAR is the first architecture to close the quality gap with AR models while delivering 4.71x to 5.91x more tokens per second.
Authors: Longfei Zuo, Barbara Plank, Siyao Peng
Abstract: High-quality datasets are critical for training and evaluating reliable NLP models. In tasks like natural language inference (NLI), human label variation (HLV) arises when multiple labels are valid for the same instance, making it difficult to separate annotation errors from plausible variation. An earlier framework VARIERR (Weber-Genzel et al., 2024) asks multiple annotators to explain their label decisions in the first round and flag errors via validity judgments in the second round. However, conducting two rounds of manual annotation is costly and may limit the coverage of plausible labels or explanations. Our study proposes a new framework, EVADE, for generating and validating explanations to detect errors using large language models (LLMs). We perform a comprehensive analysis comparing human- and LLM-detected errors for NLI across distribution comparison, validation overlap, and impact on model fine-tuning. Our experiments demonstrate that LLM validation refines generated explanation distributions to more closely align with human annotations, and that removing LLM-detected errors from training data yields improvements in fine-tuning performance than removing errors identified by human annotators. This highlights the potential to scale error detection, reducing human effort while improving dataset quality under label variation.
Authors: Shengmin Piao, Sanghyun Park
Abstract: Recent advances in large reasoning models have been driven by reinforcement learning and test-time scaling, accompanied by growing interest in latent rather than purely textual reasoning. However, existing latent reasoning methods lack mechanisms to ensure stable evolution of latent representations and a systematic way to interleave implicit and explicit reasoning. We introduce SpiralThinker, a unified framework that performs iterative updates over latent representations, enabling extended implicit reasoning without generating additional tokens. A progressive alignment objective combined with structured annotations maintains coherence between latent and textual reasoning. Across mathematical, logical, and commonsense reasoning tasks, SpiralThinker achieves the best overall performance among latent reasoning approaches, consistently surpassing previous methods across all benchmarks. Detailed analyses reveal that both iteration and alignment are indispensable, the numbers of latent tokens and iterations exhibit dataset-specific optima, and appropriate alignment proves critical for an effective iterative process. Overall, SpiralThinker bridges iterative computation and latent reasoning, demonstrating that aligned iterative updates can reliably steer reasoning in the latent space.
Authors: Zhouxing Tan, Ruochong Xiong, Yulong Wan, Jinlong Ma, Hanlin Xue, Qichun Deng, Haifeng Jing, Zhengtong Zhang, Depei Liu, Shiyuan Luo, Junfei Liu
Abstract: Emotional support is a core capability in human-AI interaction, with applications including psychological counseling, role play, and companionship. However, existing evaluations of large language models (LLMs) often rely on short, static dialogues and fail to capture the dynamic and long-term nature of emotional support. To overcome this limitation, we shift from snapshot-based evaluation to trajectory-based assessment, adopting a user-centered perspective that evaluates models based on their ability to improve and stabilize user emotional states over time. Our framework constructs a large-scale benchmark consisting of 328 emotional contexts and 1,152 disturbance events, simulating realistic emotional shifts under evolving dialogue scenarios. To encourage psychologically grounded responses, we constrain model outputs using validated emotion regulation strategies such as situation selection and cognitive reappraisal. User emotional trajectories are modeled as a first-order Markov process, and we apply causally-adjusted emotion estimation to obtain unbiased emotional state tracking. Based on this framework, we introduce three trajectory-level metrics: Baseline Emotional Level (BEL), Emotional Trajectory Volatility (ETV), and Emotional Centroid Position (ECP). These metrics collectively capture user emotional dynamics over time and support comprehensive evaluation of long-term emotional support performance of LLMs. Extensive evaluations across a diverse set of LLMs reveal significant disparities in emotional support capabilities and provide actionable insights for model development.
Authors: Sam Bayless, Stefano Buliani, Darion Cassel, Byron Cook, Duncan Clough, R\'emi Delmas, Nafi Diallo, Ferhat Erata, Nick Feng, Dimitra Giannakopoulou, Aman Goel, Aditya Gokhale, Joe Hendrix, Marc Hudak, Dejan Jovanovi\'c, Andrew M. Kent, Benjamin Kiesl-Reiter, Jeffrey J. Kuna, Nadia Labai, Joseph Lilien, Divya Raghunathan, Zvonimir Rakamari\'c, Niloofar Razavi, Michael Tautschnig, Ali Torkamani, Nathaniel Weir, Michael W. Whalen, Jianan Yao
Abstract: Large Language Models perform well at natural language interpretation and reasoning, but their inherent stochasticity limits their adoption in regulated industries like finance and healthcare that operate under strict policies. To address this limitation, we present a two-stage neurosymbolic framework that (1) uses LLMs with optional human guidance to formalize natural language policies, allowing fine-grained control of the formalization process, and (2) uses inference-time autoformalization to validate logical correctness of natural language statements against those policies. When correctness is paramount, we perform multiple redundant formalization steps at inference time, cross checking the formalizations for semantic equivalence. Our benchmarks demonstrate that our approach exceeds 99% soundness, indicating a near-zero false positive rate in identifying logical validity. Our approach produces auditable logical artifacts that substantiate the verification outcomes and can be used to improve the original text.
Authors: Gailun Zeng, Ziyang Luo, Hongzhan Lin, Yuchen Tian, Kaixin Li, Ziyang Gong, Jianxiong Guo, Jing Ma
Abstract: The ability of critique is vital for models to self-improve and serve as reliable AI assistants. While extensively studied in language-only settings, multimodal critique of Large Multimodal Models (LMMs) remains underexplored despite their growing capabilities in tasks like captioning and visual reasoning. In this work, we introduce MM-CRITIC, a holistic benchmark for evaluating the critique ability of LMMs across multiple dimensions: basic, correction, and comparison. Covering 8 main task types and over 500 tasks, MM-CRITIC collects responses from various LMMs with different model sizes and is composed of 4471 samples. To enhance the evaluation reliability, we integrate expert-informed ground answers into scoring rubrics that guide GPT-4o in annotating responses and generating reference critiques, which serve as anchors for trustworthy judgments. Extensive experiments validate the effectiveness of MM-CRITIC and provide a comprehensive assessment of leading LMMs' critique capabilities under multiple dimensions. Further analysis reveals some key insights, including the correlation between response quality and critique, and varying critique difficulty across evaluation dimensions. Our code is available at https://github.com/MichealZeng0420/MM-Critic.
Authors: Chao Wang, Yuqing Cai, Renzeng Duojie, Jin Zhang, Yutong Liu, Nyima Tashi
Abstract: In this work, we propose a streaming speech recognition framework for Amdo Tibetan, built upon a hybrid CTC/Atten-tion architecture with a context-aware dynamic chunking mechanism. The proposed strategy adaptively adjusts chunk widths based on encoding states, enabling flexible receptive fields, cross-chunk information exchange, and robust adaptation to varying speaking rates, thereby alleviating the context truncation problem of fixed-chunk methods. To further capture the linguistic characteristics of Tibetan, we construct a lexicon grounded in its orthographic principles, providing linguistically motivated modeling units. During decoding, an external language model is integrated to enhance semantic consistency and improve recognition of long sentences. Experimental results show that the proposed framework achieves a word error rate (WER) of 6.23% on the test set, yielding a 48.15% relative improvement over the fixed-chunk baseline, while significantly reducing recognition latency and maintaining performance close to global decoding.
Authors: Wenda Wei, Yu-An Liu, Ruqing Zhang, Jiafeng Guo, Lixin Su, Shuaiqiang Wang, Dawei Yin, Maarten de Rijke, Xueqi Cheng
Abstract: Retrieval-augmented generation (RAG) has proven to be effective in mitigating hallucinations in large language models, yet its effectiveness remains limited in complex, multi-step reasoning scenarios.Recent efforts have incorporated search-based interactions into RAG, enabling iterative reasoning with real-time retrieval. Most approaches rely on outcome-based supervision, offering no explicit guidance for intermediate steps. This often leads to reward hacking and degraded response quality. We propose Bi-RAR, a novel retrieval-augmented reasoning framework that evaluates each intermediate step jointly in both forward and backward directions. To assess the information completeness of each step, we introduce a bidirectional information distance grounded in Kolmogorov complexity, approximated via language model generation probabilities. This quantification measures both how far the current reasoning is from the answer and how well it addresses the question. To optimize reasoning under these bidirectional signals, we adopt a multi-objective reinforcement learning framework with a cascading reward structure that emphasizes early trajectory alignment. Empirical results on seven question answering benchmarks demonstrate that Bi-RAR surpasses previous methods and enables efficient interaction and reasoning with the search engine during training and inference.
Authors: Ritsu Sakabe, Hwichan Kim, Tosho Hirasawa, Mamoru Komachi
Abstract: Computational humor is a frontier for creating advanced and engaging natural language processing (NLP) applications, such as sophisticated dialogue systems. While previous studies have benchmarked the humor capabilities of Large Language Models (LLMs), they have often relied on single-dimensional evaluations, such as judging whether something is simply ``funny.'' This paper argues that a multifaceted understanding of humor is necessary and addresses this gap by systematically evaluating LLMs through the lens of Oogiri, a form of Japanese improvisational comedy games. To achieve this, we expanded upon existing Oogiri datasets with data from new sources and then augmented the collection with Oogiri responses generated by LLMs. We then manually annotated this expanded collection with 5-point absolute ratings across six dimensions: Novelty, Clarity, Relevance, Intelligence, Empathy, and Overall Funniness. Using this dataset, we assessed the capabilities of state-of-the-art LLMs on two core tasks: their ability to generate creative Oogiri responses and their ability to evaluate the funniness of responses using a six-dimensional evaluation. Our results show that while LLMs can generate responses at a level between low- and mid-tier human performance, they exhibit a notable lack of Empathy. This deficit in Empathy helps explain their failure to replicate human humor assessment. Correlation analyses of human and model evaluation data further reveal a fundamental divergence in evaluation criteria: LLMs prioritize Novelty, whereas humans prioritize Empathy. We release our annotated corpus to the community to pave the way for the development of more emotionally intelligent and sophisticated conversational agents.
Authors: Jieun Han, Daniel Lee, Haneul Yoo, Jinsung Yoon, Junyeong Park, Suin Kim, So-Yeon Ahn, Alice Oh
Abstract: Personalized learning has gained attention in English as a Foreign Language (EFL) education, where engagement and motivation play crucial roles in reading comprehension. We propose a novel approach to generating personalized English reading comprehension tests tailored to students' interests. We develop a structured content transcreation pipeline using OpenAI's gpt-4o, where we start with the RACE-C dataset, and generate new passages and multiple-choice reading comprehension questions that are linguistically similar to the original passages but semantically aligned with individual learners' interests. Our methodology integrates topic extraction, question classification based on Bloom's taxonomy, linguistic feature analysis, and content transcreation to enhance student engagement. We conduct a controlled experiment with EFL learners in South Korea to examine the impact of interest-aligned reading materials on comprehension and motivation. Our results show students learning with personalized reading passages demonstrate improved comprehension and motivation retention compared to those learning with non-personalized materials.
Authors: Jing Xiong, Liyang Fan, Hui Shen, Zunhai Su, Min Yang, Lingpeng Kong, Ngai Wong
Abstract: Rotary Position Embedding (RoPE) in Transformer models has inherent limits that weaken length extrapolation. We reinterpret the attention map with positional encoding as a noisy feature map, and propose Denoising Positional Encoding (DoPE), a training-free method based on truncated matrix entropy to detect outlier frequency bands in the feature map. Leveraging the noise characteristics of the feature map, we further reparameterize it with a parameter-free Gaussian distribution to achieve robust extrapolation. Our method theoretically reveals the underlying cause of the attention sink phenomenon and its connection to truncated matrix entropy. Experiments on needle-in-a-haystack and many-shot in-context learning tasks demonstrate that DoPE significantly improves retrieval accuracy and reasoning stability across extended contexts (up to 64K tokens). The results show that the denoising strategy for positional embeddings effectively mitigates attention sinks and restores balanced attention patterns, providing a simple yet powerful solution for improving length generalization. Our project page is Project: https://The-physical-picture-of-LLMs.github.io
Authors: Kangning Zhang, Wenxiang Jiao, Kounianhua Du, Yuan Lu, Weiwen Liu, Weinan Zhang, Lei Zhang, Yong Yu
Abstract: Augmenting Large Language Models (LLMs) with external tools enables them to execute complex, multi-step tasks. However, tool learning is hampered by the static synthetic data pipelines where data generation and model training are executed as two separate, non-interactive processes. This approach fails to adaptively focus on a model's specific weaknesses and allows noisy labels to persist, degrading training efficiency. We introduce LoopTool, a fully automated, model-aware data evolution framework that closes this loop by tightly integrating data synthesis and model training. LoopTool iteratively refines both the data and the model through three synergistic modules: (1) Greedy Capability Probing (GCP) diagnoses the model's mastered and failed capabilities; (2) Judgement-Guided Label Verification (JGLV) uses an open-source judge model to find and correct annotation errors, progressively purifying the dataset; and (3) Error-Driven Data Expansion (EDDE) generates new, challenging samples based on identified failures. This closed-loop process operates within a cost-effective, open-source ecosystem, eliminating dependence on expensive closed-source APIs. Experiments show that our 8B model trained with LoopTool significantly surpasses its 32B data generator and achieves new state-of-the-art results on the BFCL-v3 and ACEBench benchmarks for its scale. Our work demonstrates that closed-loop, self-refining data pipelines can dramatically enhance the tool-use capabilities of LLMs.
Authors: Daiki Shirafuji, Koji Tanaka, Tatsuhiko Saito
Abstract: Recently, Large Language Models (LLMs) are gaining increased attention in the domain of Table Question Answering (TQA), particularly for extracting information from tables in documents. However, directly entering entire tables as long text into LLMs often leads to incorrect answers because most LLMs cannot inherently capture complex table structures. In this paper, we propose a cell extraction method for TQA without manual identification, even for complex table headers. Our approach estimates table headers by computing similarities between a given question and individual cells via a hybrid retrieval mechanism that integrates a language model and TF-IDF. We then select as the answer the cells at the intersection of the most relevant row and column. Furthermore, the language model is trained using contrastive learning on a small dataset of question-header pairs to enhance performance. We evaluated our approach in the TQA dataset from the U4 shared task at NTCIR-18. The experimental results show that our pipeline achieves an accuracy of 74.6\%, outperforming existing LLMs such as GPT-4o mini~(63.9\%). In the future, although we used traditional encoder models for retrieval in this study, we plan to incorporate more efficient text-search models to improve performance and narrow the gap with human evaluation results.
Authors: Amal Sunny, Advay Gupta, Vishnu Sreekumar
Abstract: Recent work has proposed using Large Language Models (LLMs) to quantify narrative flow through a measure called sequentiality, which combines topic and contextual terms. A recent critique argued that the original results were confounded by how topics were selected for the topic-based component, and noted that the metric had not been validated against ground-truth measures of flow. That work proposed using only the contextual term as a more conceptually valid and interpretable alternative. In this paper, we empirically validate that proposal. Using two essay datasets with human-annotated trait scores, ASAP++ and ELLIPSE, we show that the contextual version of sequentiality aligns more closely with human assessments of discourse-level traits such as Organization and Cohesion. While zero-shot prompted LLMs predict trait scores more accurately than the contextual measure alone, the contextual measure adds more predictive value than both the topic-only and original sequentiality formulations when combined with standard linguistic features. Notably, this combination also outperforms the zero-shot LLM predictions, highlighting the value of explicitly modeling sentence-to-sentence flow. Our findings support the use of context-based sequentiality as a validated, interpretable, and complementary feature for automated essay scoring and related NLP tasks.
Authors: Francois Meyer, Jan Buys
Abstract: Subword segmentation is typically applied in preprocessing and stays fixed during training. Alternatively, it can be learned during training to optimise the training objective. In this paper we study the learning dynamics of subword segmentation: if a language model can dynamically optimise tokenisation, how do its subwords evolve during pretraining and finetuning? To explore this, we extend the subword segmental language model (SSLM), a framework for learning subwords during training, to support pretraining and finetuning. We train models for three typologically diverse languages to study learning dynamics across the morphological spectrum: Isi-Xhosa is conjunctive (long word forms composed of many morphemes), Setswana is disjunctive (morphemes written as separate words), and English represents a typological middle ground. We analyse subword dynamics from a linguistic perspective, tracking morphology, productivity, and fertility. We identify four stages of subword learning, with the morphologically complex isi-Xhosa exhibiting greater instability. During finetuning, subword boundaries shift to become finer-grained. Lastly, we show that learnable subwords offers a promising approach to improve text generation and cross-lingual transfer for low-resource, morphologically complex languages.
Authors: Akseli Reunamo, Laura-Maria Peltonen, Hans Moen, Sampo Pyysalo
Abstract: This paper reports on pretraining ModernBERT encoder models in six different sizes, ranging from 51M to 475M parameters, with a focus on limited multilingualism, emphasizing languages relevant to Finland. Our models are competitive with, or superior to, existing multilingual models. They outperform monolingual models on tasks that require a context longer than 512 tokens. We present empirical results on using different data in the final stage of training. The code and models are publicly released.
Authors: Jiarui Liu, Kaustubh Dhole, Yingheng Wang, Haoyang Wen, Sarah Zhang, Haitao Mao, Gaotang Li, Neeraj Varshney, Jingguo Liu, Xiaoman Pan
Abstract: Reinforcement learning with verifiable rewards (RLVR) has recently emerged as a promising framework for aligning language models with complex reasoning objectives. However, most existing methods optimize only for final task outcomes, leaving models vulnerable to collapse when negative rewards dominate early training. This challenge is especially pronounced in honesty alignment, where models must not only solve answerable queries but also identify when conclusions cannot be drawn from the given premises. Deductive reasoning provides an ideal testbed because it isolates reasoning capability from reliance on external factual knowledge. To investigate honesty alignment, we curate two multi-step deductive reasoning datasets from graph structures, one for linear algebra and one for logical inference, and introduce unanswerable cases by randomly perturbing an edge in half of the instances. We find that GRPO, with or without supervised fine tuning initialization, struggles on these tasks. Through extensive experiments across three models, we evaluate stabilization strategies and show that curriculum learning provides some benefit but requires carefully designed in distribution datasets with controllable difficulty. To address these limitations, we propose Anchor, a reinforcement learning method that injects ground truth trajectories into rollouts, preventing early training collapse. Our results demonstrate that this method stabilizes learning and significantly improves the overall reasoning performance, underscoring the importance of training dynamics for enabling reliable deductive reasoning in aligned language models.
Authors: Xuanchen Li, Chenrui Cui, Tianrui Wang, Meng Ge, Zikang Huang, Jin Li, Yizhou Peng, Longbiao Wang, Jianwu Dang, Nyima Tashi
Abstract: Speech Large Language Models (SpeechLLMs) have achieved breakthroughs in multilingual speech-to-text translation (S2TT). However, existing approaches often overlook semantic commonalities across source languages, leading to biased translation performance. In this work, we propose \textbf{POTSA} (Parallel Optimal Transport for Speech Alignment), a new framework based on cross-lingual parallel speech pairs and Optimal Transport (OT), designed to bridge high- and low-resource translation gaps. First, we introduce a Bias Compensation module to coarsely align initial speech representations across languages. Second, we impose token-level OT constraints on a Q-Former using parallel speech pairs to establish fine-grained consistency of representations. Then, we apply a layer scheduling strategy to focus OT constraints on the most semantically beneficial layers. Experiments on the FLEURS dataset show that our method achieves SOTA performance, with +0.93 BLEU on average over five common languages and +5.05 BLEU on zero-shot languages, using only 10 hours of parallel speech per source language.
Authors: Yu Li, Zhe Yang, Yi Huang, Xin Liu, Guilin Qi
Abstract: Recent advancements in large language models (LLMs) have demonstrated remarkable text generation capabilities. However, controlling specific attributes of generated text remains challenging without architectural modifications or extensive fine-tuning. Current methods typically toggle a single, basic attribute but struggle with precise multi-attribute control. In scenarios where attribute requirements conflict, existing methods lack coordination mechanisms, causing interference between desired attributes. Furthermore, these methods fail to incorporate iterative optimization processes in the controlled generation pipeline. To address these limitations, we propose Conflict-aware, Composite, and Collaborative Controlled Text Generation (C$^3$TG), a two-phase framework for fine-grained, multi-dimensional text attribute control. During generation, C$^3$TG selectively pairs the LLM with the required attribute classifiers from the 17 available dimensions and employs weighted KL-divergence to adjust token probabilities. The optimization phase then leverages an energy function combining classifier scores and penalty terms to resolve attribute conflicts through iterative feedback, enabling precise control over multiple dimensions simultaneously while preserving natural text flow. Experiments show that C$^3$TG significantly outperforms baselines across multiple metrics including attribute accuracy, linguistic fluency, and output diversity, while simultaneously reducing toxicity. These results establish C$^3$TG as an effective and flexible solution for multi-dimensional text attribute control that requires no costly model modifications.
Authors: John Joon Young Chung, Vishakh Padmakumar, Melissa Roemmele, Yi Wang, Yuqian Sun, Tiffany Wang, Shm Garanganao Almeda, Brett A. Halperin, Yuwen Lu, Max Kreminski
Abstract: People have different creative writing preferences, and large language models (LLMs) for these tasks can benefit from adapting to each user's preferences. However, these models are often trained over a dataset that considers varying personal tastes as a monolith. To facilitate developing personalized creative writing LLMs, we introduce LiteraryTaste, a dataset of reading preferences from 60 people, where each person: 1) self-reported their reading habits and tastes (stated preference), and 2) annotated their preferences over 100 pairs of short creative writing texts (revealed preference). With our dataset, we found that: 1) people diverge on creative writing preferences, 2) finetuning a transformer encoder could achieve 75.8% and 67.7% accuracy when modeling personal and collective revealed preferences, and 3) stated preferences had limited utility in modeling revealed preferences. With an LLM-driven interpretability pipeline, we analyzed how people's preferences vary. We hope our work serves as a cornerstone for personalizing creative writing technologies.
Authors: Marry Kong, Rina Buoy, Sovisal Chenda, Nguonly Taing
Abstract: Khmer polarity classification is a fundamental natural language processing task that assigns a positive, negative, or neutral label to a given Khmer text input. Existing Khmer models typically predict the label without explaining the rationale behind the prediction. This paper proposes an explainable Khmer polarity classifier by fine-tuning an instruction-based reasoning Qwen-3 model. The notion of explainability in this paper is limited to self-explanations, which the model uses to rationalize its predictions. Experimental results show that the fine-tuned model not only predicts labels accurately but also provides reasoning by identifying polarity-related keywords or phrases to support its predictions. In addition, we contribute a new Khmer polarity dataset consisting of short- to medium-length casual, romanized, and mixed-code Khmer expressions. This dataset was constructed using both heuristic rules and human curation and is publicly available through a gated Hugging Face repository (rinabuoy/khmerpolarity_nonreasoning). The fine-tuned Qwen-3 models are also made available in the same Hugging Face account.
Authors: Zhuoqun Huang, Neil G. Marchant, Olga Ohrimenko, Benjamin I. P. Rubinstein
Abstract: We consider the problem of certified robustness for sequence classification against edit distance perturbations. Naturally occurring inputs of varying lengths (e.g., sentences in natural language processing tasks) present a challenge to current methods that employ fixed-rate deletion mechanisms and lead to suboptimal performance. To this end, we introduce AdaptDel methods with adaptable deletion rates that dynamically adjust based on input properties. We extend the theoretical framework of randomized smoothing to variable-rate deletion, ensuring sound certification with respect to edit distance. We achieve strong empirical results in natural language tasks, observing up to 30 orders of magnitude improvement to median cardinality of the certified region, over state-of-the-art certifications.
Authors: Arka Mukherjee, Shreya Ghosh
Abstract: Contemporary vision-language models (VLMs) perform well on existing multimodal reasoning benchmarks (78-85\% accuracy on MMMU, MathVista). Yet, these results fail to sufficiently distinguish true scientific reasoning articulation capabilities from pattern-matching. To address this gap, we introduce \textbf{mmJEE-Eval}, a multimodal bilingual (English and Hindi) benchmark comprising 1,460 questions from India's JEE Advanced examination (2019-2025) spanning pre-college Physics, Chemistry, and Mathematics domains. Our evaluation of 17 state-of-the-art models reveals that while frontier VLMs (GPT-5, Gemini 2.5 Pro/Flash) achieve 77-84\% accuracy on held-out 2025 questions, open-source models plateau at 37-45\% despite scaling to 400B parameters, a significant difference not observed on existing benchmarks. While closed frontiers from Google and OpenAI show high problem-solving accuracies (up to 100\% pass@3 scores), they fully collapse when the reasoning load is increased meta-cognitively (GPT-5 fixes just 5.2\% errors). Systematic ablations show mmJEE-Eval's difficulty stems from complexity and reasoning depth rather than memorization. Effectively, our benchmark segregates superior training and reasoning methodologies where alternatives fail. We publicly release our code and data: https://mmjee-eval.github.io
Authors: Shiyu Ji, Yixuan Wang, Yijun Liu, Qingfu Zhu, Wanxiang Che
Abstract: Test-time scaling improves the inference performance of Large Language Models (LLMs) but also incurs substantial computational costs. Although recent studies have reduced token consumption through dynamic self-consistency, they remain constrained by the high latency of sequential requests. In this paper, we propose SeerSC, a dynamic self-consistency framework that simultaneously improves token efficiency and latency by integrating System 1 and System 2 reasoning. Specifically, we utilize the rapid System 1 to compute the answer entropy for given queries. This score is then used to evaluate the potential of samples for scaling, enabling dynamic self-consistency under System 2. Benefiting from the advance and accurate estimation provided by System 1, the proposed method can reduce token usage while simultaneously achieving a significant decrease in latency through parallel generation. It outperforms existing methods, achieving up to a 47% reduction in token consumption and a 43% reduction in inference latency without significant performance loss.
Authors: Martin Vejvar, Yasutaka Fujimoto
Abstract: We present Spider4SSC dataset and S2CLite parsing tool. S2CLite is a lightweight, ontology-agnostic parser that translates SPARQL queries into Cypher queries, enabling both in-situ and large-scale SPARQL to Cypher translation. Unlike existing solutions, S2CLite is purely rule-based (inspired by traditional programming language compilers) and operates without requiring an RDF graph or external tools. Experiments conducted on the BSBM42 and Spider4SPARQL datasets show that S2CLite significantly reduces query parsing errors, achieving a total parsing accuracy of 77.8% on Spider4SPARQL compared to 44.2% by the state-of-the-art S2CTrans. Furthermore, S2CLite achieved a 96.6\% execution accuracy on the intersecting subset of queries parsed by both parsers, outperforming S2CTrans by 7.3%. We further use S2CLite to parse Spider4SPARQL queries to Cypher and generate Spider4SSC, a unified Text-to-Query language (SQL, SPARQL, Cypher) dataset with 4525 unique questions and 3 equivalent sets of 2581 matching queries (SQL, SPARQL and Cypher). We open-source S2CLite for further development on GitHub (github.com/vejvarm/S2CLite) and provide the clean Spider4SSC dataset for download.
Authors: Rhitabrat Pokharel, Ameeta Agrawal
Abstract: The use of large language models (LLMs) for evaluating outputs is becoming an increasingly effective and scalable approach. However, it remains uncertain whether this capability extends beyond task-specific evaluations to more general assessments of text quality, particularly in multilingual contexts. In this study, we introduce, MTQ-Eval, a novel framework for multilingual text quality evaluation that learns from examples of both high- and low-quality texts, adjusting its internal representations. To develop MTQ-Eval, we first automatically generate text quality preference data and then use it to train open-source base LLMs to align with ratings of high- and low-quality text. Our comprehensive evaluation across 115 languages demonstrates the improved performance of the proposed model. Upon further analysis, we find that this enhanced evaluation capability also leads to notable improvements in downstream tasks.
Authors: Hossein A. Rahmani, Satyapriya Krishna, Xi Wang, Mohammadmehdi Naghiaei, Emine Yilmaz
Abstract: Large language models have recently demonstrated remarkable abilities to self-correct their responses through iterative refinement, often referred to as self-consistency or self-reflection. However, the dynamics of this self-correction mechanism may differ substantially depending on whether the model is tasked with open-ended text generation or with selecting the most appropriate response from multiple predefined options. In this paper, we conduct a systematic investigation of these two paradigms by comparing performance trends and error-correction behaviors across various natural language understanding and reasoning tasks, covering language models of different scales and families. Our experimental results reveal distinct patterns of improvement and failure modes: \textit{While open-ended generation often benefits from the flexibility of re-interpretation and compositional refinement, multiple-choice selection can leverage clearer solution boundaries but may be limited by the provided options}. This contrast also reflects the dual demands faced by emerging agentic LLM applications: effective agents must not only generate and refine open-ended plans or explanations, but also make reliable discrete choices when operating within constrained action spaces. Our findings, therefore, highlight that the design of self-correction mechanisms should take into account the interaction between task structure and output space, with implications for both knowledge-intensive reasoning and decision-oriented applications of LLMs.
Authors: Ruibo Deng, Duanyu Feng, Wenqiang Lei
Abstract: Offline preference optimization offers a simpler and more stable alternative to RLHF for aligning language models. However, their effectiveness is critically dependent on ranking accuracy, a metric where further gains are highly impactful. This limitation arises from a fundamental problem that we identify and formalize as the Overfitting-Underfitting Dilemma: current margin designs cause models to apply excessive, wasteful gradients to correctly ranked samples (overfitting) while providing insufficient corrective signals for misranked ones (underfitting). To resolve this dilemma, we propose Adaptive Margin-attached Preference Optimization (AMaPO), a simple yet principled algorithm. AMaPO employs an instance-wise adaptive margin, refined by Z-normalization and exponential scaling, which dynamically reallocates learning effort by amplifying gradients for misranked samples and suppressing them for correct ones. Extensive experiments on widely used benchmarks demonstrate that AMaPO not only achieves better ranking accuracy and superior downstream alignment performance, but targeted analysis also confirms that it successfully mitigates the core overfitting and underfitting issues.
Authors: Lukas Arana, Julen Etxaniz, Ander Salaberria, Gorka Azkune
Abstract: Current Multimodal Large Language Models exhibit very strong performance for several demanding tasks. While commercial MLLMs deliver acceptable performance in low-resource languages, comparable results remain unattained within the open science community. In this paper, we aim to develop a strong MLLM for a low-resource language, namely Basque. For that purpose, we develop our own training and evaluation image-text datasets. Using two different Large Language Models as backbones, the Llama-3.1-Instruct model and a Basque-adapted variant called Latxa, we explore several data mixtures for training. We show that: i) low ratios of Basque multimodal data (around 20%) are already enough to obtain solid results on Basque benchmarks, and ii) contrary to expected, a Basque instructed backbone LLM is not required to obtain a strong MLLM in Basque. Our results pave the way to develop MLLMs for other low-resource languages by openly releasing our resources.
Authors: Bichen Wang, Yixin Sun, Junzhe Wang, Hao Yang, Xing Fu, Yanyan Zhao, Si Wei, Shijin Wang, Bing Qin
Abstract: The mismatch between the growing demand for psychological counseling and the limited availability of services has motivated research into the application of Large Language Models (LLMs) in this domain. Consequently, there is a need for a robust and unified benchmark to assess the counseling competence of various LLMs. Existing works, however, are limited by unprofessional client simulation, static question-and-answer evaluation formats, and unidimensional metrics. These limitations hinder their effectiveness in assessing a model's comprehensive ability to handle diverse and complex clients. To address this gap, we introduce \textbf{CARE-Bench}, a dynamic and interactive automated benchmark. It is built upon diverse client profiles derived from real-world counseling cases and simulated according to expert guidelines. CARE-Bench provides a multidimensional performance evaluation grounded in established psychological scales. Using CARE-Bench, we evaluate several general-purpose LLMs and specialized counseling models, revealing their current limitations. In collaboration with psychologists, we conduct a detailed analysis of the reasons for LLMs' failures when interacting with clients of different types, which provides directions for developing more comprehensive, universal, and effective counseling models.
Authors: Wolfgang Otto, Lu Gan, Sharmila Upadhyaya, Saurav Karmakar, Stefan Dietze
Abstract: Research in Machine Learning (ML) and AI evolves rapidly. Information Extraction (IE) from scientific publications enables to identify information about research concepts and resources on a large scale and therefore is a pathway to improve understanding and reproducibility of ML-related research. To extract and connect fine-grained information in ML-related research, e.g. method training and data usage, we introduce GSAP-ERE. It is a manually curated fine-grained dataset with 10 entity types and 18 semantically categorized relation types, containing mentions of 63K entities and 35K relations from the full text of 100 ML publications. We show that our dataset enables fine-tuned models to automatically extract information relevant for downstream tasks ranging from knowledge graph (KG) construction, to monitoring the computational reproducibility of AI research at scale. Additionally, we use our dataset as a test suite to explore prompting strategies for IE using Large Language Models (LLM). We observe that the performance of state-of-the-art LLM prompting methods is largely outperformed by our best fine-tuned baseline model (NER: 80.6%, RE: 54.0% for the fine-tuned model vs. NER: 44.4%, RE: 10.1% for the LLM). This disparity of performance between supervised models and unsupervised usage of LLMs suggests datasets like GSAP-ERE are needed to advance research in the domain of scholarly information extraction.
Authors: Triet M. Le, Arjun Chandra, C. Anton Rytting, Valerie P. Karuzis, Vladimir Rife, William A. Simpson
Abstract: Predicting an individual's personalities from their generated texts is a challenging task, especially when the text volume is large. In this paper, we introduce a straightforward yet effective novel strategy called targeted preselection of texts (TPoT). This method semantically filters the texts as input to a deep learning model, specifically designed to predict a Big Five personality trait, facet, or item, referred to as the BIG5-TPoT model. By selecting texts that are semantically relevant to a particular trait, facet, or item, this strategy not only addresses the issue of input text limits in large language models but also improves the Mean Absolute Error and accuracy metrics in predictions for the Stream of Consciousness Essays dataset.
Authors: R\'emi Cardon, A. Seza Do\u{g}ru\"oz
Abstract: Readability is a key concept in the current era of abundant written information. To help making texts more readable and make information more accessible to everyone, a line of researched aims at making texts accessible for their target audience: automatic text simplification (ATS). Lately, there have been studies on the correlations between automatic evaluation metrics in ATS and human judgment. However, the correlations between those two aspects and commonly available readability measures (such as readability formulas or linguistic features) have not been the focus of as much attention. In this work, we investigate the place of readability measures in ATS by complementing the existing studies on evaluation metrics and human judgment, on English. We first discuss the relationship between ATS and research in readability, then we report a study on correlations between readability measures and human judgment, and between readability measures and ATS evaluation metrics. We identify that in general, readability measures do not correlate well with automatic metrics and human judgment. We argue that as the three different angles from which simplification can be assessed tend to exhibit rather low correlations with one another, there is a need for a clear definition of the construct in ATS.
Authors: Mohamed Elaraby, Jyoti Prakash Maheswari
Abstract: Large Language Models (LLMs) with extended context windows promise direct reasoning over long documents, reducing the need for chunking or retrieval. Constructing annotated resources for training and evaluation, however, remains costly. Synthetic data offers a scalable alternative, and we introduce SynClaimEval, a framework for evaluating synthetic data utility in long-context claim verification -- a task central to hallucination detection and fact-checking. Our framework examines three dimensions: (i) input characteristics, by varying context length and testing generalization to out-of-domain benchmarks; (ii) synthesis logic, by controlling claim complexity and error type variation; and (iii) explanation quality, measuring the degree to which model explanations provide evidence consistent with predictions. Experiments across benchmarks show that long-context synthesis can improve verification in base instruction-tuned models, particularly when augmenting existing human-written datasets. Moreover, synthesis enhances explanation quality, even when verification scores do not improve, underscoring its potential to strengthen both performance and explainability.
Authors: Shadaab Ghani, Anne H{\aa}kansson, Oleksii Pasichnyi, Hossein Shahrokni
Abstract: Housing cooperative is a common type of multifamily building ownership in Sweden. Although this ownership structure grants decision-making autonomy, it places a burden of responsibility on cooperative's board members. Most board members lack the resources or expertise to manage properties and their energy consumption. This ignorance presents a unique challenge, especially given the EU directives that prohibit buildings rated as energy classes F and G by 2033. Conversational agents (CAs) enable human-like interactions with computer systems, facilitating human-computer interaction across various domains. In our case, CAs can be implemented to support cooperative members in making informed energy retrofitting and usage decisions. This paper introduces a Conversational agent system, called SPARA, designed to advise cooperatives on energy efficiency. SPARA functions as an energy efficiency advisor by leveraging the Retrieval-Augmented Generation (RAG) framework with a Language Model(LM). The LM generates targeted recommendations based on a knowledge base composed of email communications between professional energy advisors and cooperatives' representatives in Stockholm. The preliminary results indicate that SPARA can provide energy efficiency advice with precision 80\%, comparable to that of municipal energy efficiency (EE) experts. A pilot implementation is currently underway, where municipal EE experts are evaluating SPARA performance based on questions posed to EE experts by BRF members. Our findings suggest that LMs can significantly improve outreach by supporting stakeholders in their energy transition. For future work, more research is needed to evaluate this technology, particularly limitations to the stability and trustworthiness of its energy efficiency advice.
Authors: Kaleem Ullah Qasim, Jiashu Zhang
Abstract: Recursive reasoning models achieve remarkable performance on complex reasoning tasks through iterative refinement, enabling tiny networks to match large language models thousands of times their size. However, training remains computationally expensive, prior work reporting approximately 36 GPU-hours per dataset, limiting broader adoption and research. We propose CGAR, a novel training methodology that applies curriculum learning to architectural depth rather than traditional data ordering. CGAR introduces two synergistic components: Progressive Depth Curriculum dynamically adjusts recursion depth from shallow to deep configurations during training, preventing early overfitting while reducing computational cost, and Hierarchical Supervision Weighting applies exponentially decaying importance to supervision steps, aligning loss weighting with observed gradient magnitude decay. On Sudoku-Extreme with 423,168 test puzzles, CGAR achieves 1.71x training speedup (10.93 to 6.38 hours, 42% cost reduction) with only 0.63% accuracy drop (86.65% to 86.02%). Systematic ablations reveal Progressive Depth Curriculum alone achieves 2.26x speedup with 85.47% accuracy, demonstrating a rare Pareto improvement where architectural curriculum simultaneously enhances training efficiency and solution quality. CGAR-trained models exhibit superior inference efficiency with 100% halting accuracy and 11% fewer reasoning steps. Our work demonstrates that principled curriculum on architectural depth enables efficient training of recursive reasoning models on modest hardware. Code and models: https://github.com/Kaleemullahqasim/CGAR and https://huggingface.co/Kaleemullah/trm-cgar-sudoku
URLs: https://github.com/Kaleemullahqasim/CGAR, https://huggingface.co/Kaleemullah/trm-cgar-sudoku
Authors: Jill Walker Rettberg
Abstract: This paper analyses AI-generated podcasts produced by Google's NotebookLM, which generates audio podcasts with two chatty AI hosts discussing whichever documents a user uploads. While AI-generated podcasts have been discussed as tools, for instance in medical education, they have not yet been analysed as media. By uploading different types of text and analysing the generated outputs I show how the podcasts' structure is built around a fixed template. I also find that NotebookLM not only translates texts from other languages into a perky standardised Mid-Western American accent, it also translates cultural contexts to a white, educated, middle-class American default. This is a distinct development in how publics are shaped by media, marking a departure from the multiple public spheres that scholars have described in human podcasting from the early 2000s until today, where hosts spoke to specific communities and responded to listener comments, to an abstraction of the podcast genre.
Authors: Andreas Einwiller, Kanishka Ghosh Dastidar, Artur Romazanov, Annette Hautli-Janisz, Michael Granitzer, Florian Lemmerich
Abstract: In behavioral sciences, experiments such as the ultimatum game are conducted to assess preferences for fairness or self-interest of study participants. In the dictator game, a simplified version of the ultimatum game where only one of two players makes a single decision, the dictator unilaterally decides how to split a fixed sum of money between themselves and the other player. Although recent studies have explored behavioral patterns of AI agents based on Large Language Models (LLMs) instructed to adopt different personas, we question the robustness of these results. In particular, many of these studies overlook the role of the system prompt - the underlying instructions that shape the model's behavior - and do not account for how sensitive results can be to slight changes in prompts. However, a robust baseline is essential when studying highly complex behavioral aspects of LLMs. To overcome previous limitations, we propose the LLM agent behavior study (LLM-ABS) framework to (i) explore how different system prompts influence model behavior, (ii) get more reliable insights into agent preferences by using neutral prompt variations, and (iii) analyze linguistic features in responses to open-ended instructions by LLM agents to better understand the reasoning behind their behavior. We found that agents often exhibit a strong preference for fairness, as well as a significant impact of the system prompt on their behavior. From a linguistic perspective, we identify that models express their responses differently. Although prompt sensitivity remains a persistent challenge, our proposed framework demonstrates a robust foundation for LLM agent behavior studies. Our code artifacts are available at https://github.com/andreaseinwiller/LLM-ABS.
Authors: Ismail Lamaakal, Chaymae Yahyati, Yassine Maleh, Khalid El Makkaoui, Ibrahim Ouahbi
Abstract: We present BayesQ, an uncertainty-guided post-training quantization framework that is the first to optimize quantization under the posterior expected loss. BayesQ fits a lightweight Gaussian posterior over weights (diagonal Laplace by default; optional K-FAC/low-rank), whitens by the posterior covariance, designs codebooks to minimize posterior-expected distortion, and allocates mixed precision via a greedy knapsack that maximizes marginal expected-loss reduction per bit under a global budget. For scalar quantizers, posterior-expected MSE yields closed-form tables; task-aware proxies are handled by short Monte Carlo on a small calibration set. An optional calibration-only distillation aligns the quantized model with the posterior predictive teacher. At matched average bits/weight of 3.0/3.5/4.0, BayesQ improves over strong PTQ baselines on ResNet-50 (ImageNet) and BERT-base (GLUE) e.g., vs. GPTQ by $+1.5/+0.7/+0.3$ top-1 percentage points on RN50 and $+1.1/+0.4/+0.2$ GLUE points on BERT, while requiring one-time preprocessing comparable to a GPTQ pass. BayesQ reframes low-bit quantization as uncertainty-aware risk minimization in a practical, post-training pipeline.
Authors: Graham L. Bishop (University of California, San Diego)
Abstract: Current bioacoustic AI systems achieve impressive cross-species performance by processing animal communication through transformer architectures, foundation model paradigms, and other computational approaches. However, these approaches overlook a fundamental question: what happens when one form of recursive cognition--AI systems with their attention mechanisms, iterative processing, and feedback loops--encounters the recursive communicative processes of other species? Drawing on philosopher Yuk Hui's work on recursivity and contingency, I argue that AI systems are not neutral pattern detectors but recursive cognitive agents whose own information processing may systematically obscure or distort other species' communicative structures. This creates a double contingency problem: each species' communication emerges through contingent ecological and evolutionary conditions, while AI systems process these signals through their own contingent architectural and training conditions. I propose that addressing this challenge requires reconceptualizing bioacoustic AI from universal pattern recognition toward diplomatic encounter between different forms of recursive cognition, with implications for model design, evaluation frameworks, and research methodologies.
Authors: Yingtong Dou, Zhimeng Jiang, Tianyi Zhang, Mingzhi Hu, Zhichao Xu, Shubham Jain, Uday Singh Saini, Xiran Fan, Jiarui Sun, Menghai Pan, Junpeng Wang, Xin Dai, Liang Wang, Chin-Chia Michael Yeh, Yujie Fan, Vineeth Rakesh, Huiyuan Chen, Mangesh Bendre, Zhongfang Zhuang, Xiaoting Li, Prince Aboagye, Vivian Lai, Minghua Xu, Hao Yang, Yiwei Cai, Mahashweta Das, Yuzhong Chen
Abstract: We present TransactionGPT (TGPT), a foundation model for consumer transaction data within one of world's largest payment networks. TGPT is designed to understand and generate transaction trajectories while simultaneously supporting a variety of downstream prediction and classification tasks. We introduce a novel 3D-Transformer architecture specifically tailored for capturing the complex dynamics in payment transaction data. This architecture incorporates design innovations that enhance modality fusion and computational efficiency, while seamlessly enabling joint optimization with downstream objectives. Trained on billion-scale real-world transactions, TGPT significantly improves downstream classification performance against a competitive production model and exhibits advantages over baselines in generating future transactions. We conduct extensive empirical evaluations utilizing a diverse collection of company transaction datasets spanning multiple downstream tasks, thereby enabling a thorough assessment of TGPT's effectiveness and efficiency in comparison to established methodologies. Furthermore, we examine the incorporation of LLM-derived embeddings within TGPT and benchmark its performance against fine-tuned LLMs, demonstrating that TGPT achieves superior predictive accuracy as well as faster training and inference. We anticipate that the architectural innovations and practical guidelines from this work will advance foundation models for transaction-like data and catalyze future research in this emerging field.
Authors: Maryam Dialameh, Hossein Rajabzadeh, Weiwei Zhang, Walid Ahmed, Hyock Ju Kwon
Abstract: We present Bayesian Mixture of Experts (Bayesian-MoE), a post-hoc uncertainty estimation framework for fine-tuned large language models (LLMs) based on Mixture-of-Experts architectures. Our method applies a structured Laplace approximation to the second linear layer of each expert, enabling calibrated uncertainty estimation without modifying the original training procedure or introducing new parameters. Unlike prior approaches, which apply Bayesian inference to added adapter modules, Bayesian-MoE directly targets the expert pathways already present in MoE models, leveraging their modular design for tractable block-wise posterior estimation. We use Kronecker-factored low-rank approximations to model curvature and derive scalable estimates of predictive uncertainty and marginal likelihood. Experiments on common-sense reasoning benchmarks with Qwen1.5-MoE and DeepSeek-MoE demonstrate that Bayesian-MoE improves both expected calibration error (ECE) and negative log-likelihood (NLL) over baselines, confirming its effectiveness for reliable downstream decision-making.
Authors: Alvin Chauhan
Abstract: Although Large Language Models (LLMs) show exceptional fluency, efforts persist to extract stronger reasoning capabilities from them. Drawing on search-based interpretations of LLM computation, this paper advances a systematic framework for understanding LLM reasoning and optimization. Namely, that enhancing reasoning is best achieved by structuring a multi-agent pipeline to ensure a traversal of the search space in a gradual, incremental, and sequential (GIS) manner. Stated succinctly, high-quality reasoning is a controlled, incremental search. To test this framework, we investigate the efficacy of recursive refinement (RR)--an iterative process of self-criticism, adversarial stress-testing, and integrating critical feedback--as a practical method for implementing GIS search. We designed an experiment comparing a simple, linear pipeline against a complex, explicitly structured pipeline leveraging a recursive refinement layer. The multi-agent models were constructed to reflect the historical personas of three US Founding Fathers (Hamilton, Jefferson, and Madison) using RAG-powered corpora and were prompted to generate responses to three contemporary political issues. Model performance was evaluated using a two-tiered approach: a quantitative score from an LLM arbiter agent and qualitative human judgment. Our results revealed that the complex model consistently outperformed the simple model across all nine test cases with an average arbiter-outputted score of 88.3 versus 71.7. The complex model's arguments were superior in analytical depth, structural nuance, and strategic framing. We conclude that recursive refinement is a robust architectural feature for enhancing LLM reasoning via GIS search.
Authors: Elliot Meyerson, Giuseppe Paolo, Roberto Dailey, Hormoz Shahrzad, Olivier Francon, Conor F. Hayes, Xin Qiu, Babak Hodjat, Risto Miikkulainen
Abstract: LLMs have achieved remarkable breakthroughs in reasoning, insights, and tool use, but chaining these abilities into extended processes at the scale of those routinely executed by humans, organizations, and societies has remained out of reach. The models have a persistent error rate that prevents scale-up: for instance, recent experiments in the Towers of Hanoi benchmark domain showed that the process inevitably becomes derailed after at most a few hundred steps. Thus, although LLM research is often still benchmarked on tasks with relatively few dependent logical steps, there is increasing attention on the ability (or inability) of LLMs to perform long range tasks. This paper describes MAKER, the first system that successfully solves a task with over one million LLM steps with zero errors, and, in principle, scales far beyond this level. The approach relies on an extreme decomposition of a task into subtasks, each of which can be tackled by focused microagents. The high level of modularity resulting from the decomposition allows error correction to be applied at each step through an efficient multi-agent voting scheme. This combination of extreme decomposition and error correction makes scaling possible. Thus, the results suggest that instead of relying on continual improvement of current LLMs, massively decomposed agentic processes (MDAPs) may provide a way to efficiently solve problems at the level of organizations and societies.
Authors: PAN Team, Jiannan Xiang, Yi Gu, Zihan Liu, Zeyu Feng, Qiyue Gao, Yiyan Hu, Benhao Huang, Guangyi Liu, Yichi Yang, Kun Zhou, Davit Abrahamyan, Arif Ahmad, Ganesh Bannur, Junrong Chen, Kimi Chen, Mingkai Deng, Ruobing Han, Xinqi Huang, Haoqiang Kang, Zheqi Li, Enze Ma, Hector Ren, Yashowardhan Shinde, Rohan Shingre, Ramsundar Tanikella, Kaiming Tao, Dequan Yang, Xinle Yu, Cong Zeng, Binglin Zhou, Hector Liu, Zhiting Hu, Eric P. Xing
Abstract: A world model enables an intelligent agent to imagine, predict, and reason about how the world evolves in response to its actions, and accordingly to plan and strategize. While recent video generation models produce realistic visual sequences, they typically operate in the prompt-to-full-video manner without causal control, interactivity, or long-horizon consistency required for purposeful reasoning. Existing world modeling efforts, on the other hand, often focus on restricted domains (e.g., physical, game, or 3D-scene dynamics) with limited depth and controllability, and struggle to generalize across diverse environments and interaction formats. In this work, we introduce PAN, a general, interactable, and long-horizon world model that predicts future world states through high-quality video simulation conditioned on history and natural language actions. PAN employs the Generative Latent Prediction (GLP) architecture that combines an autoregressive latent dynamics backbone based on a large language model (LLM), which grounds simulation in extensive text-based knowledge and enables conditioning on language-specified actions, with a video diffusion decoder that reconstructs perceptually detailed and temporally coherent visual observations, to achieve a unification between latent space reasoning (imagination) and realizable world dynamics (reality). Trained on large-scale video-action pairs spanning diverse domains, PAN supports open-domain, action-conditioned simulation with coherent, long-term dynamics. Extensive experiments show that PAN achieves strong performance in action-conditioned world simulation, long-horizon forecasting, and simulative reasoning compared to other video generators and world models, taking a step towards general world models that enable predictive simulation of future world states for reasoning and acting.
Authors: Ziwei Wang, Leyang Yang, Xiaoxuan Tang, Sheng Zhou, Dajun Chen, Wei Jiang, Yong Li
Abstract: Advances in Multimodal Large Language Models have significantly enhanced Graphical User Interface (GUI) automation. Equipping GUI agents with reliable episodic reasoning capabilities is essential for bridging the gap between users' concise task descriptions and the complexities of real-world execution. Current methods integrate Reinforcement Learning (RL) with System-2 Chain-of-Thought, yielding notable gains in reasoning enhancement. For long-horizon GUI tasks, historical interactions connect each screen to the goal-oriented episode chain, and effectively leveraging these clues is crucial for the current decision. However, existing native GUI agents exhibit weak short-term memory in their explicit reasoning, interpreting the chained interactions as discrete screen understanding, i.e., unawareness of the historical interactions within the episode. This history-agnostic reasoning challenges their performance in GUI automation. To alleviate this weakness, we propose a History-Aware Reasoning (HAR) framework, which encourages an agent to reflect on its own errors and acquire episodic reasoning knowledge from them via tailored strategies that enhance short-term memory in long-horizon interaction. The framework mainly comprises constructing a reflective learning scenario, synthesizing tailored correction guidelines, and designing a hybrid RL reward function. Using the HAR framework, we develop a native end-to-end model, HAR-GUI-3B, which alters the inherent reasoning mode from history-agnostic to history-aware, equipping the GUI agent with stable short-term memory and reliable perception of screen details. Comprehensive evaluations across a range of GUI-related benchmarks demonstrate the effectiveness and generalization of our method.
Authors: Jiarui Liu, Weihao Xuan, Zhijing Jin, Mona Diab
Abstract: Multimodal Large Language Models (MLLMs) often suffer from hallucinations, particularly errors in object existence, attributes, or relations, which undermine their reliability. We introduce TACO (Verified Atomic Confidence Estimation), a simple framework that mitigates hallucinations through self-verification and confidence calibration without relying on external vision experts. TACO decomposes responses into atomic queries, paraphrases them to reduce sensitivity to wording, and estimates confidence using self-consistency (black-box) or self-confidence (gray-box) aggregation, before refining answers with a language model. Experiments on five benchmarks (POPE, MME, HallusionBench, AMBER, and MM-Hal Bench) with two MLLMs (\texttt{LLaVA-1.5-7B} and \texttt{CogVLM2}) show that TACO consistently outperforms direct prompting and Visual Contrastive Decoding, reduces systematic biases, and improves confidence calibration, demonstrating its effectiveness in enhancing the faithfulness of MLLMs.
Authors: Jiliang Hu, Zuchao Li, Baoyuan Qi, Liu Guoming, Ping Wang
Abstract: Significant progress has been made in spoken question answering (SQA) in recent years. However, many existing methods, including large audio language models, struggle with processing long audio. Follow the success of retrieval augmented generation, a speech-related retriever shows promising in help preprocessing long-form speech. But the performance of existing speech-related retrievers is lacking. To address this challenge, we propose CLSR, an end-to-end contrastive language-speech retriever that efficiently extracts question-relevant segments from long audio recordings for downstream SQA task. Unlike conventional speech-text contrastive models, CLSR incorporates an intermediate step that converts acoustic features into text-like representations prior to alignment, thereby more effectively bridging the gap between modalities. Experimental results across four cross-modal retrieval datasets demonstrate that CLSR surpasses both end-to-end speech related retrievers and pipeline approaches combining speech recognition with text retrieval, providing a robust foundation for advancing practical long-form SQA applications.
Authors: Stine Beltoft, Lukas Galke
Abstract: Artificial intelligence (AI) and large language models (LLM) are reshaping science, with most recent advances culminating in fully-automated scientific discovery pipelines. But qualitative research has been left behind. Researchers in qualitative methods are hesitant about AI adoption. Yet when they are willing to use AI at all, they have little choice but to rely on general-purpose tools like ChatGPT to assist with interview interpretation, data annotation, and topic modeling - while simultaneously acknowledging these system's well-known limitations of being biased, opaque, irreproducible, and privacy-compromising. This creates a critical gap: while AI has substantially advanced quantitative methods, the qualitative dimensions essential for meaning-making and comprehensive scientific understanding remain poorly integrated. We argue for developing dedicated qualitative AI systems built from the ground up for interpretive research. Such systems must be transparent, reproducible, and privacy-friendly. We review recent literature to show how existing automated discovery pipelines could be enhanced by robust qualitative capabilities, and identify key opportunities where safe qualitative AI could advance multidisciplinary and mixed-methods research.
Authors: Adam \v{S}torek, Vikas Upadhyay, Marianne Menglin Liu, Daniel W. Peterson, Anshul Mittal, Sujeeth Bharadwaj, Fahad Shah, Dan Roth
Abstract: LLMs now tackle a wide range of software-related tasks, yet we show that their performance varies markedly both across and within these tasks. Routing user queries to the appropriate LLMs can therefore help improve response quality while reducing cost. Prior work, however, has focused mainly on general-purpose LLM routing via black-box models. We introduce Routesplain, the first LLM router for software-related tasks, including multilingual code generation and repair, input/output prediction, and computer science QA. Unlike existing routing approaches, Routesplain first extracts human-interpretable concepts from each query (e.g., task, domain, reasoning complexity) and only routes based on these concepts, thereby providing intelligible, faithful rationales. We evaluate Routesplain on 16 state-of-the-art LLMs across eight software-related tasks; Routesplain outperforms individual models both in terms of accuracy and cost, and equals or surpasses all black-box baselines, with concept-level intervention highlighting avenues for further router improvements.
Authors: Renda Li, Hailang Huang, Fei Wei, Feng Xiong, Yong Wang, Xiangxiang Chu
Abstract: Reinforcement learning (RL) has demonstrated considerable potential for enhancing reasoning in large language models (LLMs). However, existing methods suffer from Gradient Starvation and Policy Degradation when training directly on samples with mixed difficulty. To mitigate this, prior approaches leverage Chain-of-Thought (CoT) data, but the construction of high-quality CoT annotations remains labor-intensive. Alternatively, curriculum learning strategies have been explored but frequently encounter challenges, such as difficulty mismatch, reliance on manual curriculum design, and catastrophic forgetting. To address these issues, we propose AdaCuRL, a Adaptive Curriculum Reinforcement Learning framework that integrates coarse-to-fine difficulty estimation with adaptive curriculum scheduling. This approach dynamically aligns data difficulty with model capability and incorporates a data revisitation mechanism to mitigate catastrophic forgetting. Furthermore, AdaCuRL employs adaptive reference and sparse KL strategies to prevent Policy Degradation. Extensive experiments across diverse reasoning benchmarks demonstrate that AdaCuRL consistently achieves significant performance improvements on both LLMs and MLLMs.
Authors: Gus Cooney, Andrew Reece
Abstract: Conversation is a subject of increasing interest in the social, cognitive, and computational sciences. Yet as conversational datasets continue to increase in size and complexity, researchers lack scalable methods to segment speech-to-text transcripts into conversational "turns"-the basic building blocks of social interaction. We discuss this challenge and then introduce "NaturalTurn," a turn-segmentation algorithm designed to accurately capture the dynamics of conversational exchange. NaturalTurn operates by distinguishing speakers' primary conversational turns from listeners' secondary utterances, such as backchannels, brief interjections, and other forms of parallel speech that characterize human conversation. Using data from a large conversation corpus, we show that NaturalTurn captures conversational turns more accurately than a baseline model. For example, it produces turns with durations and gaps that match empirical literature, reveals stronger linguistic alignment patterns between speakers, and uncovers otherwise hidden relationships between turn-taking and affective outcomes. NaturalTurn thus represents a pragmatic development in machine-generated transcript-processing methods, or "turn models", that will enable researchers to link turn-taking dynamics with important outcomes of social interaction, a central goal of conversation science.
Authors: Ruihan Wu, Chhavi Yadav, Russ Salakhutdinov, Kamalika Chaudhuri
Abstract: Machine unlearning has emerged as an important component in developing safe and trustworthy models. Prior work on fact unlearning in LLMs has mostly focused on removing a specified target fact robustly, but often overlooks its deductive connections to other knowledge. We propose a new setting for fact unlearning, deep unlearning, where the goal is not only to remove a target fact but also to prevent it from being deduced via retained knowledge in the LLM and logical reasoning. We propose three novel metrics: Success-DU and Recall to measure unlearning efficacy, and Accuracy to measure the remainder model utility. To benchmark this setting, we leverage both (1) an existing real-world knowledge dataset, MQuAKE, that provides one-step deduction instances, and (2) newly construct a novel semi-synthetic dataset, Eval-DU, that allows multiple steps of realistic deductions among synthetic facts. Experiments reveal that current methods struggle with deep unlearning: they either fail to deeply unlearn, or excessively remove unrelated facts. Our results suggest that targeted algorithms may have to be developed for robust/deep fact unlearning in LLMs.
Authors: Lawrence K. Q. Yan, Qian Niu, Ming Li, Yichao Zhang, Caitlyn Heqi Yin, Cheng Fei, Benji Peng, Ziqian Bi, Pohsun Feng, Keyu Chen, Tianyang Wang, Yunze Wang, Silin Chen, Ming Liu, Junyu Liu, Xinyuan Song, Riyang Bao, Zekun Jiang, Ziyuan Qin
Abstract: With the increasing application of large language models (LLMs) in the medical domain, evaluating these models' performance using benchmark datasets has become crucial. This paper presents a comprehensive survey of various benchmark datasets employed in medical LLM tasks. These datasets span multiple modalities including text, image, and multimodal benchmarks, focusing on different aspects of medical knowledge such as electronic health records (EHRs), doctor-patient dialogues, medical question-answering, and medical image captioning. The survey categorizes the datasets by modality, discussing their significance, data structure, and impact on the development of LLMs for clinical tasks such as diagnosis, report generation, and predictive decision support. Key benchmarks include MIMIC-III, MIMIC-IV, BioASQ, PubMedQA, and CheXpert, which have facilitated advancements in tasks like medical report generation, clinical summarization, and synthetic data generation. The paper summarizes the challenges and opportunities in leveraging these benchmarks for advancing multimodal medical intelligence, emphasizing the need for datasets with a greater degree of language diversity, structured omics data, and innovative approaches to synthesis. This work also provides a foundation for future research in the application of LLMs in medicine, contributing to the evolving field of medical artificial intelligence.
Authors: Yifan Zhang
Abstract: The rapid advancement of large language models (LLMs) such as GPT-3, PaLM, and Llama has significantly transformed natural language processing, showcasing remarkable capabilities in understanding and generating language. However, a fundamental bottleneck persists: these models often struggle with tasks requiring complex, multi-step reasoning, particularly in mathematical problem-solving. This deficiency stems from the critical scarcity of large-scale, high-quality, domain-specific datasets necessary for cultivating sophisticated reasoning abilities. To overcome this challenge, we introduce Template-based Data Generation (TDG), a novel and scalable paradigm that harnesses frontier LLMs (GPT-4) to automatically generate parameterized meta-templates, which in turn synthesize a virtually infinite stream of high-quality problems and solutions. Using this paradigm, we create TemplateMath Part I: TemplateGSM, a foundational dataset of over 7 million synthetically generated grade school math problems. Each problem is accompanied by a programmatically verifiable solution, offering an unprecedented level of quality at scale. This resource not only resolves the data scarcity issue for supervised fine-tuning but also provides a robust mechanism for model alignment through Reinforcement Learning with Verifiable Rewards (RLVR). Our approach elevates data augmentation by leveraging GPT-4 to generate meta-templates, ensuring diverse and complex problem structures. By providing a scalable solution to the data and verification bottleneck, TDG and TemplateGSM pave the way for a new generation of LLMs with powerful, reliable reasoning skills. Project Page: https://github.com/iiis-ai/TemplateMath
Authors: Hanning Zhang, Juntong Song, Juno Zhu, Yuanhao Wu, Tong Zhang, Cheng Niu
Abstract: Reward Modeling is critical in evaluating and improving the generation of Large Language Models (LLMs). While numerous recent works have shown its feasibility in improving safety, helpfulness, reasoning, and instruction-following ability, its capability and generalization to open-ended long-context generation is still rarely explored. In this paper, we introduce OpenGenAlign, a framework and a high-quality dataset designed to develop reward models to evaluate and improve hallucination-free, comprehensive, reliable, and efficient open-ended long-context generation. We define four key metrics to assess generation quality and develop an automated pipeline to evaluate the outputs of multiple LLMs across long-context QA, Data-to-Text, and Summarization scenarios using o3, ending up with 33K high-quality preference data with a human agreement rate of 81\%. Experimental results first demonstrate that existing reward models perform suboptimally on the held-out benchmark. And Our trained reward model achieves superior performance in the benchmark and effectively improves the generation quality of the policy models using Reinforcement Learning (RL). Additionally, OpenGenAlign could be used for effective guided generation in existing datasets. Furthermore, we demonstrate that the OpenGenAlign could be integrated with reward data from other domains to achieve better performance.
Authors: Richard Futrell, Kyle Mahowald
Abstract: Language models can produce fluent, grammatical text. Nonetheless, some maintain that language models don't really learn language and also that, even if they did, that would not be informative for the study of human learning and processing. On the other side, there have been claims that the success of LMs obviates the need for studying linguistic theory and structure. We argue that both extremes are wrong. LMs can contribute to fundamental questions about linguistic structure, language processing, and learning. They force us to rethink arguments and ways of thinking that have been foundational in linguistics. While they do not replace linguistic structure and theory, they serve as model systems and working proofs of concept for gradient, usage-based approaches to language. We offer an optimistic take on the relationship between language models and linguistics.
Authors: Royson Lee, Minyoung Kim, Fady Rezk, Rui Li, Stylianos I. Venieris, Timothy Hospedales
Abstract: Federated learning (FL) has enabled the training of multilingual large language models (LLMs) on diverse and decentralized multilingual data, especially on low-resource languages. To improve client-specific performance, personalization via the use of parameter-efficient fine-tuning (PEFT) modules such as LoRA is common. This involves a personalization strategy (PS), such as the design of the PEFT adapter structures (e.g., in which layers to add LoRAs and what ranks) and choice of hyperparameters (e.g., learning rates) for fine-tuning. Instead of manual PS configuration, we propose FedP$^2$EFT, a federated learning-to-personalize method for multilingual LLMs in cross-device FL settings. Unlike most existing PEFT structure selection methods, which are prone to overfitting low-data regimes, FedP$^2$EFT collaboratively learns the optimal personalized PEFT structure for each client via Bayesian sparse rank selection. Evaluations on both simulated and real-world multilingual FL benchmarks demonstrate that FedP$^2$EFT largely outperforms existing personalized fine-tuning methods, while complementing other existing FL methods.
Authors: Lucile Favero, Juan Antonio P\'erez-Ortiz, Tanja K\"aser, Nuria Oliver
Abstract: Argument mining algorithms analyze the argumentative structure of essays, making them a valuable tool for enhancing education by providing targeted feedback on the students' argumentation skills. While current methods often use encoder or encoder-decoder deep learning architectures, decoder-only models remain largely unexplored, offering a promising research direction. This paper proposes leveraging open-source, small Large Language Models (LLMs) for argument mining through few-shot prompting and fine-tuning. These models' small size and open-source nature ensure accessibility, privacy, and computational efficiency, enabling schools and educators to adopt and deploy them locally. Specifically, we perform three tasks: segmentation of student essays into arguments, classification of the arguments by type, and assessment of their quality. We empirically evaluate the models on the Feedback Prize - Predicting Effective Arguments dataset of grade 6-12 students essays and demonstrate how fine-tuned small LLMs outperform baseline methods in segmenting the essays and determining the argument types while few-shot prompting yields comparable performance to that of the baselines in assessing quality. This work highlights the educational potential of small, open-source LLMs to provide real-time, personalized feedback, enhancing independent learning and writing skills while ensuring low computational cost and privacy.
Authors: Xiaomin Li, Zhou Yu, Ziji Zhang, Yingying Zhuang, Swair Shah, Narayanan Sadagopan, Anurag Beniwal
Abstract: Large language models (LLMs) have demonstrated remarkable performance across diverse tasks by encoding vast amounts of factual knowledge. However, they are still prone to hallucinations, generating incorrect or misleading information, often accompanied by high uncertainty. Existing methods for hallucination detection primarily focus on quantifying internal uncertainty, which arises from missing or conflicting knowledge within the model. However, hallucinations can also stem from external uncertainty, where ambiguous user queries lead to multiple possible interpretations. In this work, we introduce Semantic Volume, a novel mathematical measure for quantifying both external and internal uncertainty in LLMs. Our approach perturbs queries and responses, embeds them in a semantic space, and computes the Gram matrix determinant of the embedding vectors, capturing their dispersion as a measure of uncertainty. Our framework provides a generalizable and unsupervised uncertainty detection method without requiring internal access to LLMs. We conduct extensive experiments on both external and internal uncertainty detections, demonstrating that our Semantic Volume method consistently outperforms existing baselines in both tasks. Additionally, we provide theoretical insights linking our measure to differential entropy, unifying and extending previous sampling-based uncertainty measures such as the semantic entropy. Semantic Volume is shown to be a robust and interpretable approach to improving the reliability of LLMs by systematically detecting uncertainty in both user queries and model responses.
Authors: Yang Chen, Hui Wang, Shiyao Wang, Junyang Chen, Jiabei He, Jiaming Zhou, Xi Yang, Yequan Wang, Yonghua Lin, Yong Qin
Abstract: While voice technologies increasingly serve aging populations, current systems exhibit significant performance gaps due to inadequate training data capturing elderly-specific vocal characteristics like presbyphonia and dialectal variations. The limited data available on super-aged individuals in existing elderly speech datasets, coupled with overly simple recording styles and annotation dimensions, exacerbates this issue. To address the critical scarcity of speech data from individuals aged 75 and above, we introduce SeniorTalk, a carefully annotated Chinese spoken dialogue dataset. This dataset contains 55.53 hours of speech from 101 natural conversations involving 202 participants, ensuring a strategic balance across gender, region, and age. Through detailed annotation across multiple dimensions, it can support a wide range of speech tasks. We perform extensive experiments on speaker verification, speaker diarization, speech recognition, and speech editing tasks, offering crucial insights for the development of speech technologies targeting this age group.
Authors: Jian Zhang, Zhangqi Wang, Haiping Zhu, Kangda Cheng, Kai He, Bo Li, Qika Lin, Jun Liu, Erik Cambria
Abstract: Large language models (LLMs) typically operate in a question-answering paradigm, where the quality of the input prompt critically affects the response. Automated Prompt Optimization (APO) aims to overcome the cognitive biases of manually crafted prompts and explore a broader prompt design space. However, existing APO methods often suffer from rigid template structures and inefficient exploration in the prompt space. To this end, we propose a Multi-Agent Adaptive Reasoning with Socratic guidance framework (MARS) for APO. MARS consists of five complementary agents and formulates the optimization process as a Partially Observable Markov Decision Process (POMDP), enabling adaptive prompt refinement through explicit state modeling and interactive feedback. Specifically, a Planner agent generates flexible optimization trajectories, a Teacher-Critic-Student triad engages in Socratic-style dialogue to iteratively optimize the prompt based on pseudo-gradient signals in the text space, and a Target agent executes the prompt in downstream tasks to provide performance feedback. MARS integrates reasoning, feedback, and state transition into a unified hidden-state evolution process, improving both the effectiveness and interpretability of optimization. Extensive experiments on multiple datasets demonstrate that MARS outperforms existing APO methods in terms of optimization performance, search efficiency, and interpretability.
Authors: Richard A. Blythe, Casimir Fisch
Abstract: Establishing a communication system is hard because the intended meaning of a signal is unknown to its receiver when first produced, and the signaller also has no idea how that signal will be interpreted. Most theoretical accounts of the emergence of communication systems rely on feedback to reinforce behaviours that have led to successful communication in the past. However, providing such feedback requires already being able to communicate the meaning that was intended or interpreted. Therefore these accounts cannot explain how communication can be bootstrapped from non-communicative behaviours. Here we present a model that shows how a communication system, capable of expressing an unbounded number of meanings, can emerge as a result of individual behavioural differences in a large population without any pre-existing means to determine communicative success. The two key cognitive capabilities responsible for this outcome are behaving predictably in a given situation, and an alignment of psychological states ahead of signal production that derives from shared intentionality. Since both capabilities can exist independently of communication, our results are compatible with theories in which large flexible socially-learned communication systems like language are the product of a general but well-developed capacity for social cognition.
Authors: Dylan Bouchard, Mohit Singh Chauhan
Abstract: Hallucinations are a persistent problem with Large Language Models (LLMs). As these models become increasingly used in high-stakes domains, such as healthcare and finance, the need for effective hallucination detection is crucial. To this end, we outline a versatile framework for closed-book hallucination detection that practitioners can apply to real-world use cases. To achieve this, we adapt a variety of existing uncertainty quantification (UQ) techniques, including black-box UQ, white-box UQ, and LLM-as-a-Judge, transforming them as necessary into standardized response-level confidence scores ranging from 0 to 1. To enhance flexibility, we propose a tunable ensemble approach that incorporates any combination of the individual confidence scores. This approach enables practitioners to optimize the ensemble for a specific use case for improved performance. To streamline implementation, the full suite of scorers is offered in this paper's companion Python toolkit, UQLM. To evaluate the performance of the various scorers, we conduct an extensive set of experiments using several LLM question-answering benchmarks. We find that our tunable ensemble typically surpasses its individual components and outperforms existing hallucination detection methods. Our results demonstrate the benefits of customized hallucination detection strategies for improving the accuracy and reliability of LLMs.
Authors: Shuzheng Si, Haozhe Zhao, Cheng Gao, Yuzhuo Bai, Zhitong Wang, Bofei Gao, Kangyang Luo, Wenhao Li, Yufei Huang, Gang Chen, Fanchao Qi, Minjia Zhang, Baobao Chang, Maosong Sun
Abstract: Teaching large language models (LLMs) to be faithful in the provided context is crucial for building reliable information-seeking systems. Therefore, we propose a systematic framework, CANOE, to reduce faithfulness hallucinations of LLMs across different downstream tasks without human annotations. Specifically, we first synthesize short-form question-answering (QA) data with four diverse tasks to construct high-quality and easily verifiable training data without human annotation. Also, we propose Dual-GRPO, a rule-based reinforcement learning method that includes three tailored rule-based rewards derived from synthesized short-form QA data, while simultaneously optimizing both short-form and long-form response generation. Notably, Dual-GRPO eliminates the need to manually label preference data to train reward models and avoids over-optimizing short-form generation when relying only on the synthesized short-form QA data. Experimental results show that CANOE greatly improves the faithfulness of LLMs across 11 different tasks, even outperforming the most advanced LLMs, e.g., GPT-4o and OpenAI o1.
Authors: Yiming Gao, Bin Wang, Chengwei Wei, Shuo Sun, AiTi Aw
Abstract: Large language models (LLMs) have demonstrated strong instruction-following capabilities in text-based tasks. However, this ability often deteriorates in multimodal models after alignment with non-text modalities such as images or audio. While several recent efforts have investigated instruction-following performance in text and vision-language models, instruction-following in audio-based large language models remains largely unexplored. To bridge this gap, we introduce IFEval-Audio, a novel evaluation dataset designed to assess the ability to follow instructions in an audio LLM. IFEval-Audio contains 280 audio-instruction-answer triples across six diverse dimensions: Content, Capitalization, Symbol, List Structure, Length, and Format. Each example pairs an audio input with a text instruction, requiring the model to generate an output that follows a specified structure. We benchmark state-of-the-art audio LLMs on their ability to follow audio-involved instructions. The dataset is released publicly to support future research in this emerging area.
Authors: Judith Sieker, Clara Lachenmaier, Sina Zarrie{\ss}
Abstract: This paper examines how LLMs handle false presuppositions and whether certain linguistic factors influence their responses to falsely presupposed content. Presuppositions subtly introduce information as given, making them highly effective at embedding disputable or false information. This raises concerns about whether LLMs, like humans, may fail to detect and correct misleading assumptions introduced as false presuppositions, even when the stakes of misinformation are high. Using a systematic approach based on linguistic presupposition analysis, we investigate the conditions under which LLMs are more or less sensitive to adopt or reject false presuppositions. Focusing on political contexts, we examine how factors like linguistic construction, political party, and scenario probability impact the recognition of false presuppositions. We conduct experiments with a newly created dataset and examine three LLMs: OpenAI's GPT-4-o, Meta's LLama-3-8B, and MistralAI's Mistral-7B-v03. Our results show that the models struggle to recognize false presuppositions, with performance varying by condition. This study highlights that linguistic presupposition analysis is a valuable tool for uncovering the reinforcement of political misinformation in LLM responses.
Authors: Jinwen Chen, Hainan Zhang, Fei Sun, Qinnan Zhang, Sijia Wen, Ziwei Wang, Zhiming Zheng
Abstract: Stealthy data poisoning during fine-tuning can backdoor large language models (LLMs), threatening downstream safety. Existing detectors either use classifier-style probability signals--ill-suited to generation--or rely on rewriting, which can degrade quality and even introduce new triggers. We address the practical need to efficiently remove poisoned examples before or during fine-tuning. We observe a robust signal in the response space: after applying TF-IDF to model responses, poisoned examples form compact clusters (driven by consistent malicious outputs), while clean examples remain dispersed. We leverage this with RFTC--Reference-Filtration + TF-IDF Clustering. RFTC first compares each example's response with that of a reference model and flags those with large deviations as suspicious; it then performs TF-IDF clustering on the suspicious set and identifies true poisoned examples using intra-class distance. On two machine translation datasets and one QA dataset, RFTC outperforms prior detectors in both detection accuracy and the downstream performance of the fine-tuned models. Ablations with different reference models further validate the effectiveness and robustness of Reference-Filtration.
Authors: Haitao Li, Ziyu Li, Yiheng Mao, Ziyi Liu, Zhoujian Sun, Zhengxing Huang
Abstract: The advent of multimodal large language models (MLLMs) has sparked interest in their application to electrocardiogram (ECG) analysis. However, existing ECG-focused MLLMs primarily focus on report generation tasks, often limited to single 12-lead, short-duration (10s) ECG inputs, thereby underutilizing the potential of MLLMs. To this end, we aim to develop a MLLM for ECG analysis that supports a broader range of tasks and more flexible ECG inputs. However, existing ECG-QA datasets are often monotonous. To address this gap, we first constructed the anyECG dataset, which encompasses a wide variety of tasks, including report generation, abnormal waveform localization, and open-ended question answering. In addition to standard hospital ECGs, we introduced long-duration reduced-lead ECGs for home environments and multiple ECG comparison scenarios commonly encountered in clinical practice. Furthermore, we propose the anyECG-chat model, which supports dynamic-length ECG inputs and multiple ECG inputs. We trained the model using a three-stage curriculum training recipe with the anyECG dataset. A comprehensive evaluation was conducted, demonstrating that anyECG-chat is capable of supporting various practical application scenarios, including not only common report generation tasks but also abnormal waveform localization for long-duration reduced-lead ECGs in home environments and comprehensive comparative analysis of multiple ECGs. Our code and data are available at: https://github.com/CuCl-2/anyECG-chat.
Authors: Boyang Xue, Qi Zhu, Rui Wang, Sheng Wang, Hongru Wang, Minda Hu, Fei Mi, Yasheng Wang, Lifeng Shang, Qun Liu, Kam-Fai Wong
Abstract: Although demonstrating remarkable performance on reasoning tasks, Large Language Models (LLMs) still tend to fabricate unreliable responses when confronted with problems that are unsolvable or beyond their capability, severely undermining the reliability. Prior studies of LLM reliability have primarily focused on knowledge tasks to identify unanswerable questions, while mathematical reasoning tasks have remained unexplored due to the dearth of unsolvable math problems. To systematically investigate LLM reliability in mathematical reasoning tasks, we formulate the reliability evaluation for both solvable and unsolvable problems. We then develop a ReliableMath dataset which incorporates open-source solvable problems and high-quality unsolvable problems synthesized by our proposed construction workflow with human evaluations. Experiments are conducted on various LLMs with several key findings uncovered. LLMs fail to directly identify unsolvable problems and always generate fabricated responses. When instructing LLMs to indicate unsolvability using a reliable prompt, the reliability of larger-sized LLMs remains on solvable problems, but notably improves on unsolvable problems yet still falls short of solvable problems. However, small LLMs rarely show any progress despite employing reliable prompts. Therefore, we further propose an alignment strategy to enhance small LLMs' reliability, which can significantly improve LLM reliability performances on both in-domain and out-of-domain tasks.
Authors: Vanessa Rebecca Wiyono, David Anugraha, Ayu Purwarianti, Genta Indra Winata
Abstract: Over 200 million people speak Indonesian, yet the language remains significantly underrepresented in preference-based research for large language models (LLMs). Most existing multilingual datasets are derived from English translations, often resulting in content that lacks cultural and linguistic authenticity. To address this gap, we introduce IndoPref, the first fully human-authored and multi-domain Indonesian preference dataset designed to evaluate the naturalness and quality of LLM-generated text. The dataset contains 522 prompts and yields 4,099 human-annotated pairwise preferences from comparisons across five instruction-tuned LLMs. All annotations are natively written in Indonesian with strong inter-annotator agreement, measured by Krippendorff's alpha. Our benchmark spans 10 diverse categories, enabling practitioners to identify LLMs' fine-grained strengths and weaknesses.
Authors: Zunhai Su, Qingyuan Li, Hao Zhang, Weihao Ye, Qibo Xue, YuLei Qian, Yuchen Xie, Ngai Wong, Kehong Yuan
Abstract: In this study, we report, for the first time, the discovery and systematic investigation of a distinct subset of experts that play a pivotal role in the MoE LLMs' forward inference. These experts are prevalent in open-source MoE LLMs, and despite their extremely limited number, pruning them results in a substantial decline in model performance (e.g., prune just three out of 6,144 causes Qwen3-30B-A3B to generate repetitive and uninformative outputs).We refer to these experts as Super Experts (SEs). Our comprehensive analysis provides progressively deeper insights into SEs: (i) SEs are characterized by rare but extreme activation outliers in the output of the down_proj, which give rise to massive activations in the hidden states between decoder layers. Moreover, the distribution of SEs is model-specific, data-agnostic, and remains unaffected by post-training processes. (ii) By pruning SEs, we assess their significance across a variety of tasks, revealing their considerable impact on the model's overall performance, particularly in mathematical reasoning. (iii) We further investigate why compressing SEs exerts such a pronounced impact. We show that, in MoE LLMs, SEs serve as the primary source of the systematic outlier mechanism in Transformers, and that compressing them profoundly disrupts this process, ultimately causing the collapse of attention sinks. These findings advance the understanding of the internal dynamics of MoE LLMs, filling an important gap in the current knowledge. The code is provided in https://github.com/ZunhaiSu/Super-Experts-Profilling.
Authors: Liam G. McCoy, Fateme Nateghi Haredasht, Kanav Chopra, David Wu, David JH Wu, Abass Conteh, Sarita Khemani, Saloni Kumar Maharaj, Vishnu Ravi, Arth Pahwa, Yingjie Weng, Leah Rosengaus, Lena Giang, Kelvin Zhenghao Li, Olivia Jee, Daniel Shirvani, Ethan Goh, Jonathan H. Chen
Abstract: This study evaluates the capacity of large language models (LLMs) to generate structured clinical consultation templates for electronic consultation. Using 145 expert-crafted templates developed and routinely used by Stanford's eConsult team, we assess frontier models -- including o3, GPT-4o, Kimi K2, Claude 4 Sonnet, Llama 3 70B, and Gemini 2.5 Pro -- for their ability to produce clinically coherent, concise, and prioritized clinical question schemas. Through a multi-agent pipeline combining prompt optimization, semantic autograding, and prioritization analysis, we show that while models like o3 achieve high comprehensiveness (up to 92.2\%), they consistently generate excessively long templates and fail to correctly prioritize the most clinically important questions under length constraints. Performance varies across specialties, with significant degradation in narrative-driven fields such as psychiatry and pain medicine. Our findings demonstrate that LLMs can enhance structured clinical information exchange between physicians, while highlighting the need for more robust evaluation methods that capture a model's ability to prioritize clinically salient information within the time constraints of real-world physician communication.
Authors: Keyu Wang, Jin Li, Shu Yang, Zhuoran Zhang, Di Wang
Abstract: Large Language Models (LLMs) often exhibit sycophantic behavior, agreeing with user-stated opinions even when those contradict factual knowledge. While prior work has documented this tendency, the internal mechanisms that enable such behavior remain poorly understood. In this paper, we provide a mechanistic account of how sycophancy arises within LLMs. We first systematically study how user opinions induce sycophancy across different model families. We find that simple opinion statements reliably induce sycophancy, whereas user expertise framing has a negligible impact. Through logit-lens analysis and causal activation patching, we identify a two-stage emergence of sycophancy: (1) a late-layer output preference shift and (2) deeper representational divergence. We also verify that user authority fails to influence behavior because models do not encode it internally. In addition, we examine how grammatical perspective affects sycophantic behavior, finding that first-person prompts (``I believe...'') consistently induce higher sycophancy rates than third-person framings (``They believe...'') by creating stronger representational perturbations in deeper layers. These findings highlight that sycophancy is not a surface-level artifact but emerges from a structural override of learned knowledge in deeper layers, with implications for alignment and truthful AI systems.
Authors: Juyuan Wang, Rongchen Zhao, Wei Wei, Yufeng Wang, Mo Yu, Jie Zhou, Jin Xu, Liyan Xu
Abstract: Narrative comprehension on long stories and novels has been a challenging domain attributed to their intricate plotlines and entangled, often evolving relations among characters and entities. Given the LLM's diminished reasoning over extended context and its high computational cost, retrieval-based approaches remain a pivotal role in practice. However, traditional RAG methods could fall short due to their stateless, single-step retrieval process, which often overlooks the dynamic nature of capturing interconnected relations within long-range context. In this work, we propose ComoRAG, holding the principle that narrative reasoning is not a one-shot process, but a dynamic, evolving interplay between new evidence acquisition and past knowledge consolidation, analogous to human cognition on reasoning with memory-related signals in the brain. Specifically, when encountering a reasoning impasse, ComoRAG undergoes iterative reasoning cycles while interacting with a dynamic memory workspace. In each cycle, it generates probing queries to devise new exploratory paths, then integrates the retrieved evidence of new aspects into a global memory pool, thereby supporting the emergence of a coherent context for the query resolution. Across four challenging long-context narrative benchmarks (200K+ tokens), ComoRAG outperforms strong RAG baselines with consistent relative gains up to 11% compared to the strongest baseline. Further analysis reveals that ComoRAG is particularly advantageous for complex queries requiring global context comprehension, offering a principled, cognitively motivated paradigm towards retrieval-based stateful reasoning. Our framework is made publicly available at https://github.com/EternityJune25/ComoRAG.
Authors: Yuangang Li, Yiqing Shen, Yi Nian, Jiechao Gao, Ziyi Wang, Chenxiao Yu, Shawn Li, Jie Wang, Xiyang Hu, Yue Zhao
Abstract: Large language models (LLMs) exhibit logically inconsistent hallucinations that appear coherent yet violate reasoning principles, with recent research suggesting an inverse relationship between causal reasoning capabilities and such hallucinations. However, existing reasoning approaches in LLMs, such as Chain-of-Thought (CoT) and its graph-based variants, operate at the linguistic token level rather than modeling the underlying causal relationships between variables, lacking the ability to represent conditional independencies or satisfy causal identification assumptions. To bridge this gap, we introduce causal-DAG construction and reasoning (CDCR-SFT), a supervised fine-tuning framework that trains LLMs to explicitly construct variable-level directed acyclic graph (DAG) and then perform reasoning over it. Moreover, we present a dataset comprising 25,368 samples (CausalDR), where each sample includes an input question, explicit causal DAG, graph-based reasoning trace, and validated answer. Experiments on four LLMs across eight tasks show that CDCR-SFT improves the causal reasoning capability with the state-of-the-art 95.33% accuracy on CLADDER (surpassing human performance of 94.8% for the first time) and reduces the hallucination on HaluEval with 10% improvements. It demonstrates that explicit causal structure modeling in LLMs can effectively mitigate logical inconsistencies in LLM outputs. Code is available at https://github.com/MrLYG/CDCR-SFT.
Authors: Huanxuan Liao, Yixing Xu, Shizhu He, Guanchen Li, Xuanwu Yin, Dong Li, Emad Barsoum, Jun Zhao, Kang Liu
Abstract: Long-context inference in large language models (LLMs) is increasingly constrained by the KV cache bottleneck: memory usage grows linearly with sequence length, while attention computation scales quadratically. Existing approaches address this issue by compressing the KV cache along the temporal axis through strategies such as token eviction or merging to reduce memory and computational overhead. However, these methods often neglect fine-grained importance variations across feature dimensions (i.e., the channel axis), thereby limiting their ability to effectively balance efficiency and model accuracy. In reality, we observe that channel saliency varies dramatically across both queries and positions: certain feature channels carry near-zero information for a given query, while others spike in relevance. To address this oversight, we propose SPARK, a training-free plug-and-play method that applies unstructured sparsity by pruning KV at the channel level, while dynamically restoring the pruned entries during attention score computation. Notably, our approach is orthogonal to existing KV compression and quantization techniques, making it compatible for integration with them to achieve further acceleration. By reducing channel-level redundancy, SPARK enables processing of longer sequences within the same memory budget. For sequences of equal length, SPARK not only preserves or improves model accuracy but also reduces KV cache storage by over 30% compared to eviction-based methods. Furthermore, even with an aggressive pruning ratio of 80%, SPARK maintains performance with less degradation than 5% compared to the baseline eviction method, demonstrating its robustness and effectiveness. Our code will be available at https://github.com/Xnhyacinth/SparK.
Authors: Jun Bai, Minghao Tong, Yang Liu, Zixia Jia, Zilong Zheng
Abstract: Context faithfulness is essential for reliable reasoning in context-dependent scenarios. However, large language models often struggle to ground their outputs in the provided context, resulting in irrelevant responses. Inspired by the emergent expert specialization observed in mixture-of-experts architectures, this work investigates whether certain experts exhibit specialization in context utilization, offering a potential pathway toward targeted optimization for improved context faithfulness. To explore this, we propose Router Lens, a method that accurately identifies context-faithful experts. Our analysis reveals that these experts progressively amplify attention to relevant contextual information, thereby enhancing context grounding. Building on this insight, we introduce Context-faithful Expert Fine-Tuning (CEFT), a lightweight optimization approach that selectively fine-tunes context-faithful experts. Experiments across a wide range of benchmarks and models demonstrate that CEFT matches or surpasses the performance of full fine-tuning while being significantly more efficient.
Authors: Mutsumi Sasaki, Go Kamoda, Ryosuke Takahashi, Kosuke Sato, Kentaro Inui, Keisuke Sakaguchi, Benjamin Heinzerling
Abstract: Temporal reasoning and knowledge are essential capabilities for language models (LMs). While much prior work has analyzed and improved temporal reasoning in LMs, most studies have focused solely on the Gregorian calendar. However, many non-Gregorian systems, such as the Japanese, Hijri, and Hebrew calendars, are in active use and reflect culturally grounded conceptions of time. If and how well current LMs can accurately handle such non-Gregorian calendars has not been evaluated so far. Here, we present a systematic evaluation of how well language models handle one such non-Gregorian system: the Japanese wareki. We create datasets that require temporal knowledge and reasoning in using wareki dates. Evaluating open and closed LMs, we find that some models can perform calendar conversions, but GPT-4o, Deepseek V3, and even Japanese-centric models struggle with Japanese calendar arithmetic and knowledge involving wareki dates. Error analysis suggests corpus frequency of Japanese calendar expressions and a Gregorian bias in the model's knowledge as possible explanations. Our results show the importance of developing LMs that are better equipped for culture-specific tasks such as calendar understanding.
Authors: Krithi Shailya, Akhilesh Kumar Mishra, Gokul S Krishnan, Balaraman Ravindran
Abstract: Large Language Models (LLMs) are increasingly used as daily recommendation systems for tasks like education planning, yet their recommendations risk perpetuating societal biases. This paper empirically examines geographic, demographic, and economic biases in university and program suggestions from three open-source LLMs: LLaMA-3.1-8B, Gemma-7B, and Mistral-7B. Using 360 simulated user profiles varying by gender, nationality, and economic status, we analyze over 25,000 recommendations. Results show strong biases: institutions in the Global North are disproportionately favored, recommendations often reinforce gender stereotypes, and institutional repetition is prevalent. While LLaMA-3.1 achieves the highest diversity, recommending 481 unique universities across 58 countries, systemic disparities persist. To quantify these issues, we propose a novel, multi-dimensional evaluation framework that goes beyond accuracy by measuring demographic and geographic representation. Our findings highlight the urgent need for bias consideration in educational LMs to ensure equitable global access to higher education.
Authors: Ahmed Alajrami, Xingwei Tan, Nikolaos Aletras
Abstract: Instruction-tuning plays a vital role in enhancing the task-solving abilities of large language models (LLMs), improving their usability in generating helpful responses on various tasks. However, previous work has demonstrated that they are sensitive to minor variations in instruction phrasing. In this paper, we explore whether introducing perturbations in instruction-tuning data can enhance LLMs' resistance against noisy instructions. We focus on how instruction-tuning with perturbations, such as removing stop words or shuffling words, affects LLMs' performance on the original and perturbed versions of widely-used benchmarks (MMLU, BBH, GSM8K). We further assess learning dynamics and potential shifts in model behavior. Surprisingly, our results suggest that instruction-tuning on perturbed instructions can, in some cases, improve downstream performance. These findings highlight the importance of including perturbed instructions in instruction-tuning, which can make LLMs more resilient to noisy user inputs.
Authors: Dinghong Song, Jierui Xu, Weichu Yang, Pengfei Su, Dong Li
Abstract: AI accelerators, customized to AI workloads, provide cost-effective and high-performance solutions for training and inference. Trainium, an AI accelerator recently developed by Amazon Web Services (AWS), provides an attractive option for LLM training and inference through its heterogeneous architecture. However, leveraging Trainium architecture for high performance can be challenging because of its systolic array architecture and special requirement on data layout. In this paper, we design high-performance matrix multiplication (matmul), a critical compute kernel, for LLM inference on Trainium. We introduce a series of techniques customized to Trainium based on kernel fusion and novel caching strategies to reduce data movement across the software-managed memory hierarchy, maximize SRAM bandwidth, and avoid expensive matrix transpose. Evaluating with nine datasets and four recent LLMs, we show that our system largely outperforms the state-of-the-art matmul implemented by AWS on Trainium: at the level of matmul kernel, it achieves an average 1.35x speedup (up to 2.22x), which translates to an average 1.66x speedup (up to 2.49x) for end-to-end LLM inference.
Authors: Dinghong Song, Yuan Feng, Yiwei Wang, Shangye Chen, Cyril Guyot, Filip Blagojevic, Hyeran Jeon, Pengfei Su, Dong Li
Abstract: Large Language Models (LLMs) are widely used in generative applications such as chatting, code generation, and reasoning. However, many realworld workloads such as classification, question answering, recommendation, and text embedding rely solely on the prefill stage of inference, where the model encodes input sequences without performing autoregressive decoding. In these prefill only scenarios, the self-attention computation becomes the primary performance bottleneck due to its quadratic complexity with respect to sequence length. In this paper, we observe that semantically different sentences often produce similar attention maps across layers and heads. Building on this insight, we propose AttnCache, a framework that accelerates the prefill stage of LLM inference by retrieving and reusing similar attention maps. Based on an attention map memorization database, AttnCache employs efficient caching and similarity search techniques to identify and reuse pre-cached attention maps during inference, thereby reducing the computational overhead of self-attention. Experimental results show that AttnCache achieves an average of 1.2x end-to-end and 2x attention speedup on CPU, and 1.6x end-to-end and 3x attention speedup on GPU, with negligible accuracy degradation.
Authors: Zishuo Zheng, Vidhisha Balachandran, Chan Young Park, Faeze Brahman, Sachin Kumar
Abstract: As large language model (LLM) based systems take on high-stakes roles in real-world decision-making, they must reconcile competing instructions from multiple sources (e.g., model developers, users, and tools) within a single prompt context. Thus, enforcing an instruction hierarchy (IH) in LLMs, where higher-level directives override lower-priority requests, is critical for the reliability and controllability of LLMs. In this work, we reframe instruction hierarchy resolution as a reasoning task. Specifically, the model must first "think" about the relationship between a given user prompt and higher-priority (system) instructions before generating a response. To enable this capability via training, we construct VerIH, an instruction hierarchy dataset of constraint-following tasks with verifiable answers. This dataset comprises both aligned and conflicting system-user instructions. We show that lightweight reinforcement learning with VerIH effectively transfers general reasoning capabilities of models to instruction prioritization. Our finetuned models achieve consistent improvements on instruction following and instruction hierarchy benchmarks. This reasoning ability also generalizes to safety-critical settings beyond the training distribution. By treating safety issues as resolving conflicts between adversarial user inputs and predefined higher-priority policies, our trained model enhances robustness against jailbreak and prompt injection attacks. These results demonstrate that reasoning over instruction hierarchies provides a practical path to reliable LLMs, where updates to system prompts yield controllable and robust changes in model behavior.
Authors: Zihao Yi, Qingxuan Jiang, Ruotian Ma, Xingyu Chen, Qu Yang, Mengru Wang, Fanghua Ye, Ying Shen, Zhaopeng Tu, Xiaolong Li, Linus
Abstract: Large Language Models (LLMs) are increasingly tasked with creative generation, including the simulation of fictional characters. However, their ability to portray non-prosocial, antagonistic personas remains largely unexamined. We hypothesize that the safety alignment of modern LLMs creates a fundamental conflict with the task of authentically role-playing morally ambiguous or villainous characters. To investigate this, we introduce the Moral RolePlay benchmark, a new dataset featuring a four-level moral alignment scale and a balanced test set for rigorous evaluation. We task state-of-the-art LLMs with role-playing characters from moral paragons to pure villains. Our large-scale evaluation reveals a consistent, monotonic decline in role-playing fidelity as character morality decreases. We find that models struggle most with traits directly antithetical to safety principles, such as ``Deceitful'' and ``Manipulative'', often substituting nuanced malevolence with superficial aggression. Furthermore, we demonstrate that general chatbot proficiency is a poor predictor of villain role-playing ability, with highly safety-aligned models performing particularly poorly. Our work provides the first systematic evidence of this critical limitation, highlighting a key tension between model safety and creative fidelity. Our benchmark and findings pave the way for developing more nuanced, context-aware alignment methods.
Authors: Yinjie Cheng, Paul Youssef, Christin Seifert, J\"org Schl\"otterer, Zhixue Zhao
Abstract: Knowledge editing has emerged as a lightweight alternative to retraining for correcting or injecting specific facts in large language models (LLMs). Meanwhile, fine-tuning remains the default operation for adapting LLMs to new domains and tasks. Despite their widespread adoption, these two post-training interventions have been studied in isolation, leaving open a crucial question: if we fine-tune an edited model, do the edits survive? This question is motivated by two practical scenarios: removing covert or malicious edits, and preserving beneficial edits. If fine-tuning impairs edits as shown in Figure 1, current KE methods become less useful, as every fine-tuned model would require re-editing, which significantly increases the cost; if edits persist, fine-tuned models risk propagating hidden malicious edits, raising serious safety concerns. To this end, we systematically quantify edits decay after fine-tuning, investigating how fine-tuning affects knowledge editing. We evaluate two state-of-the-art editing methods (MEMIT, AlphaEdit) and three fine-tuning approaches (full-parameter, LoRA, DoRA) across five LLMs and three datasets, yielding 232 experimental configurations. Our results show that edits decay after fine-tuning, with survival varying across configurations, e.g., AlphaEdit edits decay more than MEMIT edits. Further, we propose selective-layer fine-tuning and find that fine-tuning edited layers only can effectively remove edits, though at a slight cost to downstream performance. Surprisingly, fine-tuning non-edited layers impairs more edits than full fine-tuning. Overall, our study establishes empirical baselines and actionable strategies for integrating knowledge editing with fine-tuning, and underscores that evaluating model editing requires considering the full LLM application pipeline.
Authors: Fangqi Dai, Xingjian Jiang, Zizhuang Deng
Abstract: To prevent misinformation and social issues arising from trustworthy-looking content generated by LLMs, it is crucial to develop efficient and reliable methods for identifying the source of texts. Previous approaches have demonstrated exceptional performance in detecting texts fully generated by LLMs. However, these methods struggle when confronting more advanced LLM output or text with adversarial multi-task machine revision, especially in the black-box setting, where the generating model is unknown. To address this challenge, grounded in the hypothesis that human writing possesses distinctive stylistic patterns, we propose Human Language Preference Detection (HLPD). HLPD employs a reward-based alignment process, Human Language Preference Optimization (HLPO), to shift the scoring model's token distribution toward human-like writing, making the model more sensitive to human writing, therefore enhancing the identification of machine-revised text. We test HLPD in an adversarial multi-task evaluation framework that leverages a five-dimensional prompt generator and multiple advanced LLMs to create diverse revision scenarios. When detecting texts revised by GPT-series models, HLPD achieves a 15.11% relative improvement in AUROC over ImBD, surpassing Fast-DetectGPT by 45.56%. When evaluated on texts generated by advanced LLMs, HLPD achieves the highest average AUROC, exceeding ImBD by 5.53% and Fast-DetectGPT by 34.14%. Code will be made available at https://github.com/dfq2021/HLPD.
Authors: Lynn Greschner, Meike Bauer, Sabine Weber, Roman Klinger
Abstract: The convincingness of an argument does not only depend on its structure (logos), the person who makes the argument (ethos), but also on the emotion that it causes in the recipient (pathos). While the overall intensity and categorical values of emotions in arguments have received considerable attention in the research community, we argue that the emotion an argument evokes in a recipient is subjective. It depends on the recipient's goals, standards, prior knowledge, and stance. Appraisal theories lend themselves as a link between the subjective cognitive assessment of events and emotions. They have been used in event-centric emotion analysis, but their suitability for assessing argument convincingness remains unexplored. In this paper, we evaluate whether appraisal theories are suitable for emotion analysis in arguments by considering subjective cognitive evaluations of the importance and impact of an argument on its receiver. Based on the annotations in the recently published ContArgA corpus, we perform zero-shot prompting experiments to evaluate the importance of gold-annotated and predicted emotions and appraisals for the assessment of the subjective convincingness labels. We find that, while categorical emotion information does improve convincingness prediction, the improvement is more pronounced with appraisals. This work presents the first systematic comparison between emotion models for convincingness prediction, demonstrating the advantage of appraisals, providing insights for theoretical and practical applications in computational argumentation.
Authors: Shiva Upadhye, Richard Futrell
Abstract: Contextual predictability shapes both the form and choice of words in online language production. The effects of the predictability of a word given its previous context are generally well-understood in both production and comprehension, but studies of naturalistic production have also revealed a poorly-understood backward predictability effect of a word given its future context, which may be related to future planning. Here, in two studies of naturalistic speech corpora, we investigate backward predictability effects using improved measures and more powerful language models, introducing a new principled and conceptually motivated information-theoretic predictability measure that integrates predictability from both the future and the past context. Our first study revisits classic predictability effects on word duration. Our second study investigates substitution errors within a generative framework that independently models the effects of lexical, contextual, and communicative factors on word choice, while predicting the actual words that surface as speech errors. We find that our proposed conceptually-motivated alternative to backward predictability yields qualitatively similar effects across both studies. Through a fine-grained analysis of substitution errors, we further show that different kinds of errors are suggestive of how speakers prioritize form, meaning, and context-based information during lexical planning. Together, these findings illuminate the functional roles of past and future context in how speakers encode and choose words, offering a bridge between contextual predictability effects and the mechanisms of sentence planning.
Authors: Paula Ontalvilla, Aitor Ormazabal, Gorka Azkune
Abstract: Skill composition is the ability to combine previously learned skills to solve new tasks. As neural networks acquire increasingly complex skills during their pretraining, it is not clear how successfully they can compose them. In this paper, we focus on Multimodal Large Language Models (MLLM), and study their ability to compose skills across modalities. To this end, we design three evaluation tasks which can be solved sequentially composing two modality-dependent skills, and evaluate several open MLLMs under two main settings: i) prompting the model to directly solve the task, and ii) using a two-step cascaded inference approach, which manually enforces the composition of the two skills for a given task. Even with these straightforward compositions, we find that all evaluated MLLMs exhibit a significant cross-modality skill composition gap. To mitigate the aforementioned gap, we explore two alternatives: i) use chain-of-thought prompting to explicitly instruct MLLMs for skill composition and ii) a specific fine-tuning recipe to promote skill composition. Although those strategies improve model performance, they still exhibit significant skill composition gaps, suggesting that more research is needed to improve cross-modal skill composition in MLLMs.
Authors: Leonid Boytsov, David Akinpelu, Nipun Katyal, Tianyi Lin, Fangwei Gao, Yutian Zhao, Jeffrey Huang, Eric Nyberg
Abstract: We tested over 20 Transformer models for ranking long documents (including recent LongP models trained with FlashAttention and RankGPT models "powered" by OpenAI and Anthropic cloud APIs). We compared them with the simple FirstP baseline, which applied the same model to truncated input (up to 512 tokens). On MS MARCO, TREC DL, and Robust04 no long-document model outperformed FirstP by more than 5% (on average). We hypothesized that this lack of improvement is not due to inherent model limitations, but due to benchmark positional bias (most relevant passages tend to occur early in documents), which is known to exist in MS MARCO. To confirm this, we analyzed positional relevance distributions across four long-document corpora (with six query sets) and observed the same early-position bias. Surprisingly, we also found bias in six BEIR collections, which are typically categorized as short-document datasets. We then introduced a new diagnostic dataset, MS MARCO FarRelevant, where relevant spans were deliberately placed beyond the first 512 tokens. On this dataset, many long-context models (including RankGPT) performed at random-baseline level, suggesting overfitting to positional bias. We also experimented with debiasing training data, but with limited success. Our findings (1) highlight the need for careful benchmark design in evaluating long-context models for document ranking, (2) identify model types that are more robust to positional bias, and (3) motivate further work on approaches to debias training data. We release our code and data to support further research.
Authors: Yupei Liu, Yuqi Jia, Runpeng Geng, Jinyuan Jia, Neil Zhenqiang Gong
Abstract: A prompt injection attack aims to inject malicious instruction/data into the input of an LLM-Integrated Application such that it produces results as an attacker desires. Existing works are limited to case studies. As a result, the literature lacks a systematic understanding of prompt injection attacks and their defenses. We aim to bridge the gap in this work. In particular, we propose a framework to formalize prompt injection attacks. Existing attacks are special cases in our framework. Moreover, based on our framework, we design a new attack by combining existing ones. Using our framework, we conduct a systematic evaluation on 5 prompt injection attacks and 10 defenses with 10 LLMs and 7 tasks. Our work provides a common benchmark for quantitatively evaluating future prompt injection attacks and defenses. To facilitate research on this topic, we make our platform public at https://github.com/liu00222/Open-Prompt-Injection.
Authors: Ruihao Gong, Yifu Ding, Zining Wang, Chengtao Lv, Xingyu Zheng, Jinyang Du, Haotong Qin, Jinyang Guo, Michele Magno, Xianglong Liu
Abstract: Large language models (LLMs) have achieved remarkable advancements in natural language processing, showcasing exceptional performance across various tasks. However, the expensive memory and computational requirements present significant challenges for their practical deployment. Low-bit quantization has emerged as a critical approach to mitigate these challenges by reducing the bit-width of model parameters, activations, and gradients, thus decreasing memory usage and computational demands. This paper presents a comprehensive survey of low-bit quantization methods tailored for LLMs, covering the fundamental principles, system implementations, and algorithmic strategies. An overview of basic concepts and new data formats specific to low-bit LLMs is first introduced, followed by a review of frameworks and systems that facilitate low-bit LLMs across various hardware platforms. Then, we categorize and analyze techniques and toolkits for efficient low-bit training and inference of LLMs. Finally, we conclude with a discussion of future trends and potential advancements of low-bit LLMs. Our systematic overview from basic, system, and algorithm perspectives can offer valuable insights and guidelines for future works to enhance the efficiency and applicability of LLMs through low-bit quantization.
Authors: Can Cui, Yunsheng Ma, Sung-Yeon Park, Zichong Yang, Yupeng Zhou, Juanwu Lu, Juntong Peng, Jiaru Zhang, Ruqi Zhang, Lingxi Li, Yaobin Chen, Jitesh H. Panchal, Amr Abdelraouf, Rohit Gupta, Kyungtae Han, Ziran Wang
Abstract: With the broader adoption and highly successful development of Large Language Models (LLMs), there has been growing interest and demand for applying LLMs to autonomous driving technology. Driven by their natural language understanding and reasoning capabilities, LLMs have the potential to enhance various aspects of autonomous driving systems, from perception and scene understanding to interactive decision-making. In this paper, we first introduce the novel concept of designing Large Language Models for Autonomous Driving (LLM4AD), followed by a review of existing LLM4AD studies. Then, we propose a comprehensive benchmark for evaluating the instruction-following and reasoning abilities of LLM4AD systems, which includes LaMPilot-Bench, CARLA Leaderboard 1.0 Benchmark in simulation and NuPlanQA for multi-view visual question answering. Furthermore, we conduct extensive real-world experiments on autonomous vehicle platforms, examining both on-cloud and on-edge LLM deployment for personalized decision-making and motion control. Next, we explore the future trends of integrating language diffusion models into autonomous driving, exemplified by the proposed ViLaD (Vision-Language Diffusion) framework. Finally, we discuss the main challenges of LLM4AD, including latency, deployment, security and privacy, safety, trust and transparency, and personalization.
Authors: Tatsuki Koga, Ruihan Wu, Zhiyuan Zhang, Kamalika Chaudhuri
Abstract: With the recent remarkable advancement of large language models (LLMs), there has been a growing interest in utilizing them in the domains with highly sensitive data that lies outside their training data. For this purpose, retrieval-augmented generation (RAG) is particularly effective -- it assists LLMs by directly providing relevant information from the external knowledge sources. However, without extra privacy safeguards, RAG outputs risk leaking sensitive information from the external data source. In this work, we explore RAG under differential privacy (DP), a formal guarantee of data privacy. The main challenge with differentially private RAG is how to generate long accurate answers within a moderate privacy budget. We address this by proposing an algorithm that smartly spends privacy budget only for the tokens that require the sensitive information and uses the non-private LLM for other tokens. Our extensive empirical evaluations reveal that our algorithm outperforms the non-RAG baseline under a reasonable privacy budget of $\epsilon\approx 10$ across different models and datasets.
Authors: Xiwen Chen, Wenhui Zhu, Peijie Qiu, Hao Wang, Huayu Li, Haiyu Wu, Aristeidis Sotiras, Yalin Wang, Abolfazl Razi
Abstract: Vision-language models (VLMs) such as CLIP demonstrate strong performance but struggle when adapted to downstream tasks. Prompt learning has emerged as an efficient and effective strategy to adapt VLMs while preserving their pre-trained knowledge. However, existing methods still lead to overfitting and degrade zero-shot generalization. To address this challenge, we propose an optimal transport (OT)-guided prompt learning framework that mitigates forgetting by preserving the structural consistency of feature distributions between pre-trained and fine-tuned models. Unlike conventional point-wise constraints, OT naturally captures cross-instance relationships and expands the feasible parameter space for prompt tuning, allowing a better trade-off between adaptation and generalization. Our approach enforces joint constraints on both vision and text representations, ensuring a holistic feature alignment. Extensive experiments on benchmark datasets demonstrate that our simple yet effective method can outperform existing prompt learning strategies in base-to-novel generalization, cross-dataset evaluation, and domain generalization without additional augmentation or ensemble techniques. The code is available at https://github.com/ChongQingNoSubway/Prompt-OT
Authors: Marios Kokkodis, Richard Demsyn-Jones, Vijay Raghavan
Abstract: Are traditional classification approaches irrelevant in this era of AI hype? We show that there are multiclass classification problems where predictive models holistically outperform LLM prompt-based frameworks. Given text and images from home-service project descriptions provided by Thumbtack customers, we build embeddings-based softmax models that predict the professional category (e.g., handyman, bathroom remodeling) associated with each problem description. We then compare against prompts that ask state-of-the-art LLM models to solve the same problem. We find that the embeddings approach outperforms the best LLM prompts in terms of accuracy, calibration, latency, and financial cost. In particular, the embeddings approach has 49.5\% higher accuracy than the prompting approach, and its superiority is consistent across text-only, image-only, and text-image problem descriptions. Furthermore, it yields well-calibrated probabilities, which we later use as confidence signals to provide contextualized user experience during deployment. On the contrary, prompting scores are overly uninformative. Finally, the embeddings approach is 14 and 81 times faster than prompting in processing images and text respectively, while under realistic deployment assumptions, it can be up to 10 times cheaper. Based on these results, we deployed a variation of the embeddings approach, and through A/B testing we observed performance consistent with our offline analysis. Our study shows that for multiclass classification problems that can leverage proprietary datasets, an embeddings-based approach may yield unequivocally better results. Hence, scientists, practitioners, engineers, and business leaders can use our study to go beyond the hype and consider appropriate predictive models for their classification use cases.
Authors: Alfred Clemedtson, Borun Shi
Abstract: Large language models have shown remarkable language processing and reasoning ability but are prone to hallucinate when asked about private data. Retrieval-augmented generation (RAG) retrieves relevant data that fit into an LLM's context window and prompts the LLM for an answer. GraphRAG extends this approach to structured Knowledge Graphs (KGs) and questions regarding entities multiple hops away. The majority of recent GraphRAG methods either overlook the retrieval step or have ad hoc retrieval processes that are abstract or inefficient. This prevents them from being adopted when the KGs are stored in graph databases supporting graph query languages. In this work, we present GraphRAFT, a retrieve-and-reason framework that finetunes LLMs to generate provably correct Cypher queries to retrieve high-quality subgraph contexts and produce accurate answers. Our method is the first such solution that can be taken off-the-shelf and used on KGs stored in native graph DBs. Benchmarks suggest that our method is sample-efficient and scales with the availability of training data. Our method achieves significantly better results than all state-of-the-art models across all four standard metrics on two challenging Q&As on large text-attributed KGs.
Authors: Yurun Yuan, Fan Chen, Zeyu Jia, Alexander Rakhlin, Tengyang Xie
Abstract: Policy-based methods currently dominate reinforcement learning (RL) pipelines for large language model (LLM) reasoning, leaving value-based approaches largely unexplored. We revisit the classical paradigm of Bellman Residual Minimization and introduce Trajectory Bellman Residual Minimization (TBRM), an algorithm that naturally adapts this idea to LLMs, yielding a simple yet effective off-policy algorithm that optimizes a single trajectory-level Bellman objective using the model's own logits as $Q$-values. TBRM removes the need for critics, importance-sampling ratios, or clipping, and operates with only one rollout per prompt. We prove convergence to the near-optimal KL-regularized policy from arbitrary off-policy data via an improved change-of-trajectory-measure analysis. Experiments on standard mathematical-reasoning benchmarks show that TBRM consistently outperforms policy-based baselines, like PPO and GRPO, with comparable or lower computational and memory overhead. Our results indicate that value-based RL might be a principled and efficient alternative for enhancing reasoning capabilities in LLMs.
Authors: Amit Agarwal, Srikant Panda, Kulbhushan Pachauri
Abstract: In this work, we propose Few Shot Domain Adapting Graph (FS-DAG), a scalable and efficient model architecture for visually rich document understanding (VRDU) in few-shot settings. FS-DAG leverages domain-specific and language/vision specific backbones within a modular framework to adapt to diverse document types with minimal data. The model is robust to practical challenges such as handling OCR errors, misspellings, and domain shifts, which are critical in real-world deployments. FS-DAG is highly performant with less than 90M parameters, making it well-suited for complex real-world applications for Information Extraction (IE) tasks where computational resources are limited. We demonstrate FS-DAG's capability through extensive experiments for information extraction task, showing significant improvements in convergence speed and performance compared to state-of-the-art methods. Additionally, this work highlights the ongoing progress in developing smaller, more efficient models that do not compromise on performance. Code : https://github.com/oracle-samples/fs-dag
Authors: Laura Kopf, Nils Feldhus, Kirill Bykov, Philine Lou Bommer, Anna Hedstr\"om, Marina M. -C. H\"ohne, Oliver Eberle
Abstract: Automated interpretability research aims to identify concepts encoded in neural network features to enhance human understanding of model behavior. Within the context of large language models (LLMs) for natural language processing (NLP), current automated neuron-level feature description methods face two key challenges: limited robustness and the assumption that each neuron encodes a single concept (monosemanticity), despite increasing evidence of polysemanticity. This assumption restricts the expressiveness of feature descriptions and limits their ability to capture the full range of behaviors encoded in model internals. To address this, we introduce Polysemantic FeatuRe Identification and Scoring Method (PRISM), a novel framework specifically designed to capture the complexity of features in LLMs. Unlike approaches that assign a single description per neuron, common in many automated interpretability methods in NLP, PRISM produces more nuanced descriptions that account for both monosemantic and polysemantic behavior. We apply PRISM to LLMs and, through extensive benchmarking against existing methods, demonstrate that our approach produces more accurate and faithful feature descriptions, improving both overall description quality (via a description score) and the ability to capture distinct concepts when polysemanticity is present (via a polysemanticity score).
Authors: Sadia Kamal, Lalu Prasad Yadav Prakash, S M Rafiuddin, Mohammed Rakib, Atriya Sen, Sagnik Ray Choudhury
Abstract: The Political Compass Test (PCT) and similar surveys are commonly used to assess political bias in auto-regressive LLMs. Our rigorous statistical experiments show that while changes to standard generation parameters have minimal effect on PCT scores, prompt phrasing and fine-tuning individually and together can significantly influence results. Interestingly, fine-tuning on politically rich vs. neutral datasets does not lead to different shifts in scores. We also generalize these findings to a similar popular test called 8 Values. Humans do not change their responses to questions when prompted differently (``answer this question'' vs ``state your opinion''), or after exposure to politically neutral text, such as mathematical formulae. But the fact that the models do so raises concerns about the validity of these tests for measuring model bias, and paves the way for deeper exploration into how political and social views are encoded in LLMs.
Authors: Fengran Mo, Jinghan Zhang, Yuchen Hui, Jia Ao Sun, Zhichao Xu, Zhan Su, Jian-Yun Nie
Abstract: Conversational search aims to satisfy users' complex information needs via multiple-turn interactions. The key challenge lies in revealing real users' search intent from the context-dependent queries. Previous studies achieve conversational search by fine-tuning a conversational dense retriever with relevance judgments between pairs of context-dependent queries and documents. However, this training paradigm encounters data scarcity issues. To this end, we propose ConvMix, a mixed-criteria framework to augment conversational dense retrieval, which covers more aspects than existing data augmentation frameworks. We design a two-sided relevance judgment augmentation schema in a scalable manner via the aid of large language models. Besides, we integrate the framework with quality control mechanisms to obtain semantically diverse samples and near-distribution supervisions to combine various annotated data. Experimental results on five widely used benchmarks show that the conversational dense retriever trained by our ConvMix framework outperforms previous baseline methods, which demonstrates our superior effectiveness.
Authors: Bingning Huang, Tu Nguyen, Matthieu Zimmer
Abstract: Recent advances in reasoning with large language models (LLMs) have shown the effectiveness of Monte Carlo Tree Search (MCTS) for generating high-quality intermediate trajectories, particularly in math and symbolic domains. Inspired by this, we explore how MCTS-derived trajectories-traditionally used for training value or reward models-can be repurposed to improve policy optimization in preference-based reinforcement learning (RL). Specifically, we focus on Group Relative Policy Optimization (GRPO), a recent algorithm that enables preference-consistent policy learning without value networks. We reframe GRPO into a staged training paradigm, leveraging a teacher's MCTS rollouts to construct a tree-structured curriculum of prefixes. This introduces the novel challenge of computing advantages for training samples that originate from different prefixes, each with a distinct expected return. To address this, we propose Staged Advantage Estimation (SAE), a framework for computing low-variance, prefix-aware advantages by projecting rewards onto a constraint set that respects the tree's hierarchy. Our empirical results on mathematical reasoning tasks show that SAE improves final accuracy over standard GRPO. This outcome is grounded in our theoretical analysis, which confirms that SAE reduces gradient variance-a principled path to improved sample efficiency. We demonstrate this through practical SAE implementations, comparing efficient heuristics against a formal quadratic program.
Authors: Jielong Tang, Shuang Wang, Zhenxing Wang, Jianxing Yu, Jian Yin
Abstract: Grounded Multimodal Named Entity Recognition (GMNER) extends traditional NER by jointly detecting textual mentions and grounding them to visual regions. While existing supervised methods achieve strong performance, they rely on costly multimodal annotations and often underperform in low-resource domains. Multimodal Large Language Models (MLLMs) show strong generalization but suffer from Domain Knowledge Conflict, producing redundant or incorrect mentions for domain-specific entities. To address these challenges, we propose ReFineG, a three-stage collaborative framework that integrates small supervised models with frozen MLLMs for low-resource GMNER. In the Training Stage, a domain-aware NER data synthesis strategy transfers LLM knowledge to small models with supervised training while avoiding domain knowledge conflicts. In the Refinement Stage, an uncertainty-based mechanism retains confident predictions from supervised models and delegates uncertain ones to the MLLM. In the Grounding Stage, a multimodal context selection algorithm enhances visual grounding through analogical reasoning. In the CCKS2025 GMNER Shared Task, ReFineG ranked second with an F1 score of 0.6461 on the online leaderboard, demonstrating its effectiveness with limited annotations.
Authors: Wenlong Deng, Yi Ren, Yushu Li, Boying Gong, Danica J. Sutherland, Xiaoxiao Li, Christos Thrampoulidis
Abstract: Reinforcement learning with verifiable rewards has significantly advanced the reasoning capabilities of large language models, yet how to explicitly steer training toward exploration or exploitation remains an open problem. We introduce Token Hidden Reward (THR), a token-level metric that quantifies each token's influence on the likelihood of correct responses under Group Relative Policy Optimization (GRPO). We find that training dynamics are dominated by a small subset of tokens with high absolute THR values. Most interestingly, tokens with positive THR strengthen confidence in correct outputs, thus favoring exploitation, while tokens with negative THR preserve probability mass for alternative outputs, enabling exploration. This insight suggests a natural intervention: a THR-guided reweighting algorithm that modulates GRPO's learning signals to explicitly bias training toward exploitation or exploration. We validate the efficacy of this algorithm on diverse math reasoning benchmarks. By amplifying tokens with positive THR value and weakening negative ones, our algorithm improves greedy-decoding accuracy, favoring exploitation. The reverse strategy yields consistent gains in Pass@K accuracy, favoring exploration. We further demonstrate that our algorithm integrates seamlessly with other RL objectives such as GSPO and generalizes across architectures including Llama. These findings establish THR as a principled and fine-grained mechanism for dynamically controlling exploration and exploitation in RL-tuned LLMs, providing new tools for targeted fine-tuning in reasoning-intensive applications.
Authors: Jianhui Yang, Yiming Jin, Pengkun Jiao, Chenhe Dong, Zerui Huang, Shaowei Yao, Xiaojiang Zhou, Dan Ou, Haihong Tang
Abstract: Query-product relevance prediction is fundamental to e-commerce search and has become even more critical in the era of AI-powered shopping, where semantic understanding and complex reasoning directly shape the user experience and business conversion. Large Language Models (LLMs) enable generative, reasoning-based approaches, typically aligned via supervised fine-tuning (SFT) or preference optimization methods like Direct Preference Optimization (DPO). However, the increasing complexity of business rules and user queries exposes the inability of existing methods to endow models with robust reasoning capacity for long-tail and challenging cases. Efforts to address this via reinforcement learning strategies like Group Relative Policy Optimization (GRPO) often suffer from sparse terminal rewards, offering insufficient guidance for multi-step reasoning and slowing convergence. To address these challenges, we propose TaoSR-AGRL, an Adaptive Guided Reinforcement Learning framework for LLM-based relevance prediction in Taobao Search Relevance. TaoSR-AGRL introduces two key innovations: (1) Rule-aware Reward Shaping, which decomposes the final relevance judgment into dense, structured rewards aligned with domain-specific relevance criteria; and (2) Adaptive Guided Replay, which identifies low-accuracy rollouts during training and injects targeted ground-truth guidance to steer the policy away from stagnant, rule-violating reasoning patterns toward compliant trajectories. TaoSR-AGRL was evaluated on large-scale real-world datasets and through online side-by-side human evaluations on Taobao Search. It consistently outperforms DPO and standard GRPO baselines in offline experiments, improving relevance accuracy, rule adherence, and training stability. The model trained with TaoSR-AGRL has been successfully deployed in the main search scenario on Taobao, serving hundreds of millions of users.
Authors: Daniel Gomm, Cornelius Wolff, Madelon Hulsebos
Abstract: Natural language interfaces to tabular data must handle ambiguities inherent to queries. Instead of treating ambiguity as a deficiency, we reframe it as a feature of cooperative interaction where users are intentional about the degree to which they specify queries. We develop a principled framework based on a shared responsibility of query specification between user and system, distinguishing unambiguous and ambiguous cooperative queries, which systems can resolve through reasonable inference, from uncooperative queries that cannot be resolved. Applying the framework to evaluations for tabular question answering and analysis, we analyze the queries in 15 popular datasets, and observe an uncontrolled mixing of query types neither adequate for evaluating a system's execution accuracy nor for evaluating interpretation capabilities. This conceptualization around cooperation in resolving queries informs how to design and evaluate natural language interfaces for tabular data analysis, for which we distill concrete directions for future research and broader implications.
Authors: Jiayi Fu, Qiyao Sun
Abstract: Large language models (LLMs) demonstrate strong capabilities in solving complex tasks when integrated with external tools. The Model Context Protocol (MCP) has become a standard interface for enabling such tool-based interactions. However, these interactions introduce substantial security concerns, particularly when the MCP server is compromised or untrustworthy. While prior benchmarks primarily focus on prompt injection attacks or analyze the vulnerabilities of LLM MCP interaction trajectories, limited attention has been given to the underlying system logs associated with malicious MCP servers. To address this gap, we present the first synthetic benchmark for evaluating LLMs ability to identify security risks from system logs. We define nine categories of MCP server risks and generate 1,800 synthetic system logs using ten state-of-the-art LLMs. These logs are embedded in the return values of 243 curated MCP servers, yielding a dataset of 2,421 chat histories for training and 471 queries for evaluation. Our pilot experiments reveal that smaller models often fail to detect risky system logs, leading to high false negatives. While models trained with supervised fine-tuning (SFT) tend to over-flag benign logs, resulting in elevated false positives, Reinforcement Learning from Verifiable Reward (RLVR) offers a better precision-recall balance. In particular, after training with Group Relative Policy Optimization (GRPO), Llama3.1-8B-Instruct achieves 83% accuracy, surpassing the best-performing large remote model by 9 percentage points. Fine-grained, per-category analysis further underscores the effectiveness of reinforcement learning in enhancing LLM safety within the MCP framework. Code and data are available at: https://github.com/PorUna-byte/MCP-RiskCue
Authors: Yuxuan Sun, Manchen Wang, Shengyi Qian, William R. Wong, Eric Gan, Pierluca D'Oro, Alejandro Castillejo Munoz, Sneha Silwal, Pedro Matias, Nitin Kamra, Satwik Kottur, Nick Raines, Xuanyi Zhao, Joy Chen, Joseph Greer, Andrea Madotto, Allen Bolourchi, James Valori, Kevin Carlberg, Karl Ridgeway, Joseph Tighe
Abstract: AI agents capable of controlling user interfaces have the potential to transform human interaction with digital devices. To accelerate this transformation, two fundamental building blocks are essential: high-quality datasets that enable agents to achieve complex and human-relevant goals, and robust evaluation methods that allow researchers and practitioners to rapidly enhance agent performance. In this paper, we introduce DigiData, a large-scale, high-quality, diverse, multi-modal dataset designed for training mobile control agents. Unlike existing datasets, which derive goals from unstructured interactions, DigiData is meticulously constructed through comprehensive exploration of app features, resulting in greater diversity and higher goal complexity. Additionally, we present DigiData-Bench, a benchmark for evaluating mobile control agents on real-world complex tasks. We demonstrate that the commonly used step-accuracy metric falls short in reliably assessing mobile control agents and, to address this, we propose dynamic evaluation protocols and AI-powered evaluations as rigorous alternatives for agent assessment. Our contributions aim to significantly advance the development of mobile control agents, paving the way for more intuitive and effective human-device interactions.
Authors: Cheng Yuan, Jiawei Shao, Chi Zhang, Xuelong Li
Abstract: Recent years have witnessed the rapid advancements of large language models (LLMs) and their expanding applications, leading to soaring demands for computational resources. The widespread adoption of test-time scaling further aggravates the tension between model capability and resource consumption, highlighting the importance of inference efficiency. However, a unified metric that accurately reflects an LLM's efficiency across different model sizes and architectures remains absent. Motivated by the correlation between compression and intelligence, we introduce information capacity, a measure of model efficiency based on text compression performance relative to computational complexity. Larger models can predict the next token more accurately, achieving greater compression gains but at higher computational costs. Empirical evaluations on mainstream open-source models show that models of varying sizes within a series exhibit consistent information capacity. This metric enables a fair efficiency comparison across model series and accurate performance prediction within a model series. A distinctive feature of information capacity is that it incorporates tokenizer efficiency, which affects both input and output token counts but is often neglected in LLM evaluations. We assess the information capacity of 49 models on 5 heterogeneous datasets and observe consistent results on the influences of tokenizer efficiency, pretraining data, and the mixture-of-experts architecture.