Cultural fine-tuning has become the de facto paradigm for building culture-aware large language models (LLMs), yet existing optimization exclusively for alignment scores provides an incomplete portrait of cultural fidelity by systematically obscuring inherent cultural diversity. This unidimensional evaluation lens prompts a fundamental question: do models genuinely perceive distinct cultural nuances, or do they merely memorize dominant cultural values? To address this, we propose a synergistic evaluation framework that jointly formalizes cultural alignment and diversity. Through extensive benchmarking of six mainstream LLMs on the World Values Survey, this framework uncovers a systematic and critical trade-off: the pursuit of cultural alignment consistently incurs an acute expense of diversity, leading to severe "cultural flattening." Investigating this behavioral shift, we demonstrate that these superficial alignment gains stem from models artificially anchoring to dominant majorities, converging onto a monolithic response pattern that wipes out the heterogeneous distributions inherent to human groups. Crucially, our mechanistic analysis suggests that this diversity collapse is not merely a behavioral anomaly but more likely a structural consequence of the low-rank bias inherent in neural network optimization. Therefore, our findings expose the limitations of current post-training paradigms and call for a shift toward alignment objectives that preserve cross-cultural pluralism.

World models offer a principled way to equip long-horizon LLM agents with foresight: predictions of action consequences before execution. However, unreliable foresight can be ignored, misused, or even degrade downstream decision-making. In this paper, we introduce WorldEvolver, a self-evolving world model framework that revises its deployment-time context while keeping the downstream agent and all model parameters frozen. WorldEvolver integrates three modules: (i) Episodic Memory, which exploits real action transitions through retrieval-based simulation; (ii) Semantic Memory, which extracts persistent heuristic rules from prediction-observation mismatches; and (iii) Selective Foresight, which filters low-confidence predictions before integrating them into agent reasoning context. We evaluate WorldEvolver on ALFWorld and ScienceWorld, measuring world model prediction accuracy on Word2World and downstream agent success rate on AgentBoard. Extensive experiments show that WorldEvolver achieves the highest prediction accuracy across three backbones and leads other world model baselines on downstream agent success rate, demonstrating that test-time memory revision enhances both predictive fidelity and planning performance.

Cross-cultural mediation by large language models (LLMs) requires deciding both when to intervene and how to respond in culturally grounded conflicts. Progress on this problem has been limited by the lack of (1) mediation datasets with measurable downstream effects and (2) principled metrics for evaluating intercultural stance change. To address these gaps, we introduce CC-Mediation, a cross-cultural mediation benchmark of 1,661 ten-turn dialogues grounded in the Developmental Model of Intercultural Sensitivity (DMIS), containing culturally grounded conflicts, mediation interventions, and post-intervention trajectories. We further propose two DMIS-based evaluation metrics: Trajectory AUC, which measures the persistence of intercultural improvement over time, and a signed Wasserstein-1 distance, which measures the magnitude and direction of shifts in intercultural stance. Both metrics show strong agreement with human judgment. Using CC-Mediation, we find that current LLMs have limitations on both axes: intervention timing (when) failure stems from a positional prior that ignores dialogue content, while mediation strategy (how) failure arises from a late-layer elicitation collapse rather than a knowledge deficit.

Large Audio-Language Models (LALMs) integrate audio perception and language understanding within a unified framework, enabling a wide range of real-world applications. Despite recent advances, evaluation for LALMs remains heavily underspecified relative to real-world requirements: most lack true linguistic and cultural authenticity, while others fail to capture acoustic realism. To bridge this gap, we propose GlobeAudio, a multilingual and multicultural benchmark designed to evaluate naturalistic audio understanding. GlobeAudio consists of 5,637 multiple-choice questions across six typologically diverse languages, expertly crafted by native speakers grounded on naturally occurring audio. In order to do well, models must possess higher-level auditory reasoning skills and culturally grounded interpretation. We systematically evaluate representative closed-source and open-source LALMs, as well as cascaded ASR–LLM pipelines. Our experiments reveal substantial performance gaps under natural acoustic conditions, particularly for open-source models and low-resource languages. These findings highlight critical limitations of current LALMs and underscore the importance of naturalistic audio evaluation for future audio-language systems. GlobeAudio can be found at https://huggingface.co/datasets/iNLP-Lab/GlobeAudio.

Multicultural multi-agent systems are increasingly deployed in globally diverse settings, where different agents are grounded in different cultural backgrounds. Existing cultural evaluation focuses on value alignment: how closely a single agent matches a target culture. Yet alignment is a per-agent property and cannot reveal whether a system, taken as a whole, preserves the cultural plurality it is meant to represent. We propose value diversity as a system-level evaluation axis for multicultural agent systems, defined through the dissimilarity between culturally conditioned agents’ responses on a shared value survey. Using the World Values Survey, we evaluate 19 cultures and 18 backbone models across a wide range of system configurations. We find that diversity is largely uncorrelated with alignment, indicating that the two capture complementary system properties, and that current multicultural agent systems fall substantially below human societies in value diversity. Mixed-backbone systems narrow this gap but do not close it, and the gap persists across culture compositions and agent scales. Social interaction further erodes diversity by driving agents toward consensus, and a participatory budgeting case study shows that this homogenization narrows the breadth of collective decision-making. Together, our results establish value diversity as a distinct evaluation axis for multicultural multi-agent systems and reveal a persistent homogenization tendency in current LLM-based societies. Our code and data are publicly available at https://github.com/iNLP-Lab/MultiAgent-Diversity.

Large Audio-Language Models (LALMs) have demonstrated strong performance in spoken question answering (QA), with existing evaluations primarily focusing on answer accuracy and robustness to acoustic perturbations. However, such evaluations implicitly assume that spoken inputs remain semantically answerable, an assumption that often fails in real-world interaction when essential information is missing. In this work, we introduce a repair-aware evaluation setting that explicitly distinguishes between answerable and unanswerable audio inputs. We define answerability as a property of the input itself and construct paired evaluation conditions using a semantic-acoustic masking protocol. Based on this setting, we propose the Evaluability Awareness and Repair (EAR) score, a non-compensatory metric that jointly evaluates task competence under answerable conditions and repair behavior under unanswerable conditions. Experiments on two spoken QA benchmarks across diverse LALMs reveal a consistent gap between answer accuracy and conversational reliability: while many models perform well when inputs are answerable, most fail to recognize semantic unanswerability and initiate appropriate conversational repair. These findings expose a limitation of prevailing accuracy-centric evaluation practices and motivate reliability assessments that treat unanswerable inputs as cues for repair and continued interaction. The core code and dataset are publicly available at https://github.com/sheunghung/EAR.

Large language models (LLMs)-based multi-agent systems have recently shown strong potential for machine translation (MT). However, their application to multi-domain translation (MDT) remains under-explored, particularly in addressing cross-domain word ambiguity. To investigate whether multi-agent approaches can help disambiguation in MDT, we propose a multi-agent collaborative disambiguation framework for MDT (MACD), which leverages the collaborative capabilities of LLMs for disambiguation. MACD consists of four cooperating agents responsible for domain allocation, general translation, domain disambiguation, and translation fusion. Experimental results show that MACD significantly improves translation performance across multiple domains and enhances disambiguation accuracy. Our approach reveals several findings on multi-agent collaboration in resolving word ambiguities.

Recent work suggests that LLMs “know what they don’t know”, positing that hallucinated and factually correct outputs arise from distinct internal processes and can therefore be distinguished using internal signals.However, hallucinations have multifaceted causes: beyond simple knowledge gaps, they can emerge from training incentives that encourage models to exploit statistical shortcuts or spurious associations learned during pretraining.In this paper, we argue that when LLMs rely on such learned associations to produce hallucinations, their internal processes are mechanistically similar to those of factual recall, as both stem from strong statistical correlations encoded in the model’s parameters.To verify this, we propose a novel taxonomy categorizing hallucinations into Unassociated Hallucinations (UHs), where outputs lack parametric grounding, and Associated Hallucinations (AHs), which are driven by spurious associations. Through mechanistic analysis, we compare their computational processes and hidden-state geometries with factually correct outputs.Our results show that hidden states primarily reflect whether the model is recalling parametric knowledge rather than the truthfulness of the output itself. Consequently, AHs exhibit hidden-state geometries that largely overlap with factual outputs, rendering standard detection methods ineffective. In contrast, UHs exhibit distinctive, clustered representations that facilitate reliable detection.

Existing psychological counseling datasets often suffer from monolithic client personas, insufficient therapeutic depth, and a lack of process controllability. To address these critical limitations, we propose PsyChain, a chain-of-agents framework that evolves static counseling corpora into high-fidelity dialogues through collaborative simulation which explicitly models client personality, stage progression, safety monitoring, and expert supervision. PsyChain involves a Client Profiler that extracts life scenarios and pairs them with psychological personality archetypes to synthesize diverse profiles.To simulate the complete counseling process, five specialized agents—Process Monitor, Client Speaker, Safety Monitor, Counselor Supervisor, and Counselor Speaker—collaborate and interact autonomously at each dialogue turn to ensure therapeutic professionalism and safety.We apply this to construct PsyChainD, a Chinese dataset of 10,456 dialogues featuring systematically diverse client profiles. Extensive evaluation across client side, counselor side and overall quality shows substantial improvements. The model trained on PsyChainD achieves 61-91% win rates against domain-specific baselines in pairwise evaluation and the highest average score in human evaluation, indicating potential for real-world counseling.

Large language models (LLMs) are rigorously aligned to refuse harmful requests, a process that inherently cultivates a latent capacity to evaluate and recognize unsafe content. In this work, we reveal that this advanced safety awareness inadvertently introduces a fatal vulnerability. We introduce Posterior Attack, a single-query jailbreak that bypasses guardrails by prompting the model to generate the exact harmful response its internal classifier would normally flag as unsafe. Through extensive empirical evaluation across 30 open-source LLMs (up to 35B parameters in size) and frontier models (e.g., GPT-5, Claude 4.6), we observe a striking phenomenon: models with superior safety-judgment capabilities are disproportionately more susceptible to this exploitation. To explain this, we formalize the Safety Paradox, analytically showing that monotonic improvements in safety alignment naturally amplify posterior vulnerability. Finally, we establish a causal link via reinforcement learning interventions, exemplifying that artificially degrading a model’s safety judgment immunizes it against the attack, whereas enhancing judgment exacerbates the vulnerability. Our findings highlight potential flaws in current alignment paradigms, indicating that defense mechanisms may require further structural refinement.

Output diversity is crucial for Large Language Models as it underpins pluralism and creativity.In this work, we reveal that controlling the language used during model thinking—the language of thought—provides a novel and structural source of output diversity.Our preliminary study shows that different thinking languages occupy distinct regions in a model’s thinking space.Based on this observation, we study two repeated sampling strategies under multilingual thinking—Single-Language Sampling and Mixed-Language Sampling—and conduct diversity evaluation on outputs that are controlled to be in English, regardless of the thinking language used.Across extensive experiments, we demonstrate that switching the thinking language from English to non-English languages consistently increases output diversity, with a clear and consistent positive correlation such that languages farther from English in the thinking space yield larger gains.We further show that aggregating samples across multiple thinking languages yields additional improvements through compositional effects, and that scaling sampling with linguistic heterogeneity expands the model’s diversity ceiling.Finally, we show that these findings translate into practical benefits in pluralistic alignment scenarios, leading to broader coverage of cultural knowledge and value orientations in LLM outputs. Our code is publicly available at https://github.com/iNLP-Lab/Multilingual-LoT-Diversity.

As Large Language Models (LLMs) increasingly operate as Deep Research (DR) Agents capable of autonomous investigation and information synthesis, reliable evaluation of their task performance has become a critical bottleneck. Current benchmarks predominantly rely on static datasets, which suffer from several limitations: limited task generality, temporal misalignment, and data contamination. To address these, we introduce DR-Arena, a fully automated evaluation framework that pushes DR agents to their capability limits through dynamic investigation. DR-Arena constructs real-time Information Trees from fresh web trends to ensure the evaluation rubric is synchronized with the live world state, and employs an automated Examiner to generate structured tasks testing two orthogonal capabilities: Deep reasoning and Wide coverage. DR-Arena further adopts Adaptive Evolvement Loop, a state-machine controller that dynamically escalates task complexity based on real-time performance, demanding deeper deduction or wider aggregation until a decisive capability boundary emerges. Experiments with six advanced DR agents demonstrate that DR-Arena achieves a Spearman correlation of 0.94 with the LMSYS Search Arena leaderboard. This represents state-of-the-art alignment with human preferences without any manual efforts, validating DR-Arena as a reliable alternative for costly human adjudication.

Large language models (LLMs) have demonstrated multilingual capabilities, yet they are mostly English-centric due to the imbalanced training corpora. While prior works have leveraged this bias to enhance multilingual performance through translation, they have been largely limited to natural language processing (NLP) tasks. In this work, we extend the evaluation to real-world user queries and non-English-centric LLMs, offering a broader examination of multilingual performance. Our key contribution lies in demonstrating that while translation into English can boost the performance of English-centric LLMs on NLP tasks, it is not universally optimal. For culture-related tasks that need deep language understanding, prompting in the native language proves more effective as it better captures the nuances of culture and language. Our experiments expose varied behaviors across LLMs and tasks in the multilingual context, underscoring the need for a more comprehensive approach to multilingual evaluation. Therefore, we call for greater efforts in developing and evaluating LLMs that go beyond English-centric paradigms.

This study introduces two novel benchmarks, SeaExam and SeaBench, designed to evaluate the capabilities of Large Language Models (LLMs) in Southeast Asian (SEA) application scenarios. Unlike existing multilingual datasets primarily derived from English translations, these benchmarks are constructed based on real-world scenarios from SEA regions. SeaExam draws from regional educational exams to form a comprehensive dataset that encompasses subjects such as local history and literature. In contrast, SeaBench is crafted around multi-turn, open-ended tasks that reflect daily interactions within SEA communities. Our evaluations demonstrate that SeaExam and SeaBench more effectively discern LLM performance on SEA language tasks compared to their translated benchmarks. This highlights the importance of using real-world queries to assess the multilingual capabilities of LLMs.

The tendency of Large Language Models (LLMs) to generate hallucinations raises concerns regarding their reliability. Therefore, confidence estimations indicating the extent of trustworthiness of the generations become essential. However, current LLM confidence estimations in languages other than English remain underexplored. This paper addresses this gap by introducing a comprehensive investigation of Multilingual Confidence estimation (MlingConf) on LLMs, focusing on both language-agnostic (LA) and language-specific (LS) tasks to explore the performance and language dominance effects of multilingual confidence estimations on different tasks. The benchmark comprises four meticulously checked and human-evaluated high-quality multilingual datasets for LA tasks and one for the LS task tailored to specific social, cultural, and geographical contexts of a language. Our experiments reveal that on LA tasks English exhibits notable linguistic dominance in confidence estimations than other languages, while on LS tasks, using question-related language to prompt LLMs demonstrates better linguistic dominance in multilingual confidence estimations. The phenomena inspire a simple yet effective native-tone prompting strategy by employing language-specific prompts for LS tasks, effectively improving LLMs’ reliability and accuracy in LS scenarios.

Instruction tuning is vital for enhancing the performance of large language models (LLMs), but existing text-to-text methods, referred to as TextTuning, struggle with issues such as generalization, robustness, and controllability due to their lack of explicit task structures. We introduce JsonTuning, a structure-to-structure approach that uses JSON structures to represent tasks. This method improves generalization by clarifying task elements and their relations, boosts robustness by minimizing ambiguity, and enhances controllability by allowing precise control over outputs. We conduct an extensive comparative analysis between JsonTuning and TextTuning using various language models and benchmarks. Our findings reveal that JsonTuning consistently surpasses TextTuning in terms of performance, robustness, and controllability across different scenarios. By overcoming the limitations of TextTuning, JsonTuning demonstrates significant potential for developing more effective and reliable LLMs capable of handling diverse scenarios.

Large Language Models (LLMs) have transformed natural language processing, yet their substantial model sizes often demand significant computational resources. To preserve computing resources and accelerate inference speed, it is crucial to prune redundant parameters, especially for experienced users who often need expert models tailored to specific downstream scenarios. However, current pruning methods primarily focus on maintaining models’ general capabilities, either requiring extensive post-training or performing poorly due to coarse-grained pruning. In this work, we design a  ̲Custom  ̲Pruning method (Cus-Prun) to prune a large general model into a smaller lightweight expert model, which is positioned along the “language”, “domain” and “task” dimensions. By identifying and pruning irrelevant neurons of each dimension, Cus-Prun creates expert models without any post-training. Our experiments demonstrate that Cus-Prun consistently outperforms other methods, achieving minimal loss in both expert and general capabilities across various models from different model families and sizes.

Recent progress in large language models (LLMs) has opened new possibilities for mental health support, yet current approaches lack realism in simulating specialized psychotherapy and fail to capture therapeutic progression over time. Narrative therapy, which helps individuals transform problematic life stories into empowering alternatives, remains underutilized due to limited access and social stigma. We address these limitations through a comprehensive framework with two core components. First, INT (Interactive Narrative Therapist) simulates expert narrative therapists by planning therapeutic stages, guiding reflection levels, and generating contextually appropriate responses through retrieval-augmentation. Second, IMA (Innovative Moment Assessment) provides a therapy-centric evaluation method that quantifies effectiveness by tracking “Innovative Moments” (IMs), critical narrative shifts in client speech signaling therapy progress. Experimental results on 260 simulated clients and 230 human participants reveal that INT consistently outperforms standard methods in therapeutic quality and depth. We further demonstrate the effectiveness of INT in synthesizing high-quality support conversations to facilitate social applications.

Although large language models (LLMs) store vast amount of knowledge in their parameters, they still have limitations in the memorization and utilization of certain knowledge, leading to undesired behaviors such as generating untruthful and inaccurate responses. This highlights the critical need to understand the knowledge boundary of LLMs, a concept that remains inadequately defined in existing research. In this survey, we propose a comprehensive definition of the LLM knowledge boundary and introduce a formalized taxonomy categorizing knowledge into four distinct types. Using this foundation, we systematically review the field through three key lenses: the motivation for studying LLM knowledge boundaries, methods for identifying these boundaries, and strategies for mitigating the challenges they present. Finally, we discuss open challenges and potential research directions in this area. We aim for this survey to offer the community a comprehensive overview, facilitate access to key issues, and inspire further advancements in LLM knowledge research.

As LLMs continuously evolve, there is an urgent need for a reliable evaluation method that delivers trustworthy results promptly. Currently, static benchmarks suffer from inflexibility and unreliability, leading users to prefer human voting platforms like Chatbot Arena. However, human evaluations require significant manual effort. Therefore, we propose Auto-Arena, an innovative framework that automates the entire evaluation process using LLM-powered agents. Firstly, an LLM examiner generates questions. Then, two LLM candidates engage in a multi-round peer battle based on the questions, aiming at revealing their true performance differences. Finally, a committee of LLM judges collaboratively discusses and decides the winner, reducing bias and enhancing fairness. During the peer battles, we observe intriguing scenarios where the LLM candidates display competitive behaviors and learn from the opponents. In our extensive experiments involving 15 recent LLMs, Auto-Arena shows a 92.14% correlation with human preferences, surpassing all previous expert-annotated benchmarks without any manual efforts. Auto-Arena offers a promising alternative to current human evaluation platforms for evaluating LLMs automatically.

Large Language Models (LLMs) have significantly advanced the fact-checking studies. However, existing automated fact-checking evaluation methods rely on static datasets and classification metrics, which fail to automatically evaluate the justification production and uncover the nuanced limitations of LLMs in fact-checking. In this work, we introduce FACT-AUDIT, an agent-driven framework that adaptively and dynamically assesses LLMs’ fact-checking capabilities. Leveraging importance sampling principles and multi-agent collaboration, FACT-AUDIT generates adaptive and scalable datasets, performs iterative model-centric evaluations, and updates assessments based on model-specific responses. By incorporating justification production alongside verdict prediction, this framework provides a comprehensive and evolving audit of LLMs’ factual reasoning capabilities, to investigate their trustworthiness. Extensive experiments demonstrate that FACT-AUDIT effectively differentiates among state-of-the-art LLMs, providing valuable insights into model strengths and limitations in model-centric fact-checking analysis.

Large language models (LLMs) have become increasingly central to AI applications worldwide, necessitating robust multilingual safety alignment to ensure secure deployment across diverse linguistic contexts. Existing preference learning methods for safety alignment, such as RLHF and DPO, are primarily monolingual and struggle with noisy multilingual data. To address these limitations, we introduce Multilingual reward gaP Optimization (MPO), a novel approach that leverages the well-aligned safety capabilities of the dominant language (e.g., English) to improve safety alignment across multiple languages. MPO directly minimizes the reward gap difference between the dominant language and target languages, effectively transferring safety capabilities while preserving the original strengths of the dominant language. Extensive experiments on three LLMs, LLaMA-3.1, Gemma-2 and Qwen2.5, validate MPO’s efficacy in multilingual safety alignment without degrading general multilingual utility.

This paper introduces a Dual Evaluation Framework to comprehensively assess the multilingual capabilities of LLMs. By decomposing the evaluation along the dimensions of linguistic medium and cultural context, this framework enables a nuanced analysis of LLMs’ ability to process questions within both native and cross-cultural contexts cross-lingually. Extensive evaluations are conducted on a wide range of models, revealing a notable “Cultural-Linguistic Synergy” phenomenon, where models exhibit better performance when questions are culturally aligned with the language. This phenomenon is further explored through interpretability probing, which shows that a higher proportion of specific neurons are activated in a language’s cultural context. This activation proportion could serve as a potential indicator for evaluating multilingual performance during model training. Our findings challenge the prevailing notion that LLMs, primarily trained on English data, perform uniformly across languages and highlight the necessity of culturally and linguistically model evaluations.

Despite the remarkable achievements of large language models (LLMs) in various tasks, there remains a linguistic bias that favors high-resource languages, such as English, often at the expense of low-resource and regional languages. To address this imbalance, we introduce SeaLLMs, an innovative series of language models that specifically focuses on Southeast Asian (SEA) languages. SeaLLMs are built upon popular English-centric models through continued pre-training with an extended vocabulary, specialized instruction and alignment tuning to better capture the intricacies of regional languages. This allows them to respect and reflect local cultural norms, customs, stylistic preferences, and legal considerations. Our comprehensive evaluation demonstrates that SeaLLM models exhibit superior performance across a wide spectrum of linguistic tasks and assistant-style instruction-following capabilities relative to comparable open-source models. Moreover, they outperform ChatGPT-3.5 in non-Latin languages, such as Thai, Khmer, Lao, and Burmese, by large margins while remaining lightweight and cost-effective to operate.

The goal-oriented document-grounded dialogue aims at responding to the user query based on the dialogue context and supporting document. Existing studies tackle this problem by decomposing it into two sub-tasks: knowledge identification and response generation. However, such pipeline methods would unavoidably suffer from the error propagation issue. This paper proposes to unify these two sub-tasks via sequentially generating the grounding knowledge and the response. We further develop a prompt-connected multi-task learning strategy to model the characteristics and connections of different tasks and introduce linear temperature scheduling to reduce the negative effect of irrelevant document information. Experimental results demonstrate the effectiveness of our framework.