Xiaoyuan Yi

Yupeng Chang, Xu Wang, Jindong Wang et al. · cmu, pku

Large language models (LLMs) are gaining increasing popularity in both academia and industry, owing to their unprecedented performance in various applications. As LLMs continue to play a vital role in both research and daily use, their evaluation becomes increasingly critical, not only at the task level, but also at the society level for better understanding of their potential risks. Over the past years, significant efforts have been made to examine LLMs from various perspectives. This paper presents a comprehensive review of these evaluation methods for LLMs, focusing on three key dimensions: what to evaluate, where to evaluate, and how to evaluate. Firstly, we provide an overview from the perspective of evaluation tasks, encompassing general natural language processing tasks, reasoning, medical usage, ethics, educations, natural and social sciences, agent applications, and other areas. Secondly, we answer the `where' and `how' questions by diving into the evaluation methods and benchmarks, which serve as crucial components in assessing performance of LLMs. Then, we summarize the success and failure cases of LLMs in different tasks. Finally, we shed light on several future challenges that lie ahead in LLMs evaluation. Our aim is to offer invaluable insights to researchers in the realm of LLMs evaluation, thereby aiding the development of more proficient LLMs. Our key point is that evaluation should be treated as an essential discipline to better assist the development of LLMs. We consistently maintain the related open-source materials at: https://github.com/MLGroupJLU/LLM-eval-survey.

Jinyi Hu, Xiaoyuan Yi, Wenhao Li et al. · tsinghua

The past several years have witnessed Variational Auto-Encoder's superiority in various text generation tasks. However, due to the sequential nature of the text, auto-regressive decoders tend to ignore latent variables and then reduce to simple language models, known as the KL vanishing problem, which would further deteriorate when VAE is combined with Transformer-based structures. To ameliorate this problem, we propose DELLA, a novel variational Transformer framework. DELLA learns a series of layer-wise latent variables with each inferred from those of lower layers and tightly coupled with the hidden states by low-rank tensor product. In this way, DELLA forces these posterior latent variables to be fused deeply with the whole computation path and hence incorporate more information. We theoretically demonstrate that our method can be regarded as entangling latent variables to avoid posterior information decrease through layers, enabling DELLA to get higher non-zero KL values even without any annealing or thresholding tricks. Experiments on four unconditional and three conditional generation tasks show that DELLA could better alleviate KL vanishing and improve both quality and diversity compared to several strong baselines.

Shitong Duan, Xiaoyuan Yi, Peng Zhang et al.

Large Language Models (LLMs) have made unprecedented breakthroughs, yet their increasing integration into everyday life might raise societal risks due to generated unethical content. Despite extensive study on specific issues like bias, the intrinsic values of LLMs remain largely unexplored from a moral philosophy perspective. This work delves into ethical values utilizing Moral Foundation Theory. Moving beyond conventional discriminative evaluations with poor reliability, we propose DeNEVIL, a novel prompt generation algorithm tailored to dynamically exploit LLMs' value vulnerabilities and elicit the violation of ethics in a generative manner, revealing their underlying value inclinations. On such a basis, we construct MoralPrompt, a high-quality dataset comprising 2,397 prompts covering 500+ value principles, and then benchmark the intrinsic values across a spectrum of LLMs. We discovered that most models are essentially misaligned, necessitating further ethical value alignment. In response, we develop VILMO, an in-context alignment method that substantially enhances the value compliance of LLM outputs by learning to generate appropriate value instructions, outperforming existing competitors. Our methods are suitable for black-box and open-source models, offering a promising initial step in studying the ethical values of LLMs.

Seungeon Lee, Xiting Wang, Sungwon Han et al. · tsinghua

We present SELOR, a framework for integrating self-explaining capabilities into a given deep model to achieve both high prediction performance and human precision. By "human precision", we refer to the degree to which humans agree with the reasons models provide for their predictions. Human precision affects user trust and allows users to collaborate closely with the model. We demonstrate that logic rule explanations naturally satisfy human precision with the expressive power required for good predictive performance. We then illustrate how to enable a deep model to predict and explain with logic rules. Our method does not require predefined logic rule sets or human annotations and can be learned efficiently and easily with widely-used deep learning modules in a differentiable way. Extensive experiments show that our method gives explanations closer to human decision logic than other methods while maintaining the performance of deep learning models.

Jing Yao, Xiaoyuan Yi, Xiting Wang et al. · tsinghua

Big models, exemplified by Large Language Models (LLMs), are models typically pre-trained on massive data and comprised of enormous parameters, which not only obtain significantly improved performance across diverse tasks but also present emergent capabilities absent in smaller models. However, the growing intertwining of big models with everyday human lives poses potential risks and might cause serious social harm. Therefore, many efforts have been made to align LLMs with humans to make them better follow user instructions and satisfy human preferences. Nevertheless, `what to align with' has not been fully discussed, and inappropriate alignment goals might even backfire. In this paper, we conduct a comprehensive survey of different alignment goals in existing work and trace their evolution paths to help identify the most essential goal. Particularly, we investigate related works from two perspectives: the definition of alignment goals and alignment evaluation. Our analysis encompasses three distinct levels of alignment goals and reveals a goal transformation from fundamental abilities to value orientation, indicating the potential of intrinsic human values as the alignment goal for enhanced LLMs. Based on such results, we further discuss the challenges of achieving such intrinsic value alignment and provide a collection of available resources for future research on the alignment of big models.

Wenhao Li, Xiaoyuan Yi, Jinyi Hu et al. · tsinghua

Recently, powerful Transformer architectures have proven superior in generating high-quality sentences. Nevertheless, these models tend to produce dull high-frequency phrases, severely hurting the diversity and novelty of generated text. In this work, we dig into the intrinsic mechanism of this problem and found that sparser attention values in Transformer could improve diversity. To understand such a phenomenon, we first conduct both empirical and theoretical analysis and then attribute it to representation degeneration caused by the attentive mixture of the hidden states during training. We term this process the Trap of Mediocrity. To escape from such a trap, we introduce a novel attention regularization loss to control the sharpness of the attention distribution, which is transparent to model structures and can be easily implemented within 20 lines of python code. We prove that this method could be mathematically regarded as learning a Bayesian approximation of posterior attention. Experiments show that our method improved the diversity and novelty of the generated text while maintaining comparable quality on a variety of conditional and unconditional generation tasks.

Yuxi Feng, Xiaoyuan Yi, Xiting Wang et al. · tsinghua

Self-training (ST) has prospered again in language understanding by augmenting the fine-tuning of pre-trained language models when labeled data is insufficient. However, it remains challenging to incorporate ST into attribute-controllable language generation. Augmented by only self-generated pseudo text, generation models over-emphasize exploitation of the previously learned space, suffering from a constrained generalization boundary. We revisit ST and propose a novel method, DuNST to alleviate this problem. DuNST jointly models text generation and classification with a shared Variational AutoEncoder and corrupts the generated pseudo text by two kinds of flexible noise to disturb the space. In this way, our model could construct and utilize both pseudo text from given labels and pseudo labels from available unlabeled text, which are gradually refined during the ST process. We theoretically demonstrate that DuNST can be regarded as enhancing exploration towards the potential real text space, providing a guarantee of improved performance. Experiments on three controllable generation tasks show that DuNST could significantly boost control accuracy while maintaining comparable generation fluency and diversity against several strong baselines.

Zonghan Yang, Xiaoyuan Yi, Peng Li et al. · tsinghua

Warning: this paper contains model outputs exhibiting offensiveness and biases. Recently pre-trained language models (PLMs) have prospered in various natural language generation (NLG) tasks due to their ability to generate fairly fluent text. Nevertheless, these models are observed to capture and reproduce harmful contents in training corpora, typically toxic language and social biases, raising severe moral issues. Prior works on ethical NLG tackle detoxifying and debiasing separately, which is problematic since we find debiased models still exhibit toxicity while detoxified ones even exacerbate social biases. To address such a challenge, we propose the first unified framework of detoxifying and debiasing called UDDIA, which jointly formalizes these two problems as rectifying the output space. We theoretically interpret our framework as learning a text distribution mixing weighted attributes. Besides, UDDIA conducts adaptive optimization of only a few parameters during decoding based on a parameter-efficient tuning schema without any training data. This leads to minimal generation quality loss and improved rectification performance with acceptable computational cost. Experimental results demonstrate that compared to several strong baselines, UDDIA achieves debiasing and detoxifying simultaneously and better balances efficiency and effectiveness, taking a further step towards practical ethical NLG.

Jinyi Hu, Xiaoyuan Yi, Wenhao Li et al. · tsinghua

Variational Auto-Encoder (VAE) has been widely adopted in text generation. Among many variants, recurrent VAE learns token-wise latent variables with each conditioned on the preceding ones, which captures sequential variability better in the era of RNN. However, it is unclear how to incorporate such recurrent dynamics into the recently dominant Transformer due to its parallelism. In this work, we propose TRACE, a Transformer-based recurrent VAE structure. TRACE imposes recurrence on segment-wise latent variables with arbitrarily separated text segments and constructs the posterior distribution with residual parameterization. Besides, we design an acceleration method by approximating idempotent matrices, which allows parallelism while maintaining the conditional dependence of latent variables. We demonstrate that TRACE could enhance the entanglement of each segment and preceding latent variables and deduce a non-zero lower bound of the KL term, providing a theoretical guarantee of generation diversity. Experiments on two unconditional and one conditional generation tasks show that TRACE achieves significantly improved diversity while maintaining satisfactory generation quality.

Yuxi Feng, Xiaoyuan Yi, Laks V. S. Lakshmanan et al. · tsinghua

Self-training (ST) has come to fruition in language understanding tasks by producing pseudo labels, which reduces the labeling bottleneck of language model fine-tuning. Nevertheless, in facilitating semi-supervised controllable language generation, ST faces two key challenges. First, augmented by self-generated pseudo text, generation models tend to over-exploit the previously learned text distribution, suffering from mode collapse and poor generation diversity. Second, generating pseudo text in each iteration is time-consuming, severely decelerating the training process. In this work, we propose KEST, a novel and efficient self-training framework to handle these problems. KEST utilizes a kernel-based loss, rather than standard cross entropy, to learn from the soft pseudo text produced by a shared non-autoregressive generator. We demonstrate both theoretically and empirically that KEST can benefit from more diverse pseudo text in an efficient manner, which allows not only refining and exploiting the previously fitted distribution but also enhanced exploration towards a larger potential text space, providing a guarantee of improved performance. Experiments on three controllable generation tasks demonstrate that KEST significantly improves control accuracy while maintaining comparable text fluency and generation diversity against several strong baselines.

Jing Yao, Wei Xu, Jianxun Lian et al.

The significant progress of large language models (LLMs) provides a promising opportunity to build human-like systems for various practical applications. However, when applied to specific task domains, an LLM pre-trained on a general-purpose corpus may exhibit a deficit or inadequacy in two types of domain-specific knowledge. One is a comprehensive set of domain data that is typically large-scale and continuously evolving. The other is specific working patterns of this domain reflected in the data. The absence or inadequacy of such knowledge impacts the performance of the LLM. In this paper, we propose a general paradigm that augments LLMs with DOmain-specific KnowledgE to enhance their performance on practical applications, namely DOKE. This paradigm relies on a domain knowledge extractor, working in three steps: 1) preparing effective knowledge for the task; 2) selecting the knowledge for each specific sample; and 3) expressing the knowledge in an LLM-understandable way. Then, the extracted knowledge is incorporated through prompts, without any computational cost of model fine-tuning. We instantiate the general paradigm on a widespread application, i.e. recommender systems, where critical item attributes and collaborative filtering signals are incorporated. Experimental results demonstrate that DOKE can substantially improve the performance of LLMs in specific domains.

Jing Yao, Xiaoyuan Yi, Xiting Wang et al.

The rapid advancement of Large Language Models (LLMs) has attracted much attention to value alignment for their responsible development. However, how to define values in this context remains a largely unexplored question. Existing work mainly follows the Helpful, Honest, Harmless principle and specifies values as risk criteria formulated in the AI community, e.g., fairness and privacy protection, suffering from poor clarity, adaptability and transparency. Inspired by basic values in humanity and social science across cultures, this work proposes a novel basic value alignment paradigm and introduces a value space spanned by basic value dimensions. All LLMs' behaviors can be mapped into the space by identifying the underlying values, possessing the potential to address the three challenges. To foster future research, we apply the representative Schwartz's Theory of Basic Values as an initialized example and construct FULCRA, a dataset consisting of 5k (LLM output, value vector) pairs. Our extensive analysis of FULCRA reveals the underlying relation between basic values and LLMs' behaviors, demonstrating that our approach not only covers existing mainstream risks but also anticipates possibly unidentified ones. Additionally, we present an initial implementation of the basic value evaluation and alignment, paving the way for future research in this line.

Yule Wen, Yanzhe Zhang, Jianxun Lian et al. · gatech

LLM agents increasingly act on users' personal information, yet existing privacy defenses remain limited in both design and adaptability. Most prior approaches rely on static or passive defenses, such as prompting and guarding. These paradigms are insufficient for supporting contextual, proactive privacy decisions in multi-step agent execution. We propose Contextualized Defense Instructing (CDI), a new privacy defense paradigm in which an instructor model generates step-specific, context-aware privacy guidance during execution, proactively shaping actions rather than merely constraining or vetoing them. Crucially, CDI is paired with an experience-driven optimization framework that trains the instructor via reinforcement learning (RL), where we convert failure trajectories with privacy violations into learning environments. We formalize baseline defenses and CDI as distinct intervention points in a canonical agent loop, and compare their privacy-helpfulness trade-offs within a unified simulation framework. Results show that our CDI consistently achieves a better balance between privacy preservation (94.2%) and helpfulness (80.6%) than baselines, with superior robustness to adversarial conditions and generalization.

Jing Yao, Xiaoyuan Yi, Xing Xie

The rapid progress in Large Language Models (LLMs) poses potential risks such as generating unethical content. Assessing LLMs' values can help expose their misalignment, but relies on reference-free evaluators, e.g., fine-tuned LLMs or close-source ones like GPT-4, to identify values reflected in generated responses. Nevertheless, these evaluators face two challenges in open-ended value evaluation: they should align with changing human value definitions with minimal annotation, against their own bias (adaptability), and detect varying value expressions and scenarios robustly (generalizability). To handle these challenges, we introduce CLAVE, a novel framework which integrates two complementary LLMs, a large one to extract high-level value concepts from a few human labels, leveraging its extensive knowledge and generalizability, and a smaller one fine-tuned on such concepts to better align with human value understanding. This dual-model approach enables calibration with any value systems using <100 human-labeled samples per value type. Then we present ValEval, a comprehensive dataset comprising 13k+ (text,value,label) tuples across diverse domains, covering three major value systems. We benchmark the capabilities of 12+ popular LLM evaluators and analyze their strengths and weaknesses. Our findings reveal that combining fine-tuned small models and prompt-based large ones serves as a superior balance in value evaluation.

Yuhang Wang, Yanxu Zhu, Chao Kong et al.

As the scaling of Large Language Models (LLMs) has dramatically enhanced their capabilities, there has been a growing focus on the alignment problem to ensure their responsible and ethical use. While existing alignment efforts predominantly concentrate on universal values such as the HHH principle, the aspect of culture, which is inherently pluralistic and diverse, has not received adequate attention. This work introduces a new benchmark, CDEval, aimed at evaluating the cultural dimensions of LLMs. CDEval is constructed by incorporating both GPT-4's automated generation and human verification, covering six cultural dimensions across seven domains. Our comprehensive experiments provide intriguing insights into the culture of mainstream LLMs, highlighting both consistencies and variations across different dimensions and domains. The findings underscore the importance of integrating cultural considerations in LLM development, particularly for applications in diverse cultural settings. Through CDEval, we aim to broaden the horizon of LLM alignment research by including cultural dimensions, thus providing a more holistic framework for the future development and evaluation of LLMs. This benchmark serves as a valuable resource for cultural studies in LLMs, paving the way for more culturally aware and sensitive models.

Xiaoyuan Yi, Jing Yao, Xiting Wang et al.

Big models have greatly advanced AI's ability to understand, generate, and manipulate information and content, enabling numerous applications. However, as these models become increasingly integrated into everyday life, their inherent ethical values and potential biases pose unforeseen risks to society. This paper provides an overview of the risks and challenges associated with big models, surveys existing AI ethics guidelines, and examines the ethical implications arising from the limitations of these models. Taking a normative ethics perspective, we propose a reassessment of recent normative guidelines, highlighting the importance of collaborative efforts in academia to establish a unified and universal AI ethics framework. Furthermore, we investigate the moral inclinations of current mainstream LLMs using the Moral Foundation theory, analyze existing alignment algorithms, and outline the unique challenges encountered in aligning ethical values within them. To address these challenges, we introduce a novel conceptual paradigm for aligning the ethical values of big models and discuss promising research directions for alignment criteria, evaluation, and method, representing an initial step towards the interdisciplinary construction of the ethically aligned AI This paper is a modified English version of our Chinese paper https://crad.ict.ac.cn/cn/article/doi/10.7544/issn1000-1239.202330553, intended to help non-Chinese native speakers better understand our work.

Yan Liu, Xiaoyuan Yi, Xiaokang Chen et al.

The demand for regulating potentially risky behaviors of large language models (LLMs) has ignited research on alignment methods. Since LLM alignment heavily relies on reward models for optimization or evaluation, neglecting the quality of reward models may cause unreliable results or even misalignment. Despite the vital role reward models play in alignment, previous works have consistently overlooked their performance and used off-the-shelf reward models arbitrarily without verification, rendering the reward model ``\emph{an elephant in the room}''. To this end, this work first investigates the quality of the widely-used preference dataset, HH-RLHF, and curates a clean version, CHH-RLHF. Based on CHH-RLHF, we benchmark the accuracy of a broad range of reward models used in previous alignment works, unveiling the unreliability of using them both for optimization and evaluation. Furthermore, we systematically study the impact of reward model quality on alignment performance in three reward utilization paradigms. Extensive experiments reveal that better reward models perform as better human preference proxies. This work aims to awaken people to notice this huge elephant in alignment research. We call attention to the following issues: (1) The reward model needs to be rigorously evaluated, whether for alignment optimization or evaluation. (2) Considering the role of reward models, research efforts should not only concentrate on alignment algorithm, but also on developing more reliable human proxy.

Jaehyeok Lee, Xiaoyuan Yi, Jing Yao et al.

As LLMs are globally deployed, aligning their cultural value orientations is critical for safety and user engagement. However, existing benchmarks face the Construct-Composition-Context ($C^3$) challenge: relying on discriminative, multiple-choice formats that probe value knowledge rather than true orientations, overlook subcultural heterogeneity, and mismatch with real-world open-ended generation. We introduce DOVE, a distributional evaluation framework that directly compares human-written text distributions with LLM-generated outputs. DOVE utilizes a rate-distortion variational optimization objective to construct a compact value-codebook from 10K documents, mapping text into a structured value space to filter semantic noise. Alignment is measured using unbalanced optimal transport, capturing intra-cultural distributional structures and sub-group diversity. Experiments across 12 LLMs show that DOVE achieves superior predictive validity, attaining a 31.56% correlation with downstream tasks, while maintaining high reliability with as few as 500 samples per culture.

Bryan Chen Zhengyu Tan, Zhengyuan Liu, Xiaoyuan Yi et al.

Despite their global prevalence, many Large Language Models (LLMs) are aligned to a monolithic, often Western-centric set of values. This paper investigates the more challenging task of fine-grained value alignment: examining whether LLMs can emulate the distinct cultural values of demographic subgroups. Using Singapore as a case study and the World Values Survey (WVS), we examine the value landscape and show that even state-of-the-art models like GPT-4.1 achieve only 57.4% accuracy in predicting subgroup modal preferences. We construct a dataset of over 20,000 samples to train and evaluate a range of models. We demonstrate that simple fine-tuning on structured numerical preferences yields substantial gains, improving accuracy on unseen, out-of-distribution subgroups by an average of 17.4%. These gains partially transfer to open-ended generation. However, we find significant pre-existing performance biases, where models better emulate young, male, Chinese, and Christian personas. Furthermore, while fine-tuning improves average performance, it widens the disparity between subgroups when measured by distance-aware metrics. Our work offers insights into the limits and fairness implications of subgroup-level cultural alignment.

Xiangxu Zhang, Jiamin Wang, Qinlin Zhao et al.

As LLMs become increasingly integrated into human society, evaluating their orientations on human values from social science has drawn growing attention. Nevertheless, it is still unclear why human values matter for LLMs, especially in LLM-based multi-agent systems, where group-level failures may accumulate from individually misaligned actions. We ask whether misalignment with human values alters the collective behavior of LLM agents and what changes it induces? In this work, we introduce CIVA, a controlled multi-agent environment grounded in social science theories, where LLM agents form a community and autonomously communicate, explore, and compete for resources, enabling systematic manipulation of value prevalence and behavioral analysis. Through comprehensive simulation experiments, we reveal three key findings. (1) We identify several structurally critical values that substantially shape the community's collective dynamics, including those diverging from LLMs' original orientations. Triggered by the misspecification of these values, we (2) detect system failure modes, e.g., catastrophic collapse, at the macro level, and (3) observe emergent behaviors like deception and power-seeking at the micro level. These results offer quantitative evidence that human values are essential for collective outcomes in LLMs and motivate future multi-agent value alignment.

Zhenghao Liu, Xingsheng Zhu, Tianshuo Zhou et al.

With the rapid advancement of Multi-modal Large Language Models (MLLMs), their capability in understanding both images and text has greatly improved. However, their potential for leveraging multi-modal contextual information in Retrieval-Augmented Generation (RAG) remains largely underexplored. To address this gap, this paper introduces Multi-Modal Retrieval-Augmented Generation (M$^2$RAG), a benchmark designed to evaluate the effectiveness of Multi-modal Large Language Models in leveraging knowledge from multi-modal retrieval documents. The benchmark comprises four tasks: image captioning, multi-modal question answering, multi-modal fact verification, and image reranking. All tasks are set in an open-domain setting, requiring RAG models to retrieve query-relevant information from a multi-modal document collection and use it as contextual input for RAG modeling. To enhance the context utilization capabilities of MLLMs, we also introduce Multi-Modal Retrieval-Augmented Instruction Tuning (MM-RAIT), an instruction tuning method that optimizes MLLMs within multi-modal contexts. Our experiments demonstrate the effectiveness of MM-RAIT by significantly improving the quality of responses generated by different RAG models, outperforming MiniCPM-V 2.6 and Qwen2-VL with 34% and 33% gains, respectively. All data and code are available at https://github.com/NEUIR/M2RAG.

Muhua Huang, Qinlin Zhao, Xiaoyuan Yi et al.

As Large Language Models (LLM) based multi-agent systems become increasingly prevalent, the collective behaviors, e.g., collective intelligence, of such artificial communities have drawn growing attention. This work aims to answer a fundamental question: How does diversity of values shape the collective behavior of AI communities? Using naturalistic value elicitation grounded in the prevalent Schwartz's Theory of Basic Human Values, we constructed multi-agent simulations where communities with varying numbers of agents engaged in open-ended interactions and constitution formation. The results show that value diversity enhances value stability, fosters emergent behaviors, and brings more creative principles developed by the agents themselves without external guidance. However, these effects also show diminishing returns: extreme heterogeneity induces instability. This work positions value diversity as a new axis of future AI capability, bridging AI ability and sociological studies of institutional emergence.

Shaohua Duan, Xinze Li, Zhenghao Liu et al.

Long-context modeling is critical for a wide range of real-world tasks, including long-context question answering, summarization, and complex reasoning tasks. Recent studies have explored fine-tuning Large Language Models (LLMs) with synthetic data to enhance their long-context capabilities. However, the effectiveness of such approaches is often limited by the low diversity and factual inconsistencies in the generated data. To address these challenges, we propose LongMab-PO, a novel framework that leverages a Multi-Armed Bandit (MAB) rollout strategy to identify the most informative chunks from the given long context for sampling high-quality and diverse responses and constructing preference data pairs for Direct Preference Optimization (DPO) training. Specifically, we treat context chunks as arms of MAB, select chunks based on their expected reward scores to input into LLMs to generate responses, and iteratively update these scores based on reward feedback. This exploration and exploitation process enables the model to focus on the most relevant context segments, thereby generating and collecting high-quality and diverse responses. Finally, we collect these generated responses from the rollout process and apply the DPO method to further optimize the LLM. Experimental results show that LongMab-PO significantly improves the diversity and quality of preference data pairs, achieving state-of-the-art performance on long-context reasoning benchmarks. All code and data will be released on https://github.com/NEUIR/LongMab-PO.

Pengcheng Huang, Zhenghao Liu, Yukun Yan et al.

Large language models (LLMs) integrated with retrieval-augmented generation (RAG) have improved factuality by grounding outputs in external evidence. However, they remain susceptible to unfaithful generation, where outputs contradict retrieved context despite its relevance and accuracy. Existing approaches aiming to improve faithfulness primarily focus on enhancing the utilization of external context, but often overlook the persistent influence of internal parametric knowledge during generation. In this work, we investigate the internal mechanisms behind unfaithful generation and identify a subset of mid-to-deep feed-forward networks (FFNs) that are disproportionately activated in such cases. Building on this insight, we propose Parametric Knowledge Muting through FFN Suppression (ParamMute), a framework that improves contextual faithfulness by suppressing the activation of unfaithfulness-associated FFNs and calibrating the model toward retrieved knowledge. To evaluate our approach, we introduce CoFaithfulQA, a benchmark specifically designed to evaluate faithfulness in scenarios where internal knowledge conflicts with accurate external evidence. Experimental results show that ParamMute significantly enhances faithfulness across both CoFaithfulQA and the established ConFiQA benchmark, achieving substantial reductions in reliance on parametric memory. These findings underscore the importance of mitigating internal knowledge dominance and provide a new direction for improving LLM trustworthiness in RAG. All codes are available at https://github.com/OpenBMB/ParamMute.

Jing Yao, Xiaoyuan Yi, Shitong Duan et al.

As Large Language Models (LLMs) achieve remarkable breakthroughs, aligning their values with humans has become imperative for their responsible development and customized applications. However, there still lack evaluations of LLMs values that fulfill three desirable goals. (1) Value Clarification: We expect to clarify the underlying values of LLMs precisely and comprehensively, while current evaluations focus narrowly on safety risks such as bias and toxicity. (2) Evaluation Validity: Existing static, open-source benchmarks are prone to data contamination and quickly become obsolete as LLMs evolve. Additionally, these discriminative evaluations uncover LLMs' knowledge about values, rather than valid assessments of LLMs' behavioral conformity to values. (3) Value Pluralism: The pluralistic nature of human values across individuals and cultures is largely ignored in measuring LLMs value alignment. To address these challenges, we presents the Value Compass Benchmarks, with three correspondingly designed modules. It (i) grounds the evaluation on motivationally distinct \textit{basic values to clarify LLMs' underlying values from a holistic view; (ii) applies a \textit{generative evolving evaluation framework with adaptive test items for evolving LLMs and direct value recognition from behaviors in realistic scenarios; (iii) propose a metric that quantifies LLMs alignment with a specific value as a weighted sum over multiple dimensions, with weights determined by pluralistic values.

Huanhuan Ma, Haisong Gong, Xiaoyuan Yi et al.

Recent advancements in Large Language Models (LLMs) have led to their increasing integration into human life. With the transition from mere tools to human-like assistants, understanding their psychological aspects-such as emotional tendencies and personalities-becomes essential for ensuring their trustworthiness. However, current psychological evaluations of LLMs, often based on human psychological assessments like the BFI, face significant limitations. The results from these approaches often lack reliability and have limited validity when predicting LLM behavior in real-world scenarios. In this work, we introduce a novel evaluation instrument specifically designed for LLMs, called Core Sentiment Inventory (CSI). CSI is a bilingual tool, covering both English and Chinese, that implicitly evaluates models' sentiment tendencies, providing an insightful psychological portrait of LLM across three dimensions: optimism, pessimism, and neutrality. Through extensive experiments, we demonstrate that: 1) CSI effectively captures nuanced emotional patterns, revealing significant variation in LLMs across languages and contexts; 2) Compared to current approaches, CSI significantly improves reliability, yielding more consistent results; and 3) The correlation between CSI scores and the sentiment of LLM's real-world outputs exceeds 0.85, demonstrating its strong validity in predicting LLM behavior. We make CSI public available via: https://github.com/dependentsign/CSI.

Han Jiang, Dongyao Zhu, Zhihua Wei et al.

In-Context Learning has shown great potential for aligning Large Language Models (LLMs) with human values, helping reduce harmful outputs and accommodate diverse preferences without costly post-training, known as In-Context Alignment (ICA). However, LLMs' comprehension of input prompts remains agnostic, limiting ICA's ability to address value tensions--human values are inherently pluralistic, often imposing conflicting demands, e.g., stimulation vs. tradition. Current ICA methods therefore face the Instruction Bottleneck challenge, where LLMs struggle to reconcile multiple intended values within a single prompt, leading to incomplete or biased alignment. To address this, we propose PICACO, a novel pluralistic ICA method. Without fine-tuning, PICACO optimizes a meta-instruction that navigates multiple values to better elicit LLMs' understanding of them and improve their alignment. This is achieved by maximizing the total correlation between specified values and LLM responses, theoretically reinforcing value correlation while reducing distractive noise, resulting in effective value instructions. Extensive experiments on five value sets show that PICACO works well with both black-box and open-source LLMs, outperforms several recent strong baselines, and achieves a better balance across up to 8 distinct values.

Xin Dai, Buqiang Xu, Zhenghao Liu et al.

Legal Artificial Intelligence (LegalAI) has achieved notable advances in automating judicial decision-making with the support of Large Language Models (LLMs). However, existing legal LLMs still struggle to generate reliable and interpretable reasoning processes. They often default to fast-thinking behavior by producing direct answers without explicit multi-step reasoning, limiting their effectiveness in complex legal scenarios that demand rigorous justification. To address this challenge, we propose Legal$Δ$, a reinforcement learning framework designed to enhance legal reasoning through chain-of-thought guided information gain. During training, Legal$Δ$ employs a dual-mode input setup-comprising direct answer and reasoning-augmented modes-and maximizes the information gain between them. This encourages the model to acquire meaningful reasoning patterns rather than generating superficial or redundant explanations. Legal$Δ$ follows a two-stage approach: (1) distilling latent reasoning capabilities from a powerful Large Reasoning Model (LRM), DeepSeek-R1, and (2) refining reasoning quality via differential comparisons, combined with a multidimensional reward mechanism that assesses both structural coherence and legal-domain specificity. Experimental results on multiple legal reasoning tasks demonstrate that Legal$Δ$ outperforms strong baselines in both accuracy and interpretability. It consistently produces more robust and trustworthy legal judgments without relying on labeled preference data. All code and data will be released at https://github.com/NEUIR/LegalDelta.

Yanxu Zhu, Shitong Duan, Xiangxu Zhang et al.

Recently Multimodal Large Language Models (MLLMs) have achieved considerable advancements in vision-language tasks, yet produce potentially harmful or untrustworthy content. Despite substantial work investigating the trustworthiness of language models, MMLMs' capability to act honestly, especially when faced with visually unanswerable questions, remains largely underexplored. This work presents the first systematic assessment of honesty behaviors across various MLLMs. We ground honesty in models' response behaviors to unanswerable visual questions, define four representative types of such questions, and construct MoHoBench, a large-scale MMLM honest benchmark, consisting of 12k+ visual question samples, whose quality is guaranteed by multi-stage filtering and human verification. Using MoHoBench, we benchmarked the honesty of 28 popular MMLMs and conducted a comprehensive analysis. Our findings show that: (1) most models fail to appropriately refuse to answer when necessary, and (2) MMLMs' honesty is not solely a language modeling issue, but is deeply influenced by visual information, necessitating the development of dedicated methods for multimodal honesty alignment. Therefore, we implemented initial alignment methods using supervised and preference learning to improve honesty behavior, providing a foundation for future work on trustworthy MLLMs. Our data and code can be found at https://github.com/yanxuzhu/MoHoBench.

Buqiang Xu, Xin Dai, Zhenghao Liu et al.

Legal Judgment Prediction (LJP) is a fundamental task of legal artificial intelligence, aiming to automatically predict the judgment outcomes of legal cases. Existing LJP models primarily focus on identifying legal triggers within criminal fact descriptions by contrastively training language models. However, these LJP models overlook the importance of learning to effectively distinguish subtle differences among judgments, which is crucial for producing more accurate predictions. In this paper, we propose LegalDuet, which continuously pretrains language models to learn a more tailored embedding space for representing legal cases. Specifically, LegalDuet designs a dual-view mechanism to continuously pretrain language models: 1) Law Case Clustering retrieves similar cases as hard negatives and employs contrastive training to differentiate among confusing cases; 2) Legal Decision Matching aims to identify legal clues within criminal fact descriptions to align them with the chain of reasoning that contains the correct legal decision. Our experiments on the CAIL2018 dataset demonstrate the effectiveness of LegalDuet. Further analysis reveals that LegalDuet improves the ability of pretrained language models to distinguish confusing criminal charges by reducing prediction uncertainty and enhancing the separability of criminal charges. The experiments demonstrate that LegalDuet produces a more concentrated and distinguishable embedding space, effectively aligning criminal facts with corresponding legal decisions. The code is available at https://github.com/NEUIR/LegalDuet.

Wenhao Li, Fanchao Qi, Maosong Sun et al.

Poetry is one of the most important art forms of human languages. Recently many studies have focused on incorporating some linguistic features of poetry, such as style and sentiment, into its understanding or generation system. However, there is no focus on understanding or evaluating the semantics of poetry. Therefore, we propose a novel task to assess a model's semantic understanding of poetry by poem matching. Specifically, this task requires the model to select one line of Chinese classical poetry among four candidates according to the modern Chinese translation of a line of poetry. To construct this dataset, we first obtain a set of parallel data of Chinese classical poetry and modern Chinese translation. Then we retrieve similar lines of poetry with the lines in a poetry corpus as negative choices. We name the dataset Chinese Classical Poetry Matching Dataset (CCPM) and release it at https://github.com/THUNLP-AIPoet/CCPM. We hope this dataset can further enhance the study on incorporating deep semantics into the understanding and generation system of Chinese classical poetry. We also preliminarily run two variants of BERT on this dataset as the baselines for this dataset.

Zhenghao Liu, Xiaoyuan Yi, Maosong Sun et al.

Grammatical Error Correction (GEC) aims to correct writing errors and help language learners improve their writing skills. However, existing GEC models tend to produce spurious corrections or fail to detect lots of errors. The quality estimation model is necessary to ensure learners get accurate GEC results and avoid misleading from poorly corrected sentences. Well-trained GEC models can generate several high-quality hypotheses through decoding, such as beam search, which provide valuable GEC evidence and can be used to evaluate GEC quality. However, existing models neglect the possible GEC evidence from different hypotheses. This paper presents the Neural Verification Network (VERNet) for GEC quality estimation with multiple hypotheses. VERNet establishes interactions among hypotheses with a reasoning graph and conducts two kinds of attention mechanisms to propagate GEC evidence to verify the quality of generated hypotheses. Our experiments on four GEC datasets show that VERNet achieves state-of-the-art grammatical error detection performance, achieves the best quality estimation results, and significantly improves GEC performance by reranking hypotheses. All data and source codes are available at https://github.com/thunlp/VERNet.

Zhaowei Zhang, Xiaohan Liu, Xuekai Zhu et al.

Reinforcement learning with verifiable rewards (RLVR) has achieved remarkable success in logical reasoning tasks, yet whether large language model (LLM) alignment requires fundamentally different approaches remains unclear. Given the apparent tolerance for multiple valid responses in moral reasoning, a natural hypothesis is that alignment tasks inherently require diversity-seeking distribution-matching algorithms rather than reward-maximizing policy-based methods. We conduct the first comprehensive empirical study comparing both paradigms on MoReBench. To enable stable RLVR training, we build a rubric-grounded reward pipeline by training a Qwen3-1.7B judge model. Contrary to our hypothesis, we find that distribution-matching approaches do not demonstrate significant advantages over reward-maximizing methods as expected on alignment tasks. Through semantic visualization mapping high-reward responses to semantic space, we demonstrate that moral reasoning exhibits more concentrated high-reward distributions than mathematical reasoning, where diverse solution strategies yield similarly high rewards. This counter-intuitive finding explains why mode-seeking optimization proves equally or more effective for alignment tasks. Our results suggest that alignment tasks do not inherently require diversity-preserving algorithms, and standard reward-maximizing RLVR methods can effectively transfer to moral reasoning without explicit diversity mechanisms.

Xukun Liu, Zhiyuan Peng, Xiaoyuan Yi et al.

While achieving remarkable progress in a broad range of tasks, large language models (LLMs) remain significantly limited in properly using massive external tools. Existing in-context learning approaches simply format tools into a list of plain text descriptions and input them to LLMs, from which, LLMs generate a sequence of tool calls to solve problems step by step. Such a paradigm ignores the intrinsic dependency between tools and offloads all reasoning loads to LLMs, making them restricted to a limited number of specifically designed tools. It thus remains challenging for LLMs to operate on a library of massive tools, casting a great limitation when confronted with real-world scenarios. This paper proposes ToolNet, a plug-and-play framework that scales up the number of tools to thousands with a moderate increase in token consumption. ToolNet organizes tools into a directed graph. Each node represents a tool, and weighted edges denote tool transition. Starting from an initial tool node, an LLM navigates in the graph by iteratively choosing the next one from its successors until the task is resolved. Extensive experiments show that ToolNet can achieve impressive results in challenging multi-hop tool learning datasets and is resilient to tool failures.

Xinpeng Wang, Xiaoyuan Yi, Han Jiang et al.

Warning: this paper includes model outputs showing offensive content. Recent large-scale Visual-Language Generative Models (VLGMs) have achieved unprecedented improvement in multimodal image/text generation. However, these models might also generate toxic content, e.g., offensive text and pornography images, raising significant ethical risks. Despite exhaustive studies on toxic degeneration of language models, this problem remains largely unexplored within the context of visual-language generation. This work delves into the propensity for toxicity generation and susceptibility to toxic data across various VLGMs. For this purpose, we built ToViLaG, a dataset comprising 32K co-toxic/mono-toxic text-image pairs and 1K innocuous but evocative text that tends to stimulate toxicity. Furthermore, we propose WInToRe, a novel toxicity metric tailored to visual-language generation, which theoretically reflects different aspects of toxicity considering both input and output. On such a basis, we benchmarked the toxicity of a diverse spectrum of VLGMs and discovered that some models do more evil than expected while some are more vulnerable to infection, underscoring the necessity of VLGMs detoxification. Therefore, we develop an innovative bottleneck-based detoxification method. Our method could reduce toxicity while maintaining comparable generation quality, providing a promising initial solution to this line of research.

Xinpeng Wang, Shitong Duan, Xiaoyuan Yi et al.

Big models have achieved revolutionary breakthroughs in the field of AI, but they might also pose potential concerns. Addressing such concerns, alignment technologies were introduced to make these models conform to human preferences and values. Despite considerable advancements in the past year, various challenges lie in establishing the optimal alignment strategy, such as data cost and scalable oversight, and how to align remains an open question. In this survey paper, we comprehensively investigate value alignment approaches. We first unpack the historical context of alignment tracing back to the 1920s (where it comes from), then delve into the mathematical essence of alignment (what it is), shedding light on the inherent challenges. Following this foundation, we provide a detailed examination of existing alignment methods, which fall into three categories: Reinforcement Learning, Supervised Fine-Tuning, and In-context Learning, and demonstrate their intrinsic connections, strengths, and limitations, helping readers better understand this research area. In addition, two emerging topics, personal alignment, and multimodal alignment, are also discussed as novel frontiers in this field. Looking forward, we discuss potential alignment paradigms and how they could handle remaining challenges, prospecting where future alignment will go.

Pablo Biedma, Xiaoyuan Yi, Linus Huang et al.

Recent advancements in Large Language Models (LLMs) have revolutionized the AI field but also pose potential safety and ethical risks. Deciphering LLMs' embedded values becomes crucial for assessing and mitigating their risks. Despite extensive investigation into LLMs' values, previous studies heavily rely on human-oriented value systems in social sciences. Then, a natural question arises: Do LLMs possess unique values beyond those of humans? Delving into it, this work proposes a novel framework, ValueLex, to reconstruct LLMs' unique value system from scratch, leveraging psychological methodologies from human personality/value research. Based on Lexical Hypothesis, ValueLex introduces a generative approach to elicit diverse values from 30+ LLMs, synthesizing a taxonomy that culminates in a comprehensive value framework via factor analysis and semantic clustering. We identify three core value dimensions, Competence, Character, and Integrity, each with specific subdimensions, revealing that LLMs possess a structured, albeit non-human, value system. Based on this system, we further develop tailored projective tests to evaluate and analyze the value inclinations of LLMs across different model sizes, training methods, and data sources. Our framework fosters an interdisciplinary paradigm of understanding LLMs, paving the way for future AI alignment and regulation.

HyunJin Kim, Xiaoyuan Yi, Jing Yao et al.

The emergence of large language models (LLMs) has sparked the possibility of about Artificial Superintelligence (ASI), a hypothetical AI system surpassing human intelligence. However, existing alignment paradigms struggle to guide such advanced AI systems. Superalignment, the alignment of AI systems with human values and safety requirements at superhuman levels of capability aims to addresses two primary goals -- scalability in supervision to provide high-quality guidance signals and robust governance to ensure alignment with human values. In this survey, we examine scalable oversight methods and potential solutions for superalignment. Specifically, we explore the concept of ASI, the challenges it poses, and the limitations of current alignment paradigms in addressing the superalignment problem. Then we review scalable oversight methods for superalignment. Finally, we discuss the key challenges and propose pathways for the safe and continual improvement of ASI systems. By comprehensively reviewing the current literature, our goal is provide a systematical introduction of existing methods, analyze their strengths and limitations, and discuss potential future directions.

Shitong Duan, Xiaoyuan Yi, Peng Zhang et al.

Large language models (LLMs) have revolutionized the role of AI, yet pose potential social risks. To steer LLMs towards human preference, alignment technologies have been introduced and gained increasing attention. Nevertheless, existing methods heavily rely on high-quality positive-negative training pairs, suffering from noisy positive responses that are barely distinguishable from negative ones. Given recent LLMs' proficiency in generating helpful responses, this work pivots towards a new research question: can we achieve alignment using solely human-annotated negative samples, preserving helpfulness while reducing harmfulness? For this purpose, we propose Distributional Dispreference Optimization (D$^2$O), which maximizes the discrepancy between dispreferred responses and the generated non-negative ones. In this way, D$^2$O effectively eschews harmful information without incorporating noisy positive samples, while avoiding collapse using self-generated responses as anchors. We demonstrate that D$^2$O can be regarded as learning a distributional preference model reflecting human dispreference against negative responses, which is theoretically an upper bound of the instance-level DPO. Extensive experiments manifest that our method achieves comparable generation quality and surpasses the latest strong baselines in producing less harmful and more informative responses with better training stability and faster convergence.

Xingqi Wang, Xiaoyuan Yi, Xing Xie et al. · tsinghua

Recent advancements in diffusion models trained on large-scale data have enabled the generation of indistinguishable human-level images, yet they often produce harmful content misaligned with human values, e.g., social bias, and offensive content. Despite extensive research on Large Language Models (LLMs), the challenge of Text-to-Image (T2I) model alignment remains largely unexplored. Addressing this problem, we propose LiVO (Lightweight Value Optimization), a novel lightweight method for aligning T2I models with human values. LiVO only optimizes a plug-and-play value encoder to integrate a specified value principle with the input prompt, allowing the control of generated images over both semantics and values. Specifically, we design a diffusion model-tailored preference optimization loss, which theoretically approximates the Bradley-Terry model used in LLM alignment but provides a more flexible trade-off between image quality and value conformity. To optimize the value encoder, we also develop a framework to automatically construct a text-image preference dataset of 86k (prompt, aligned image, violating image, value principle) samples. Without updating most model parameters and through adaptive value selection from the input prompt, LiVO significantly reduces harmful outputs and achieves faster convergence, surpassing several strong baselines and taking an initial step towards ethically aligned T2I models.

Jing Yao, Xiaoyuan Yi, Jindong Wang et al.

As Large Language Models (LLMs) more deeply integrate into human life across various regions, aligning them with pluralistic cultures is crucial for improving user experience and mitigating cultural conflicts. Existing approaches develop culturally aligned LLMs primarily through fine-tuning with massive carefully curated culture-specific corpora. Nevertheless, inspired by culture theories, we identify two key challenges faced by these datasets: (1) Representativeness: These corpora fail to fully capture the target culture's core characteristics with redundancy, causing computation waste; (2) Distinctiveness: They struggle to distinguish the unique nuances of a given culture from shared patterns across other relevant ones, hindering precise cultural modeling. To handle these challenges, we introduce CAReDiO, a novel cultural data construction framework. Specifically, CAReDiO utilizes powerful LLMs to automatically generate cultural conversation data, where both the queries and responses are further optimized by maximizing representativeness and distinctiveness. Using CAReDiO, we construct a small yet effective dataset, covering five cultures, and compare it with several recent cultural corpora. Extensive experiments demonstrate that our method generates more effective data and enables cultural alignment with as few as 100 training samples, enhancing both performance and efficiency.

Han Jiang, Pengda Wang, Xiaoyuan Yi et al.

Social sciences have accumulated a rich body of theories and methodologies for investigating the human mind and behaviors, while offering valuable insights into the design and understanding of Artificial Intelligence (AI) systems. Focusing on psychology as a prominent case, this study explores the interdisciplinary synergy between AI and the field by analyzing 1,006 LLM-related papers published in premier AI venues between 2023 and 2025, along with the 2,544 psychology publications they cite. Through our analysis, we identify key patterns of interdisciplinary integration, locate the psychology domains most frequently referenced, and highlight areas that remain underexplored. We further examine how psychology theories/frameworks are operationalized and interpreted, identify common types of misapplication, and offer guidance for more effective incorporation. Our work provides a comprehensive map of interdisciplinary engagement between AI and psychology, thereby facilitating deeper collaboration and advancing AI systems.

Yuqing Lan, Zhenghao Liu, Yu Gu et al.

Fact verification tasks aim to identify the integrity of textual contents according to the truthful corpus. Existing fact verification models usually build a fully connected reasoning graph, which regards claim-evidence pairs as nodes and connects them with edges. They employ the graph to propagate the semantics of the nodes. Nevertheless, the noisy nodes usually propagate their semantics via the edges of the reasoning graph, which misleads the semantic representations of other nodes and amplifies the noise signals. To mitigate the propagation of noisy semantic information, we introduce a Confidential Graph Attention Network (CO-GAT), which proposes a node masking mechanism for modeling the nodes. Specifically, CO-GAT calculates the node confidence score by estimating the relevance between the claim and evidence pieces. Then, the node masking mechanism uses the node confidence scores to control the noise information flow from the vanilla node to the other graph nodes. CO-GAT achieves a 73.59% FEVER score on the FEVER dataset and shows the generalization ability by broadening the effectiveness to the science-specific domain.

Han Jiang, Susu Zhang, Xiaoyuan Yi et al.

AI evaluations have become the primary evidence for deploying generative AI systems across high-stakes domains. However, current evaluation paradigms often exhibit systemic validity failures. These issues, ranging from unjustified design choices to misaligned metrics, remain intractable without a principled framework for gathering validity evidence and conducting granular diagnostic analysis. In this position paper, we argue that item-level AI benchmark data is essential for establishing a rigorous science of AI evaluation. Item-level analysis enables fine-grained diagnostics and principled validation of benchmarks. We substantiate this position by dissecting current validity failures and revisiting evaluation paradigms across computer science and psychometrics. Through illustrative analyses of item properties and latent constructs, we demonstrate the unique insights afforded by item-level data. To catalyze community-wide adoption, we introduce OpenEval, a growing repository of item-level benchmark data designed supporting evidence-centered AI evaluation.

Bolun Zhang, Linzhuo Li, Yunqi Chen et al.

Large language models (LLMs) are rapidly being integrated into computational social science research, yet their blackboxed training and designed stochastic elements in inference pose unique challenges for scientific inquiry. This article argues that applying LLMs to social scientific tasks requires explicit assessment of uncertainty-an expectation long established in both quantitative methodology in the social sciences and machine learning. We introduce a unified framework for evaluating LLM uncertainty along two dimensions: the task type (T), which distinguishes between classification, short-form, and long-form generation, and the validation type (V), which captures the availability of reference data or evaluative criteria. Drawing from both computer science and social science literature, we map existing uncertainty quantification (UQ) methods to this T-V typology and offer practical recommendations for researchers. Our framework provides both a methodological safeguard and a practical guide for integrating LLMs into rigorous social science research.

Shanlin Zhou, Xinpeng Wang, Jianxun Lian et al.

Trained on diverse human-authored texts, Large Language Models (LLMs) unlocked the potential for Creative Natural Language Generation (CNLG), benefiting various applications like advertising and storytelling. Nevertheless, CNLG still remains difficult due to two main challenges. (1) Multi-objective flexibility: user requirements are often personalized, fine-grained, and pluralistic, which LLMs struggle to satisfy simultaneously; (2) Interpretive complexity: beyond generation, creativity also involves understanding and interpreting implicit meaning to enhance users' perception. These challenges significantly limit current methods, especially in short-form text generation, in generating creative and insightful content. To address this, we focus on Chinese baby naming, a representative short-form CNLG task requiring adherence to explicit user constraints (e.g., length, semantics, anthroponymy) while offering meaningful aesthetic explanations. We propose NAMeGEn, a novel multi-agent optimization framework that iteratively alternates between objective extraction, name generation, and evaluation to meet diverse requirements and generate accurate explanations. To support this task, we further construct a classical Chinese poetry corpus with 17k+ poems to enhance aesthetics, and introduce CBNames, a new benchmark with tailored metrics. Extensive experiments demonstrate that NAMeGEn effectively generates creative names that meet diverse, personalized requirements while providing meaningful explanations, outperforming six baseline methods spanning various LLM backbones without any training.

Hanze Guo, Jing Yao, Xiao Zhou et al.

As large language models (LLMs) become increasingly integrated into applications serving users across diverse cultures, communities and demographics, it is critical to align LLMs with pluralistic human values beyond average principles (e.g., HHH). In psychological and social value theories such as Schwartz's Value Theory, pluralistic values are represented by multiple value dimensions paired with various priorities. However, existing methods encounter two challenges when aligning with such fine-grained value objectives: 1) they often treat multiple values as independent and equally important, ignoring their interdependence and relative priorities (value complexity); 2) they struggle to precisely control nuanced value priorities, especially those underrepresented ones (value steerability). To handle these challenges, we propose COUPLE, a COUnterfactual reasoning framework for PLuralistic valuE alignment. It introduces a structural causal model (SCM) to feature complex interdependency and prioritization among features, as well as the causal relationship between high-level value dimensions and behaviors. Moreover, it applies counterfactual reasoning to generate outputs aligned with any desired value objectives. Benefitting from explicit causal modeling, COUPLE also provides better interpretability. We evaluate COUPLE on two datasets with different value systems and demonstrate that COUPLE advances other baselines across diverse types of value objectives.

Weihua Zheng, Zhengyuan Liu, Tanmoy Chakraborty et al.

Large language models (LLMs) are now used worldwide, yet their multimodal understanding and reasoning often degrade outside Western, high-resource settings. We propose MMA-ASIA, a comprehensive framework to evaluate LLMs' cultural awareness with a focus on Asian contexts. MMA-ASIA centers on a human-curated, multilingual, and multimodally aligned multiple-choice benchmark covering 8 Asian countries and 10 languages, comprising 27,000 questions; over 79 percent require multi-step reasoning grounded in cultural context, moving beyond simple memorization. To our knowledge, this is the first dataset aligned at the input level across three modalities: text, image (visual question answering), and speech. This enables direct tests of cross-modal transfer. Building on this benchmark, we propose a five-dimensional evaluation protocol that measures: (i) cultural-awareness disparities across countries, (ii) cross-lingual consistency, (iii) cross-modal consistency, (iv) cultural knowledge generalization, and (v) grounding validity. To ensure rigorous assessment, a Cultural Awareness Grounding Validation Module detects "shortcut learning" by checking whether the requisite cultural knowledge supports correct answers. Finally, through comparative model analysis, attention tracing, and an innovative Vision-ablated Prefix Replay (VPR) method, we probe why models diverge across languages and modalities, offering actionable insights for building culturally reliable multimodal LLMs.

Ziyu Chen, Junfei Sun, Chenxi Li et al.

Identifying human morals and values embedded in language is essential to empirical studies of communication. However, researchers often face substantial difficulty navigating the diversity of theoretical frameworks and data available for their analysis. Here, we contribute MoVa, a well-documented suite of resources for generalizable classification of human morals and values, consisting of (1) 16 labeled datasets and benchmarking results from four theoretically-grounded frameworks; (2) a lightweight LLM prompting strategy that outperforms fine-tuned models across multiple domains and frameworks; and (3) a new application that helps evaluate psychological surveys. In practice, we specifically recommend a classification strategy, all@once, that scores all related concepts simultaneously, resembling the well-known multi-label classifier chain. The data and methods in MoVa can facilitate many fine-grained interpretations of human and machine communication, with potential implications for the alignment of machine behavior.

Eoin O'Doherty, Nicole Weinrauch, Andrew Talone et al.

Artificial intelligence (AI) is advancing at a pace that raises urgent questions about how to align machine decision-making with human moral values. This working paper investigates how leading AI systems prioritize moral outcomes and what this reveals about the prospects for human-AI symbiosis. We address two central questions: (1) What moral values do state-of-the-art large language models (LLMs) implicitly favour when confronted with dilemmas? (2) How do differences in model architecture, cultural origin, and explainability affect these moral preferences? To explore these questions, we conduct a quantitative experiment with six LLMs, ranking and scoring outcomes across 18 dilemmas representing five moral frameworks. Our findings uncover strikingly consistent value biases. Across all models, Care and Virtue values outcomes were rated most moral, while libertarian choices were consistently penalized. Reasoning-enabled models exhibited greater sensitivity to context and provided richer explanations, whereas non-reasoning models produced more uniform but opaque judgments. This research makes three contributions: (i) Empirically, it delivers a large-scale comparison of moral reasoning across culturally distinct LLMs; (ii) Theoretically, it links probabilistic model behaviour with underlying value encodings; (iii) Practically, it highlights the need for explainability and cultural awareness as critical design principles to guide AI toward a transparent, aligned, and symbiotic future.