Yang Deng

Boyi Deng, Wenjie Wang, Fuli Feng et al.

Large language models (LLMs) are susceptible to red teaming attacks, which can induce LLMs to generate harmful content. Previous research constructs attack prompts via manual or automatic methods, which have their own limitations on construction cost and quality. To address these issues, we propose an integrated approach that combines manual and automatic methods to economically generate high-quality attack prompts. Specifically, considering the impressive capabilities of newly emerged LLMs, we propose an attack framework to instruct LLMs to mimic human-generated prompts through in-context learning. Furthermore, we propose a defense framework that fine-tunes victim LLMs through iterative interactions with the attack framework to enhance their safety against red teaming attacks. Extensive experiments on different LLMs validate the effectiveness of our proposed attack and defense frameworks. Additionally, we release a series of attack prompts datasets named SAP with varying sizes, facilitating the safety evaluation and enhancement of more LLMs. Our code and dataset is available on https://github.com/Aatrox103/SAP .

Pengbo Liu, Yang Deng, Ce Wang et al.

In the paper, we present an approach for learning a single model that universally segments 33 anatomical structures, including vertebrae, pelvic bones, and abdominal organs. Our model building has to address the following challenges. Firstly, while it is ideal to learn such a model from a large-scale, fully-annotated dataset, it is practically hard to curate such a dataset. Thus, we resort to learn from a union of multiple datasets, with each dataset containing the images that are partially labeled. Secondly, along the line of partial labelling, we contribute an open-source, large-scale vertebra segmentation dataset for the benefit of spine analysis community, CTSpine1K, boasting over 1,000 3D volumes and over 11K annotated vertebrae. Thirdly, in a 3D medical image segmentation task, due to the limitation of GPU memory, we always train a model using cropped patches as inputs instead a whole 3D volume, which limits the amount of contextual information to be learned. To this, we propose a cross-patch transformer module to fuse more information in adjacent patches, which enlarges the aggregated receptive field for improved segmentation performance. This is especially important for segmenting, say, the elongated spine. Based on 7 partially labeled datasets that collectively contain about 2,800 3D volumes, we successfully learn such a universal model. Finally, we evaluate the universal model on multiple open-source datasets, proving that our model has a good generalization performance and can potentially serve as a solid foundation for downstream tasks.

Weixiang Zhao, Yichen Zhang, Yingshuo Wang et al. · cmu

Experience-driven self-evolution has emerged as a promising paradigm for improving the autonomy of large language model agents, yet its reliance on self-curated experience introduces underexplored safety risks. In this study, we investigate how experience accumulation and utilization in self-evolving agents affect safety performance across web-based and embodied environments. Notably, experience gathered solely from benign tasks can still compromise safety in high-risk scenarios. Further analysis attributes this degradation to the execution-oriented nature of accumulated experience, which reinforces agents' tendency to act rather than refuse. In more realistic settings where agents encounter both benign and harmful tasks, refusal-related experience mitigates safety decline but induces over-refusal, revealing a fundamental safety-utility trade-off. Overall, our findings expose inherent limitations of current self-evolving agents and call for more principled strategies to ensure safe and reliable adaptation.

Wenxuan Zhang, Xin Li, Yang Deng et al.

As an important fine-grained sentiment analysis problem, aspect-based sentiment analysis (ABSA), aiming to analyze and understand people's opinions at the aspect level, has been attracting considerable interest in the last decade. To handle ABSA in different scenarios, various tasks are introduced for analyzing different sentiment elements and their relations, including the aspect term, aspect category, opinion term, and sentiment polarity. Unlike early ABSA works focusing on a single sentiment element, many compound ABSA tasks involving multiple elements have been studied in recent years for capturing more complete aspect-level sentiment information. However, a systematic review of various ABSA tasks and their corresponding solutions is still lacking, which we aim to fill in this survey. More specifically, we provide a new taxonomy for ABSA which organizes existing studies from the axes of concerned sentiment elements, with an emphasis on recent advances of compound ABSA tasks. From the perspective of solutions, we summarize the utilization of pre-trained language models for ABSA, which improved the performance of ABSA to a new stage. Besides, techniques for building more practical ABSA systems in cross-domain/lingual scenarios are discussed. Finally, we review some emerging topics and discuss some open challenges to outlook potential future directions of ABSA.

Zhonghua Zheng, Lizi Liao, Yang Deng et al.

The integration of emotional support into various conversational scenarios presents profound societal benefits, such as social interactions, mental health counseling, and customer service. However, there are unsolved challenges that hinder real-world applications in this field, including limited data availability and the absence of well-accepted model training paradigms. This work endeavors to navigate these challenges by harnessing the capabilities of Large Language Models (LLMs). We introduce an innovative methodology that synthesizes human insights with the computational prowess of LLMs to curate an extensive emotional support dialogue dataset. Our approach is initiated with a meticulously designed set of dialogues spanning diverse scenarios as generative seeds. By utilizing the in-context learning potential of ChatGPT, we recursively generate an ExTensible Emotional Support dialogue dataset, named ExTES. Following this, we deploy advanced tuning techniques on the LLaMA model, examining the impact of diverse training strategies, ultimately yielding an LLM meticulously optimized for emotional support interactions. An exhaustive assessment of the resultant model showcases its proficiency in offering emotional support, marking a pivotal step in the realm of emotional support bots and paving the way for subsequent research and implementations.

Nuo Chen, Yan Wang, Yang Deng et al.

This survey explores the burgeoning field of role-playing with language models, focusing on their development from early persona-based models to advanced character-driven simulations facilitated by Large Language Models (LLMs). Initially confined to simple persona consistency due to limited model capabilities, role-playing tasks have now expanded to embrace complex character portrayals involving character consistency, behavioral alignment, and overall attractiveness. We provide a comprehensive taxonomy of the critical components in designing these systems, including data, models and alignment, agent architecture and evaluation. This survey not only outlines the current methodologies and challenges, such as managing dynamic personal profiles and achieving high-level persona consistency but also suggests avenues for future research in improving the depth and realism of role-playing applications. The goal is to guide future research by offering a structured overview of current methodologies and identifying potential areas for improvement. Related resources and papers are available at https://github.com/nuochenpku/Awesome-Role-Play-Papers.

Yang Deng, Wenxuan Zhang, Weiwen Xu et al.

Recent years witnessed several advances in developing multi-goal conversational recommender systems (MG-CRS) that can proactively attract users' interests and naturally lead user-engaged dialogues with multiple conversational goals and diverse topics. Four tasks are often involved in MG-CRS, including Goal Planning, Topic Prediction, Item Recommendation, and Response Generation. Most existing studies address only some of these tasks. To handle the whole problem of MG-CRS, modularized frameworks are adopted where each task is tackled independently without considering their interdependencies. In this work, we propose a novel Unified MultI-goal conversational recommeNDer system, namely UniMIND. In specific, we unify these four tasks with different formulations into the same sequence-to-sequence (Seq2Seq) paradigm. Prompt-based learning strategies are investigated to endow the unified model with the capability of multi-task learning. Finally, the overall learning and inference procedure consists of three stages, including multi-task learning, prompt-based tuning, and inference. Experimental results on two MG-CRS benchmarks (DuRecDial and TG-ReDial) show that UniMIND achieves state-of-the-art performance on all tasks with a unified model. Extensive analyses and discussions are provided for shedding some new perspectives for MG-CRS.

Yang Deng, Wenqiang Lei, Wenxuan Zhang et al.

To facilitate conversational question answering (CQA) over hybrid contexts in finance, we present a new dataset, named PACIFIC. Compared with existing CQA datasets, PACIFIC exhibits three key features: (i) proactivity, (ii) numerical reasoning, and (iii) hybrid context of tables and text. A new task is defined accordingly to study Proactive Conversational Question Answering (PCQA), which combines clarification question generation and CQA. In addition, we propose a novel method, namely UniPCQA, to adapt a hybrid format of input and output content in PCQA into the Seq2Seq problem, including the reformulation of the numerical reasoning process as code generation. UniPCQA performs multi-task learning over all sub-tasks in PCQA and incorporates a simple ensemble strategy to alleviate the error propagation issue in the multi-task learning by cross-validating top-$k$ sampled Seq2Seq outputs. We benchmark the PACIFIC dataset with extensive baselines and provide comprehensive evaluations on each sub-task of PCQA.

Xiaoyan Zhao, Yang Deng, Min Yang et al.

Relation extraction (RE) involves identifying the relations between entities from underlying content. RE serves as the foundation for many natural language processing (NLP) and information retrieval applications, such as knowledge graph completion and question answering. In recent years, deep neural networks have dominated the field of RE and made noticeable progress. Subsequently, the large pre-trained language models have taken the state-of-the-art RE to a new level. This survey provides a comprehensive review of existing deep learning techniques for RE. First, we introduce RE resources, including datasets and evaluation metrics. Second, we propose a new taxonomy to categorize existing works from three perspectives, i.e., text representation, context encoding, and triplet prediction. Third, we discuss several important challenges faced by RE and summarize potential techniques to tackle these challenges. Finally, we outline some promising future directions and prospects in this field. This survey is expected to facilitate researchers' collaborative efforts to address the challenges of real-world RE systems.

Yang Deng, Wenxuan Zhang, Wai Lam et al.

Proactive dialogues serve as a practical yet challenging dialogue problem in the era of large language models (LLMs), where the dialogue policy planning is the key to improving the proactivity of LLMs. Most existing studies enable the dialogue policy planning of LLMs using various prompting schemes or iteratively enhance this capability in handling the given case with verbal AI feedback. However, these approaches are either bounded by the policy planning capability of the frozen LLMs or hard to be transferred to new cases. In this work, we introduce a new dialogue policy planning paradigm to strategize LLMs for proactive dialogue problems with a tunable language model plug-in as a plug-and-play dialogue policy planner, named PPDPP. Specifically, we develop a novel training framework to facilitate supervised fine-tuning over available human-annotated data as well as reinforcement learning from goal-oriented AI feedback with dynamic interaction data collected by the LLM-based self-play simulation. In this manner, the LLM-powered dialogue agent can not only be generalized to different cases after the training, but also be applicable to different applications by just substituting the learned plug-in. In addition, we propose to evaluate the policy planning capability of dialogue systems under the interactive setting. Experimental results demonstrate that PPDPP consistently and substantially outperforms existing approaches on three different proactive dialogue applications, including negotiation, emotional support, and tutoring dialogues.

Liang Chen, Yang Deng, Yatao Bian et al.

Large language models (LLMs) outperform information retrieval techniques for downstream knowledge-intensive tasks when being prompted to generate world knowledge. However, community concerns abound regarding the factuality and potential implications of using this uncensored knowledge. In light of this, we introduce CONNER, a COmpreheNsive kNowledge Evaluation fRamework, designed to systematically and automatically evaluate generated knowledge from six important perspectives -- Factuality, Relevance, Coherence, Informativeness, Helpfulness and Validity. We conduct an extensive empirical analysis of the generated knowledge from three different types of LLMs on two widely studied knowledge-intensive tasks, i.e., open-domain question answering and knowledge-grounded dialogue. Surprisingly, our study reveals that the factuality of generated knowledge, even if lower, does not significantly hinder downstream tasks. Instead, the relevance and coherence of the outputs are more important than small factual mistakes. Further, we show how to use CONNER to improve knowledge-intensive tasks by designing two strategies: Prompt Engineering and Knowledge Selection. Our evaluation code and LLM-generated knowledge with human annotations will be released to facilitate future research.

Weiwen Xu, Xin Li, Yang Deng et al.

Span identification aims at identifying specific text spans from text input and classifying them into pre-defined categories. Different from previous works that merely leverage the Subordinate (SUB) relation (i.e. if a span is an instance of a certain category) to train models, this paper for the first time explores the Peer (PR) relation, which indicates that two spans are instances of the same category and share similar features. Specifically, a novel Peer Data Augmentation (PeerDA) approach is proposed which employs span pairs with the PR relation as the augmentation data for training. PeerDA has two unique advantages: (1) There are a large number of PR span pairs for augmenting the training data. (2) The augmented data can prevent the trained model from over-fitting the superficial span-category mapping by pushing the model to leverage the span semantics. Experimental results on ten datasets over four diverse tasks across seven domains demonstrate the effectiveness of PeerDA. Notably, PeerDA achieves state-of-the-art results on six of them.

Weiwen Xu, Yang Deng, Wenqiang Lei et al.

We study automatic Contract Clause Extraction (CCE) by modeling implicit relations in legal contracts. Existing CCE methods mostly treat contracts as plain text, creating a substantial barrier to understanding contracts of high complexity. In this work, we first comprehensively analyze the complexity issues of contracts and distill out three implicit relations commonly found in contracts, namely, 1) Long-range Context Relation that captures the correlations of distant clauses; 2) Term-Definition Relation that captures the relation between important terms with their corresponding definitions; and 3) Similar Clause Relation that captures the similarities between clauses of the same type. Then we propose a novel framework ConReader to exploit the above three relations for better contract understanding and improving CCE. Experimental results show that ConReader makes the prediction more interpretable and achieves new state-of-the-art on two CCE tasks in both conventional and zero-shot settings.

Yang Deng, Wenxuan Zhang, Qian Yu et al.

Product question answering (PQA), aiming to automatically provide instant responses to customer's questions in E-Commerce platforms, has drawn increasing attention in recent years. Compared with typical QA problems, PQA exhibits unique challenges such as the subjectivity and reliability of user-generated contents in E-commerce platforms. Therefore, various problem settings and novel methods have been proposed to capture these special characteristics. In this paper, we aim to systematically review existing research efforts on PQA. Specifically, we categorize PQA studies into four problem settings in terms of the form of provided answers. We analyze the pros and cons, as well as present existing datasets and evaluation protocols for each setting. We further summarize the most significant challenges that characterize PQA from general QA applications and discuss their corresponding solutions. Finally, we conclude this paper by providing the prospect on several future directions.

Zhiyuan Hu, Yue Feng, Yang Deng et al.

Recently, the development of large language models (LLMs) has been significantly enhanced the question answering and dialogue generation, and makes them become increasingly popular in current practical scenarios. While unlike the general dialogue system which emphasizes the semantic performance, the task-oriented dialogue (ToD) systems aim to achieve the dialogue goal efficiently and successfully in multiple turns. Unfortunately, existing LLM-induced ToD systems lack the direct reward toward the final goal and do not take account of the dialogue proactivity that can strengthen the dialogue efficiency. To fill these gaps, we introduce the ProToD (Proactively Goal-Driven LLM-Induced ToD) approach, which anticipates the future dialogue actions and incorporates the goal-oriented reward signal to enhance ToD systems. Additionally, we present a novel evaluation method that assesses ToD systems based on goal-driven dialogue simulations. This method allows us to gauge user satisfaction, system efficiency and successful rate while overcoming the limitations of current Information and Success metrics. Empirical experiments conducted on the MultiWoZ 2.1 dataset demonstrate that our model can achieve superior performance using only 10% of the data compared to previous end-to-end fully supervised models. This improvement is accompanied by enhanced user satisfaction and efficiency.

Hongru Wang, Minda Hu, Yang Deng et al.

Open-domain dialogue system usually requires different sources of knowledge to generate more informative and evidential responses. However, existing knowledge-grounded dialogue systems either focus on a single knowledge source or overlook the dependency between multiple sources of knowledge, which may result in generating inconsistent or even paradoxical responses. To incorporate multiple knowledge sources and dependencies between them, we propose SAFARI, a novel framework that leverages the exceptional capabilities of large language models (LLMs) in planning, understanding, and incorporating under both supervised and unsupervised settings. Specifically, SAFARI decouples the knowledge grounding into multiple sources and response generation, which allows easy extension to various knowledge sources including the possibility of not using any sources. To study the problem, we construct a personalized knowledge-grounded dialogue dataset \textit{\textbf{K}nowledge \textbf{B}ehind \textbf{P}ersona}~(\textbf{KBP}), which is the first to consider the dependency between persona and implicit knowledge. Experimental results on the KBP dataset demonstrate that the SAFARI framework can effectively produce persona-consistent and knowledge-enhanced responses.

Liang Chen, Yatao Bian, Yang Deng et al.

Text watermarking has emerged as a pivotal technique for identifying machine-generated text. However, existing methods often rely on arbitrary vocabulary partitioning during decoding to embed watermarks, which compromises the availability of suitable tokens and significantly degrades the quality of responses. This study assesses the impact of watermarking on different capabilities of large language models (LLMs) from a cognitive science lens. Our finding highlights a significant disparity; knowledge recall and logical reasoning are more adversely affected than language generation. These results suggest a more profound effect of watermarking on LLMs than previously understood. To address these challenges, we introduce Watermarking with Mutual Exclusion (WatME), a novel approach leveraging linguistic prior knowledge of inherent lexical redundancy in LLM vocabularies to seamlessly integrate watermarks. Specifically, WatME dynamically optimizes token usage during the decoding process by applying a mutually exclusive rule to the identified lexical redundancies. This strategy effectively prevents the unavailability of appropriate tokens and preserves the expressive power of LLMs. We provide both theoretical analysis and empirical evidence showing that WatME effectively preserves the diverse capabilities of LLMs while ensuring watermark detectability.

Peixin Qin, Chen Huang, Yang Deng et al.

With the aid of large language models, current conversational recommender system (CRS) has gaining strong abilities to persuade users to accept recommended items. While these CRSs are highly persuasive, they can mislead users by incorporating incredible information in their explanations, ultimately damaging the long-term trust between users and the CRS. To address this, we propose a simple yet effective method, called PC-CRS, to enhance the credibility of CRS's explanations during persuasion. It guides the explanation generation through our proposed credibility-aware persuasive strategies and then gradually refines explanations via post-hoc self-reflection. Experimental results demonstrate the efficacy of PC-CRS in promoting persuasive and credible explanations. Further analysis reveals the reason behind current methods producing incredible explanations and the potential of credible explanations to improve recommendation accuracy.

Shuaiyi Li, Yang Deng, Wai Lam

Spatial reasoning in text plays a crucial role in various real-world applications. Existing approaches for spatial reasoning typically infer spatial relations from pure text, which overlooks the gap between natural language and symbolic structures. Graph neural networks (GNNs) have showcased exceptional proficiency in inducing and aggregating symbolic structures. However, classical GNNs face challenges in handling multi-hop spatial reasoning due to the over-smoothing issue, i.e., the performance decreases substantially as the number of graph layers increases. To cope with these challenges, we propose a novel Depth-Wise Graph Neural Network (DepWiGNN). Specifically, we design a novel node memory scheme and aggregate the information over the depth dimension instead of the breadth dimension of the graph, which empowers the ability to collect long dependencies without stacking multiple layers. Experimental results on two challenging multi-hop spatial reasoning datasets show that DepWiGNN outperforms existing spatial reasoning methods. The comparisons with the other three GNNs further demonstrate its superiority in capturing long dependency in the graph.

Suhyun Lee, Palakorn Achananuparp, Neemesh Yadav et al.

Large language models (LLMs) are increasingly explored as scalable tools for mental health counseling, yet evaluating their safety remains challenging due to the interactional and context-dependent nature of clinical harm. Existing evaluation frameworks predominantly assess isolated responses using coarse-grained taxonomies or static datasets, limiting their ability to diagnose how harms emerge and accumulate over multi-turn counseling interactions. In this work, we introduce R-MHSafe, a role-aware mental health safety taxonomy that characterizes clinically significant harm in terms of the interactional roles an AI counselor adopts, including perpetrator, instigator, facilitator, or enabler, combined with clinically grounded harm categories. Then, we propose MHSafeEval, a closed-loop, agent-based evaluation framework that formulates safety assessment as trajectory-level discovery of harm through adversarial multi-turn interactions, guided by role-aware modeling. Using R-MHSafe and MHSafeEval, we conduct a large-scale evaluation across state-of-the-art LLMs. Our results reveal substantial role-dependent and cumulative safety failures that are systematically missed by existing static benchmarks, and show that our framework significantly improves failure-mode coverage and diagnostic granularity.

Simiao Ren, Saad Lahrichi, Yang Deng et al.

Deep active learning (DAL) methods have shown significant improvements in sample efficiency compared to simple random sampling. While these studies are valuable, they nearly always assume that optimal DAL hyperparameter (HP) settings are known in advance, or optimize the HPs through repeating DAL several times with different HP settings. Here, we argue that in real-world settings, or in the wild, there is significant uncertainty regarding good HPs, and their optimization contradicts the premise of using DAL (i.e., we require labeling efficiency). In this study, we evaluate the performance of eleven modern DAL methods on eight benchmark problems as we vary a key HP shared by all methods: the pool ratio. Despite adjusting only one HP, our results indicate that eight of the eleven DAL methods sometimes underperform relative to simple random sampling and some frequently perform worse. Only three methods always outperform random sampling (albeit narrowly), and we find that these methods all utilize diversity to select samples - a relatively simple criterion. Our findings reveal the limitations of existing DAL methods when deployed in the wild, and present this as an important new open problem in the field.

Zhixiang Cheng, Hongxin Xiang, Pengsen Ma et al.

Activity cliffs, which refer to pairs of molecules that are structurally similar but show significant differences in their potency, can lead to model representation collapse and make the model challenging to distinguish them. Our research indicates that as molecular similarity increases, graph-based methods struggle to capture these nuances, whereas image-based approaches effectively retain the distinctions. Thus, we developed MaskMol, a knowledge-guided molecular image self-supervised learning framework. MaskMol accurately learns the representation of molecular images by considering multiple levels of molecular knowledge, such as atoms, bonds, and substructures. By utilizing pixel masking tasks, MaskMol extracts fine-grained information from molecular images, overcoming the limitations of existing deep learning models in identifying subtle structural changes. Experimental results demonstrate MaskMol's high accuracy and transferability in activity cliff estimation and compound potency prediction across 20 different macromolecular targets, outperforming 25 state-of-the-art deep learning and machine learning approaches. Visualization analyses reveal MaskMol's high biological interpretability in identifying activity cliff-relevant molecular substructures. Notably, through MaskMol, we identified candidate EP4 inhibitors that could be used to treat tumors. This study not only raises awareness about activity cliffs but also introduces a novel method for molecular image representation learning and virtual screening, advancing drug discovery and providing new insights into structure-activity relationships (SAR).

Weixiang Zhao, Yingshuo Wang, Yichen Zhang et al.

Self-evolving large language model (LLM) agents continually improve by accumulating and reusing past experience, yet it remains unclear whether they faithfully rely on that experience to guide their behavior. We present the first systematic investigation of experience faithfulness, the causal dependence of an agent's decisions on the experience it is given, in self-evolving LLM agents. Using controlled causal interventions on both raw and condensed forms of experience, we comprehensively evaluate four representative frameworks across 10 LLM backbones and 9 environments. Our analysis uncovers a striking asymmetry: while agents consistently depend on raw experience, they often disregard or misinterpret condensed experience, even when it is the only experience provided. This gap persists across single- and multi-agent configurations and across backbone scales. We trace its underlying causes to three factors: the semantic limitations of condensed content, internal processing biases that suppress experience, and task regimes where pretrained priors already suffice. These findings challenge prevailing assumptions about self-evolving methods and underscore the need for more faithful and reliable approaches to experience integration.

Chuang Li, Weida Liang, Hengchang Hu et al.

We tackle the challenge of integrating large language models (LLMs) with external recommender systems to enhance domain expertise in conversational recommendation (CRS). Current LLM-based CRS approaches primarily rely on zero/few-shot methods for generating item recommendations based on user queries, but this method faces two significant challenges: (1) without domain-specific adaptation, LLMs frequently recommend items not in the target item space, resulting in low recommendation accuracy; and (2) LLMs largely rely on dialogue context for content-based recommendations, neglecting the collaborative relationships among item sequences. To address these limitations, we introduce the CARE (Contextual Adaptation of Recommenders) framework. CARE (a) integrates external recommender systems as domain experts, producing candidate items through entity-level insights, and (b) customizes LLMs as rerankers to enhance the accuracy by leveraging contextual information. Our results demonstrate that incorporating CARE framework significantly enhances recommendation accuracy of LLMs by an average of 54% and 25% for ReDial and INSPIRED datasets. The most effective CARE strategy involves LLMs selecting and reranking candidate items that external recommenders provide based on contextual insights.

Tong Zhang, Peixin Qin, Yang Deng et al.

Large language models (LLMs) are increasingly used to meet user information needs, but their effectiveness in dealing with user queries that contain various types of ambiguity remains unknown, ultimately risking user trust and satisfaction. To this end, we introduce CLAMBER, a benchmark for evaluating LLMs using a well-organized taxonomy. Building upon the taxonomy, we construct ~12K high-quality data to assess the strengths, weaknesses, and potential risks of various off-the-shelf LLMs. Our findings indicate the limited practical utility of current LLMs in identifying and clarifying ambiguous user queries, even enhanced by chain-of-thought (CoT) and few-shot prompting. These techniques may result in overconfidence in LLMs and yield only marginal enhancements in identifying ambiguity. Furthermore, current LLMs fall short in generating high-quality clarifying questions due to a lack of conflict resolution and inaccurate utilization of inherent knowledge. In this paper, CLAMBER presents a guidance and promotes further research on proactive and trustworthy LLMs. Our dataset is available at https://github.com/zt991211/CLAMBER

Kaiyuan Zhai, Jiacheng Cui, Zhehao Zhang et al.

Cross-domain HVAC energy prediction is essential for scalable building energy management, particularly because collecting extensive labeled data for every new building is both costly and impractical. Yet, this task remains highly challenging due to the scarcity and heterogeneity of data across different buildings, climate zones, and seasonal patterns. In particular, buildings situated in distinct climatic regions introduce variability that often leads existing methods to overfit to spurious correlations, rely heavily on expert intervention, or compromise on data diversity. To address these limitations, we propose CaberNet, a causal and interpretable deep sequence model that learns invariant (Markov blanket) representations for robust cross-domain prediction. In a purely data-driven fashion and without requiring any prior knowledge, CaberNet integrates i) a global feature gate trained with a self-supervised Bernoulli regularization to distinguish superior causal features from inferior ones, and ii) a domain-wise training scheme that balances domain contributions, minimizes cross-domain loss variance, and promotes latent factor independence. We evaluate CaberNet on real-world datasets collected from three buildings located in three climatically diverse cities, and it consistently outperforms all baselines, achieving a 22.9% reduction in normalized mean squared error (NMSE) compared to the best benchmark. Our code is available at https://github.com/SusCom-Lab/CaberNet-CRL.

Yifei Yuan, Wenxuan Zhang, Yang Deng et al.

Social media plays a significant role in boosting the fashion industry, where a massive amount of fashion-related posts are generated every day. In order to obtain the rich fashion information from the posts, we study the task of social media fashion knowledge extraction. Fashion knowledge, which typically consists of the occasion, person attributes, and fashion item information, can be effectively represented as a set of tuples. Most previous studies on fashion knowledge extraction are based on the fashion product images without considering the rich text information in social media posts. Existing work on fashion knowledge extraction in social media is classification-based and requires to manually determine a set of fashion knowledge categories in advance. In our work, we propose to cast the task as a captioning problem to capture the interplay of the multimodal post information. Specifically, we transform the fashion knowledge tuples into a natural language caption with a sentence transformation method. Our framework then aims to generate the sentence-based fashion knowledge directly from the social media post. Inspired by the big success of pre-trained models, we build our model based on a multimodal pre-trained generative model and design several auxiliary tasks for enhancing the knowledge extraction. Since there is no existing dataset which can be directly borrowed to our task, we introduce a dataset consisting of social media posts with manual fashion knowledge annotation. Extensive experiments are conducted to demonstrate the effectiveness of our model.

Tu Ao, Yanhua Yu, Yuling Wang et al.

Large Language Models (LLMs) have impressive capabilities in text understanding and zero-shot reasoning. However, delays in knowledge updates may cause them to reason incorrectly or produce harmful results. Knowledge Graphs (KGs) provide rich and reliable contextual information for the reasoning process of LLMs by structurally organizing and connecting a wide range of entities and relations. Existing KG-based LLM reasoning methods only inject KGs' knowledge into prompts in a textual form, ignoring its structural information. Moreover, they mostly rely on close-source models or open-source models with large parameters, which poses challenges to high resource consumption. To address this, we propose a novel Lightweight and efficient Prompt learning-ReasOning Framework for KGQA (LightPROF), which leverages the full potential of LLMs to tackle complex reasoning tasks in a parameter-efficient manner. Specifically, LightPROF follows a "Retrieve-Embed-Reason process", first accurately, and stably retrieving the corresponding reasoning graph from the KG through retrieval module. Next, through a Transformer-based Knowledge Adapter, it finely extracts and integrates factual and structural information from the KG, then maps this information to the LLM's token embedding space, creating an LLM-friendly prompt to be used by the LLM for the final reasoning. Additionally, LightPROF only requires training Knowledge Adapter and can be compatible with any open-source LLM. Extensive experiments on two public KGQA benchmarks demonstrate that LightPROF achieves superior performance with small-scale LLMs. Furthermore, LightPROF shows significant advantages in terms of input token count and reasoning time.

Xuan Zhang, Wenxuan Zhang, Anxu Wang et al.

Multi-agent role-playing has recently shown promise for studying social behavior with language agents, but existing simulations are mostly monolingual and fail to model cross-lingual interaction, an essential property of real societies. We introduce MASim, the first multilingual agent-based simulation framework that supports multi-turn interaction among generative agents with diverse sociolinguistic profiles. MASim offers two key analyses: (i) global public opinion modeling, by simulating how attitudes toward open-domain hypotheses evolve across languages and cultures, and (ii) media influence and information diffusion, via autonomous news agents that dynamically generate content and shape user behavior. To instantiate simulations, we construct the MAPS benchmark, which combines survey questions and demographic personas drawn from global population distributions. Experiments on calibration, sensitivity, consistency, and cultural case studies show that MASim reproduces sociocultural phenomena and highlights the importance of multilingual simulation for scalable, controlled computational social science.

Yifei Yuan, Yang Deng, Anders Søgaard et al.

Users post numerous product-related questions on e-commerce platforms, affecting their purchase decisions. Product-related question answering (PQA) entails utilizing product-related resources to provide precise responses to users. We propose a novel task of Multilingual Cross-market Product-based Question Answering (MCPQA) and define the task as providing answers to product-related questions in a main marketplace by utilizing information from another resource-rich auxiliary marketplace in a multilingual context. We introduce a large-scale dataset comprising over 7 million questions from 17 marketplaces across 11 languages. We then perform automatic translation on the Electronics category of our dataset, naming it as McMarket. We focus on two subtasks: review-based answer generation and product-related question ranking. For each subtask, we label a subset of McMarket using an LLM and further evaluate the quality of the annotations via human assessment. We then conduct experiments to benchmark our dataset, using models ranging from traditional lexical models to LLMs in both single-market and cross-market scenarios across McMarket and the corresponding LLM subset. Results show that incorporating cross-market information significantly enhances performance in both tasks.

Yong Zhao, Yang Deng, See-Kiong Ng et al.

Large Language Models (LLMs) have demonstrated impressive capabilities in complex reasoning tasks. However, they can be easily misled by unfaithful arguments during conversations, even when their original statements are correct. To this end, we investigate the problem of maintaining faithful integrity in LLMs. This involves ensuring that LLMs adhere to their faithful statements in the face of opposing arguments and are able to correct their incorrect statements when presented with faithful arguments. In this work, we propose a novel framework, named Alignment for Faithful Integrity with Confidence Estimation (AFICE), which aims to align the LLM responses with faithful integrity. Specifically, AFICE first designs a Bilateral Confidence Estimation (BCE) approach for estimating the uncertainty of each response generated by the LLM given a specific context, which simultaneously estimate the model's confidence to the question based on the internal states during decoding as well as to the answer based on cumulative probability ratios. With the BCE, we construct a conversational preference dataset composed of context, original statement, and argument, which is adopted for aligning the LLM for faithful integrity using Direct Preference Optimization (DPO). Extensive experimental results on a wide range of benchmarks demonstrate significant improvements in the LLM's ability to maintain faithful responses when encountering opposing arguments, ensuring both the practical utility and trustworthiness of LLMs in complex interactive settings. Code and data will be released via https://github.com/zhaoy777/AFICE.git

Yang Deng, David Mould, Paul L. Rosin et al.

Existing text-to-image diffusion models, while excelling at subject synthesis, exhibit a persistent foreground bias that treats the background as a passive and under-optimized byproduct. This imbalance compromises global scene coherence and constrains compositional control. To address the limitation, we propose a training-free framework that restructures diffusion sampling to explicitly account for foreground-background interactions. Our approach consists of two key components. First, Dynamic Spatial Guidance introduces a soft, time step dependent gating mechanism that modulates foreground and background attention during the diffusion process, enabling spatially balanced generation. Second, Multi-Path Pruning performs multi-path latent exploration and dynamically filters candidate trajectories using both internal attention statistics and external semantic alignment signals, retaining trajectories that better satisfy object-background constraints. We further develop a benchmark specifically designed to evaluate object-background compositionality. Extensive evaluations across multiple diffusion backbones demonstrate consistent improvements in background coherence and object-background compositional alignment.

Shuaiyi Li, Zhisong Zhang, Yan Wang et al.

Block attention, which processes the input as separate blocks that cannot attend to one another, offers significant potential to improve KV cache reuse in long-context scenarios such as Retrieval-Augmented Generation (RAG). However, its broader application is hindered by two key challenges: the difficulty of segmenting input text into meaningful, self-contained blocks, and the inefficiency of existing block fine-tuning methods that risk degrading performance. To address these, we first construct SemanticSeg, a large and diverse semantic segmentation dataset containing over 30k instances across 16 categories-including books, code, web text, and conversations with text lengths ranging from 2k to 32k. Using this dataset, we train a lightweight segmenter to automatically partition text into human-instinct-aligned blocks with controllable granularity. Second, we propose block distillation, a training framework that is more efficient than block fine-tuning, which uses a frozen full-attention teacher model to guide the block-attention student. This framework integrates three novel components: block sink tokens to mitigate information loss at block boundaries, block dropout to leverage training signals from all blocks, and token-level loss weighting to focus learning on block-attention-sensitive tokens. Experiments across multiple models and benchmarks demonstrate that our segmenter outperforms heuristic and statistical baselines, and block distillation achieves near-full-attention performance under block attention, establishing a practical and scalable pathway for deploying block attention.

Dingzirui Wang, Xuanliang Zhang, Keyan Xu et al.

Contextual information at inference time, such as demonstrations, retrieved knowledge, or interaction history, can substantially improve large language models (LLMs) without parameter updates, yet its theoretical role remains poorly understood beyond specific settings such as in-context learning (ICL). We present a unified theoretical framework for analyzing the effect of arbitrary contextual information in Transformer-based LLMs. Our analysis characterizes contextual influence through output error dynamics. In a single-layer Transformer, we prove that the context-conditioned error vector decomposes additively into the baseline error vector and a contextual correction vector. This yields necessary geometric conditions for error reduction: the contextual correction must align with the negative baseline error and satisfy a norm constraint. We further show that the contextual correction norm admits an explicit upper bound determined by context-query relevance and complementarity. These results extend to multi-context and multi-layer Transformers. Experiments across ICL, retrieval-augmented generation, and memory evolution validate our theory and motivate a principled context selection strategy that improves performance by $0.6\%$.

Rui Liang, Yang Deng, Donghua Xie et al.

Advances in time-series forecasting are driving a shift from conventional machine learning models to foundation models (FMs) that are trained with generalized knowledge. However, existing FMs still perform poorly in the energy fields, such as building energy forecasting (BEF). This paper studies the adaptation of FM to BEF tasks. We demonstrate the shortcomings of fine-tuning FM straightforwardly from both the perspectives of FM and the data. To overcome these limitations, we propose a new \textit{contrastive curriculum learning}-based training method. Our method optimizes the ordering of training data in the context of TSFM adaptation. Experiments show that our method can improve the zero/few-shot performance by 14.6\% compared to the existing FMs. Our code and new TSFM will be available at <Anonymous Github Repo>.

Zesheng Wei, Mengxiang Li, Zilei Wang et al.

Personalized Large Language Models (LLMs) facilitate more natural, human-like interactions in human-centric applications. However, existing personalization methods are constrained by limited controllability and high resource demands. Furthermore, their reliance on static personality modeling restricts adaptability across varying situations. To address these limitations, we first demonstrate the existence of situation-dependency and consistent situation-behavior patterns within LLM personalities through a multi-perspective analysis of persona neurons. Building on these insights, we propose IRIS, a training-free, neuron-based Identify-Retrieve-Steer framework for advanced situational personality steering. Our approach comprises situational persona neuron identification, situation-aware neuron retrieval, and similarity-weighted steering. We empirically validate our framework on PersonalityBench and our newly introduced SPBench, a comprehensive situational personality benchmark. Experimental results show that our method surpasses best-performing baselines, demonstrating IRIS's generalization and robustness to complex, unseen situations and different models architecture.

Shuai Zhen, Yanhua Yu, Ruopei Guo et al.

Large language model (LLM) agents have demonstrated strong capabilities in complex interactive decision-making tasks. However, existing LLM agents typically rely on increasingly long interaction histories, resulting in high computational cost and limited scalability. In this paper, we propose STEP-HRL, a hierarchical reinforcement learning (HRL) framework that enables step-level learning by conditioning only on single-step transitions rather than full interaction histories. STEP-HRL structures tasks hierarchically, using completed subtasks to represent global progress of overall task. By introducing a local progress module, it also iteratively and selectively summarizes interaction history within each subtask to produce a compact summary of local progress. Together, these components yield augmented step-level transitions for both high-level and low-level policies. Experimental results on ScienceWorld and ALFWorld benchmarks consistently demonstrate that STEP-HRL substantially outperforms baselines in terms of performance and generalization while reducing token usage. Our code is available at https://github.com/TonyStark042/STEP-HRL.

Xubo Lin, Zezhii Deng, Shihao Wang et al.

Most existing dialogue systems are user-driven, primarily designed to fulfill user requests. However, in many critical real-world scenarios, a conversational agent must proactively extract information to achieve its own objectives rather than merely respond. To address this gap, we introduce \emph{Inquisitive Conversational Agents (ICAs)} and develop an ICA specifically tailored to U.S. Supreme Court oral arguments. We propose a Dual Hierarchical Reinforcement Learning framework featuring two cooperating RL agents, each with its own policy, to coordinate strategic dialogue management and fine-grained utterance generation. By learning when and how to ask probing questions, the agent emulates judicial questioning patterns and systematically uncovers crucial information to fulfill its legal objectives. Evaluations on a U.S. Supreme Court dataset show that our method outperforms various baselines across multiple metrics. It represents an important first step toward broader high-stakes, domain-specific applications.

Yihuai Hong, Yiran Zhao, Wei Tang et al.

Over time, a growing wave of large language models from various series has been introduced to the community. Researchers are striving to maximize the performance of language models with constrained parameter sizes. However, from a microscopic perspective, there has been limited research on how to better store knowledge in model parameters, particularly within MLPs, to enable more effective utilization of this knowledge by the model. In this work, we analyze twenty publicly available open-source large language models to investigate the relationship between their strong performance and the way knowledge is stored in their corresponding MLP parameters. Our findings reveal that as language models become more advanced and demonstrate stronger knowledge capabilities, their parameters exhibit increased specialization. Specifically, parameters in the MLPs tend to be more focused on encoding similar types of knowledge. We experimentally validate that this specialized distribution of knowledge contributes to improving the efficiency of knowledge utilization in these models. Furthermore, by conducting causal training experiments, we confirm that this specialized knowledge distribution plays a critical role in improving the model's efficiency in leveraging stored knowledge.

Weixiang Zhao, Jiahe Guo, Yang Deng et al.

Multilingual reasoning remains a significant challenge for large language models (LLMs), with performance disproportionately favoring high-resource languages. Drawing inspiration from cognitive neuroscience, which suggests that human reasoning functions largely independently of language processing, we hypothesize that LLMs similarly encode reasoning and language as separable components that can be disentangled to enhance multilingual reasoning. To evaluate this, we perform a causal intervention by ablating language-specific representations at inference time. Experiments on 10 open-source LLMs spanning 11 typologically diverse languages show that this language-specific ablation consistently boosts multilingual reasoning performance. Layer-wise analyses further confirm that language and reasoning representations can be effectively decoupled throughout the model, yielding improved multilingual reasoning capabilities, while preserving top-layer language features remains essential for maintaining linguistic fidelity. Compared to post-training such as supervised fine-tuning or reinforcement learning, our training-free ablation achieves comparable or superior results with minimal computational overhead. These findings shed light on the internal mechanisms underlying multilingual reasoning in LLMs and suggest a lightweight and interpretable strategy for improving cross-lingual generalization.

Mengfan Li, Xuanhua Shi, Yang Deng

Theory of Mind (ToM), the ability to attribute mental states to others, is a hallmark of social intelligence. While large language models (LLMs) demonstrate promising performance on standard ToM benchmarks, we observe that they often fail to generalize to complex task-specific scenarios, relying heavily on prompt scaffolding to mimic reasoning. The critical misalignment between the internal knowledge and external behavior raises a fundamental question: Do LLMs truly possess intrinsic cognition, and can they externalize this internal knowledge into stable, high-quality behaviors? To answer this, we introduce CoSToM (Causal-oriented Steering for ToM alignment), a framework that transitions from mechanistic interpretation to active intervention. First, we employ causal tracing to map the internal distribution of ToM features, empirically uncovering the internal layers' characteristics in encoding fundamental ToM semantics. Building on this insight, we implement a lightweight alignment framework via targeted activation steering within these ToM-critical layers. Experiments demonstrate that CoSToM significantly enhances human-like social reasoning capabilities and downstream dialogue quality.

Jingyao Liu, Chen Huang, Zhizhao Guan et al.

The rapid advancement in large language models (LLMs) has demonstrated significant potential in End-to-End Software Development (E2ESD). However, existing E2ESD benchmarks are limited by coarse-grained requirement specifications and unreliable evaluation protocols, hindering a true understanding of current framework capabilities. To address these limitations, we present E2EDev, a novel benchmark grounded in the principles of Behavior-Driven Development (BDD), which evaluates the capabilities of E2ESD frameworks by assessing whether the generated software meets user needs through mimicking real user interactions (Figure 1). E2EDev comprises (i) a fine-grained set of user requirements, (ii) multiple BDD test scenarios with corresponding Python step implementations for each requirement, and (iii) a fully automated testing pipeline built on the Behave framework. To ensure its quality while reducing the annotation effort, E2EDev leverages our proposed Human-in-the-Loop Multi-Agent Annotation Framework (HITL-MAA). By evaluating various E2ESD frameworks and LLM backbones with E2EDev, our analysis reveals a persistent struggle to effectively solve these tasks, underscoring the critical need for more effective and cost-efficient E2ESD solutions. Our codebase and benchmark are publicly available at https://github.com/SCUNLP/E2EDev.

Zhiyu Yang, Shuo Wang, Yukun Yan et al.

LLMs are transforming software development, yet current code generation and code repair benchmarks mainly assess syntactic and functional correctness in simple, single-error cases. LLMs' capabilities to autonomously find and fix runtime logical errors in complex data science code remain largely unexplored. To address this gap, we introduce DSDBench: the Data Science Debugging Benchmark, the first benchmark for systematic evaluation of LLMs on multi-hop error tracing and multi-bug detection in data science code debugging. DSDBench adapts datasets from existing data science task benchmarks, such as DABench and MatPlotBench, featuring realistic data science debugging tasks with automatically synthesized multi-hop, multi-bug code snippets. DSDBench includes 1,117 annotated samples with 741 cause-effect error pairs and runtime error messages. Evaluations of state-of-the-art LLMs on DSDBench show significant performance gaps, highlighting challenges in debugging logical runtime errors in data science code. DSDBench offers a crucial resource to evaluate and improve LLMs' debugging and reasoning capabilities, enabling more reliable AI-assisted data science in the future. DSDBench is publicly available at github.com/KevinCL16/DSDBench.

Hao Li, Chenghao Yang, An Zhang et al.

Open-domain dialogue systems have seen remarkable advancements with the development of large language models (LLMs). Nonetheless, most existing dialogue systems predominantly focus on brief single-session interactions, neglecting the real-world demands for long-term companionship and personalized interactions with chatbots. Crucial to addressing this real-world need are event summary and persona management, which enable reasoning for appropriate long-term dialogue responses. Recent progress in the human-like cognitive and reasoning capabilities of LLMs suggests that LLM-based agents could significantly enhance automated perception, decision-making, and problem-solving. In response to this potential, we introduce a model-agnostic framework, the Long-term Dialogue Agent (LD-Agent), which incorporates three independently tunable modules dedicated to event perception, persona extraction, and response generation. For the event memory module, long and short-term memory banks are employed to separately focus on historical and ongoing sessions, while a topic-based retrieval mechanism is introduced to enhance the accuracy of memory retrieval. Furthermore, the persona module conducts dynamic persona modeling for both users and agents. The integration of retrieved memories and extracted personas is subsequently fed into the generator to induce appropriate responses. The effectiveness, generality, and cross-domain capabilities of LD-Agent are empirically demonstrated across various illustrative benchmarks, models, and tasks. The code is released at https://github.com/leolee99/LD-Agent.

Yuexiang Xie, Fei Sun, Yang Deng et al.

Despite significant progress has been achieved in text summarization, factual inconsistency in generated summaries still severely limits its practical applications. Among the key factors to ensure factual consistency, a reliable automatic evaluation metric is the first and the most crucial one. However, existing metrics either neglect the intrinsic cause of the factual inconsistency or rely on auxiliary tasks, leading to an unsatisfied correlation with human judgments or increasing the inconvenience of usage in practice. In light of these challenges, we propose a novel metric to evaluate the factual consistency in text summarization via counterfactual estimation, which formulates the causal relationship among the source document, the generated summary, and the language prior. We remove the effect of language prior, which can cause factual inconsistency, from the total causal effect on the generated summary, and provides a simple yet effective way to evaluate consistency without relying on other auxiliary tasks. We conduct a series of experiments on three public abstractive text summarization datasets, and demonstrate the advantages of the proposed metric in both improving the correlation with human judgments and the convenience of usage. The source code is available at https://github.com/xieyxclack/factual_coco.

Yiwen Gao, Ruochen Zhao, Yang Deng et al.

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 the state-of-the-art alignment with human preferences without any manual efforts, validating DR-Arena as a reliable alternative for costly human adjudication.

Yang Deng, Lizi Liao, Zhonghua Zheng et al.

Recent research on proactive conversational agents (PCAs) mainly focuses on improving the system's capabilities in anticipating and planning action sequences to accomplish tasks and achieve goals before users articulate their requests. This perspectives paper highlights the importance of moving towards building human-centered PCAs that emphasize human needs and expectations, and that considers ethical and social implications of these agents, rather than solely focusing on technological capabilities. The distinction between a proactive and a reactive system lies in the proactive system's initiative-taking nature. Without thoughtful design, proactive systems risk being perceived as intrusive by human users. We address the issue by establishing a new taxonomy concerning three key dimensions of human-centered PCAs, namely Intelligence, Adaptivity, and Civility. We discuss potential research opportunities and challenges based on this new taxonomy upon the five stages of PCA system construction. This perspectives paper lays a foundation for the emerging area of conversational information retrieval research and paves the way towards advancing human-centered proactive conversational systems.

Yang Deng, Xuan Zhang, Wenxuan Zhang et al.

Web agents powered by Large Language Models (LLMs) have demonstrated remarkable abilities in planning and executing multi-step interactions within complex web-based environments, fulfilling a wide range of web navigation tasks. Despite these advancements, the potential for LLM-powered agents to effectively engage with sequential user instructions in real-world scenarios has not been fully explored. In this work, we introduce a new task of Conversational Web Navigation, which necessitates sophisticated interactions that span multiple turns with both the users and the environment, supported by a specially developed dataset named Multi-Turn Mind2Web (MT-Mind2Web). To tackle the limited context length of LLMs and the context-dependency issue of the conversational tasks, we further propose a novel framework, named self-reflective memory-augmented planning (Self-MAP), which employs memory utilization and self-reflection techniques. Extensive experiments are conducted to benchmark the MT-Mind2Web dataset, and validate the effectiveness of the proposed method.

Yang Deng, Yong Zhao, Moxin Li et al.

Despite the remarkable abilities of Large Language Models (LLMs) to answer questions, they often display a considerable level of overconfidence even when the question does not have a definitive answer. To avoid providing hallucinated answers to these unknown questions, existing studies typically investigate approaches to refusing to answer these questions. In this work, we propose a novel and scalable self-alignment method to utilize the LLM itself to enhance its response-ability to different types of unknown questions, being capable of not only refusing to answer but also providing explanation to the unanswerability of unknown questions. Specifically, the Self-Align method first employ a two-stage class-aware self-augmentation approach to generate a large amount of unknown question-response data. Then we conduct disparity-driven self-curation to select qualified data for fine-tuning the LLM itself for aligning the responses to unknown questions as desired. Experimental results on two datasets across four types of unknown questions validate the superiority of the Self-Align method over existing baselines in terms of three types of task formulation.

Meng Chu, Xuan Billy Zhang, Kevin Qinghong Lin et al.

As AI systems move from generating text to accomplishing goals through sustained interaction, the ability to model environment dynamics becomes a central bottleneck. Agents that manipulate objects, navigate software, coordinate with others, or design experiments require predictive environment models, yet the term world model carries different meanings across research communities. We introduce a "levels x laws" taxonomy organized along two axes. The first defines three capability levels: L1 Predictor, which learns one-step local transition operators; L2 Simulator, which composes them into multi-step, action-conditioned rollouts that respect domain laws; and L3 Evolver, which autonomously revises its own model when predictions fail against new evidence. The second identifies four governing-law regimes: physical, digital, social, and scientific. These regimes determine what constraints a world model must satisfy and where it is most likely to fail. Using this framework, we synthesize over 400 works and summarize more than 100 representative systems spanning model-based reinforcement learning, video generation, web and GUI agents, multi-agent social simulation, and AI-driven scientific discovery. We analyze methods, failure modes, and evaluation practices across level-regime pairs, propose decision-centric evaluation principles and a minimal reproducible evaluation package, and outline architectural guidance, open problems, and governance challenges. The resulting roadmap connects previously isolated communities and charts a path from passive next-step prediction toward world models that can simulate, and ultimately reshape, the environments in which agents operate.