Hongwei Feng

Zhouhong Gu, Xiaoxuan Zhu, Haoning Ye et al.

New Natural Langauge Process~(NLP) benchmarks are urgently needed to align with the rapid development of large language models (LLMs). We present Xiezhi, the most comprehensive evaluation suite designed to assess holistic domain knowledge. Xiezhi comprises multiple-choice questions across 516 diverse disciplines ranging from 13 different subjects with 249,587 questions and accompanied by Xiezhi-Specialty and Xiezhi-Interdiscipline, both with 15k questions. We conduct evaluation of the 47 cutting-edge LLMs on Xiezhi. Results indicate that LLMs exceed average performance of humans in science, engineering, agronomy, medicine, and art, but fall short in economics, jurisprudence, pedagogy, literature, history, and management. We anticipate Xiezhi will help analyze important strengths and shortcomings of LLMs, and the benchmark is released in~\url{https://github.com/MikeGu721/XiezhiBenchmark}.

Jialing Li, Zhouhong Gu, Yin Cai et al.

The burgeoning field of LLM-based Multi-Agent Systems (MAS) promises to tackle complex tasks through collaborative intelligence, yet fundamental questions regarding their scaling behavior and intrinsic collective dynamics remain underexplored. This paper systematically investigates how the performance of a homogeneous MAS evolves as the number of agents increases, isolating the variable of collaboration from model or knowledge heterogeneity. We propose the Sequential Iterative Multi-Agent System (SIMAS) framework, a minimalist architecture centered on sequential inter-agent communication, to clearly observe scaling effects. Through extensive experiments across diverse tasks and model scales, we establish that MAS performance does not scale monotonically with agent count but follows a pattern of diminishing returns, governed by a trade-off between collaborative synergy and coordination overhead. Our findings reveal that effective MAS requires a sufficiently capable base LLM, that task type critically modulates the optimal agent count, and that collective intelligence is an emergent property contingent on strategic interaction design rather than a guaranteed outcome of agent plurality. The performance degradation stems coordination overhead rather than merely long-context failure, and the scaling tendency generalizes across interaction architectures like structured debate topologies. This work provides a foundational understanding of MAS scaling laws, offering practical guidance for designing efficient collaborative systems and challenging the prevailing assumption that more agents invariably lead to better performance.

Zhouhong Gu, Sihang Jiang, Jingping Liu et al.

Taxonomy is formulated as directed acyclic concepts graphs or trees that support many downstream tasks. Many new coming concepts need to be added to an existing taxonomy. The traditional taxonomy expansion task aims only at finding the best position for new coming concepts in the existing taxonomy. However, they have two drawbacks when being applied to the real-scenarios. The previous methods suffer from low-efficiency since they waste much time when most of the new coming concepts are indeed noisy concepts. They also suffer from low-effectiveness since they collect training samples only from the existing taxonomy, which limits the ability of the model to mine more hypernym-hyponym relationships among real concepts. This paper proposes a pluggable framework called Generative Adversarial Network for Taxonomy Entering Evaluation (GANTEE) to alleviate these drawbacks. A generative adversarial network is designed in this framework by discriminative models to alleviate the first drawback and the generative model to alleviate the second drawback. Two discriminators are used in GANTEE to provide long-term and short-term rewards, respectively. Moreover, to further improve the efficiency, pre-trained language models are used to retrieve the representation of the concepts quickly. The experiments on three real-world large-scale datasets with two different languages show that GANTEE improves the performance of the existing taxonomy expansion methods in both effectiveness and efficiency.

Zhouhong Gu, Xiaoxuan Zhu, Haoning Ye et al.

Domain knowledge refers to the in-depth understanding, expertise, and familiarity with a specific subject, industry, field, or area of special interest. The existing benchmarks are all lack of an overall design for domain knowledge evaluation. Holding the belief that the real ability of domain language understanding can only be fairly evaluated by an comprehensive and in-depth benchmark, we introduces the Domma, a Domain Mastery Benchmark. DomMa targets at testing Large Language Models (LLMs) on their domain knowledge understanding, it features extensive domain coverage, large data volume, and a continually updated data set based on Chinese 112 first-level subject classifications. DomMa consist of 100,000 questions in both Chinese and English sourced from graduate entrance examinations and undergraduate exams in Chinese college. We have also propose designs to make benchmark and evaluation process more suitable to LLMs.

Zhouhong Gu, Sihang Jiang, Wenhao Huang et al.

The model's ability to understand synonymous expression is crucial in many kinds of downstream tasks. It will make the model to better understand the similarity between context, and more robust to the synonym substitution attack. However, many Pretrained Language Model (PLM) lack synonym knowledge due to limitation of small-scale synsets and PLM's pretraining objectives. In this paper, we propose a framework called Sem4SAP to mine synsets from Open Knowledge Graph (Open-KG) and using the mined synsets to do synonym-aware pretraining for language models. We propose to coarsly filter the content in Open-KG and use the frequency information to better help the clustering process under low-resource unsupervised conditions. We expand the mined synsets by migrating core semantics between synonymous expressions.We also propose two novel and effective synonym-aware pre-training methods for injecting synonym knowledge into PLMs.Extensive experiments demonstrate that Sem4SAP can dramatically outperform the original PLMs and other baselines on ten different tasks.

Yifeng Wang, Zhouhong Gu, Siwei Zhang et al.

Explainable fake news detection predicts the authenticity of news items with annotated explanations. Today, Large Language Models (LLMs) are known for their powerful natural language understanding and explanation generation abilities. However, presenting LLMs for explainable fake news detection remains two main challenges. Firstly, fake news appears reasonable and could easily mislead LLMs, leaving them unable to understand the complex news-faking process. Secondly, utilizing LLMs for this task would generate both correct and incorrect explanations, which necessitates abundant labor in the loop. In this paper, we propose LLM-GAN, a novel framework that utilizes prompting mechanisms to enable an LLM to become Generator and Detector and for realistic fake news generation and detection. Our results demonstrate LLM-GAN's effectiveness in both prediction performance and explanation quality. We further showcase the integration of LLM-GAN to a cloud-native AI platform to provide better fake news detection service in the cloud.

Tianjun Pan, Xuan Lin, Wenyan Yang et al.

Rubric-based evaluation has become a prevailing paradigm for evaluating instruction following in large language models (LLMs). Despite its widespread use, the reliability of these rubric-level evaluations remains unclear, calling for meta-evaluation. However, prior meta-evaluation efforts largely focus on the response level, failing to assess the fine-grained judgment accuracy that rubric-based evaluation relies on. To bridge this gap, we introduce RubricEval. Our benchmark features: (1) the first rubric-level meta-evaluation benchmark for instruction following, (2) diverse instructions and responses spanning multiple categories and model sources, and (3) a substantial set of 3,486 quality-controlled instances, along with Easy/Hard subsets that better differentiates judge performance. Our experiments reveal that rubric-level judging remains far from solved: even GPT-4o, a widely adopted judge in instruction-following benchmarks, achieves only 55.97% on Hard subset. Considering evaluation paradigm, rubric-level evaluation outperforms checklist-level, explicit reasoning improves accuracy, and both together reduce inter-judge variance. Through our established rubric taxonomy, we further identify common failure modes and offer actionable insights for reliable instruction-following evaluation.

Zhouhong Gu, Lin Zhang, Jiangjie Chen et al.

Detectives frequently engage in information detection and reasoning simultaneously when making decisions across various cases, especially when confronted with a vast amount of information. With the rapid development of large language models~(LLMs), evaluating how these models identify key information and reason to solve questions becomes increasingly relevant. We introduces the DetectBench, a reading comprehension dataset designed to assess a model's ability to jointly ability in key information detection and multi-hop reasoning when facing complex and implicit information. The DetectBench comprises 3,928 questions, each paired with a paragraph averaging 190 tokens in length. To enhance model's detective skills, we propose the Detective Thinking Framework. These methods encourage models to identify all possible clues within the context before reasoning. Our experiments reveal that existing models perform poorly in both information detection and multi-hop reasoning. However, the Detective Thinking Framework approach alleviates this issue.

Xintao Wang, Jian Yang, Weiyuan Li et al.

Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning and generation, serving as the foundation for advanced persona simulation and Role-Playing Language Agents (RPLAs). However, achieving authentic alignment with human cognitive and behavioral patterns remains a critical challenge for these agents. We present HumanLLM, a framework treating psychological patterns as interacting causal forces. We construct 244 patterns from ~12,000 academic papers and synthesize 11,359 scenarios where 2-5 patterns reinforce, conflict, or modulate each other, with multi-turn conversations expressing inner thoughts, actions, and dialogue. Our dual-level checklists evaluate both individual pattern fidelity and emergent multi-pattern dynamics, achieving strong human alignment (r=0.91) while revealing that holistic metrics conflate simulation accuracy with social desirability. HumanLLM-8B outperforms Qwen3-32B on multi-pattern dynamics despite 4x fewer parameters, demonstrating that authentic anthropomorphism requires cognitive modeling--simulating not just what humans do, but the psychological processes generating those behaviors.

Zhouhong Gu, Xiaoxuan Zhu, Yin Cai et al.

Large language model based multi-agent systems have demonstrated significant potential in social simulation and complex task resolution domains. However, current frameworks face critical challenges in system architecture design, cross-domain generalizability, and performance guarantees, particularly as task complexity and number of agents increases. We introduces AgentGroupChat-V2, a novel framework addressing these challenges through three core innovations: (1) a divide-and-conquer fully parallel architecture that decomposes user queries into hierarchical task forest structures enabling dependency management and distributed concurrent processing. (2) an adaptive collaboration engine that dynamically selects heterogeneous LLM combinations and interaction modes based on task characteristics. (3) agent organization optimization strategies combining divide-and-conquer approaches for efficient problem decomposition. Extensive experiments demonstrate AgentGroupChat-V2's superior performance across diverse domains, achieving 91.50% accuracy on GSM8K (exceeding the best baseline by 5.6 percentage points), 30.4% accuracy on competition-level AIME (nearly doubling other methods), and 79.20% pass@1 on HumanEval. Performance advantages become increasingly pronounced with higher task difficulty, particularly on Level 5 MATH problems where improvements exceed 11 percentage points compared to state-of-the-art baselines. These results confirm that AgentGroupChat-V2 provides a comprehensive solution for building efficient, general-purpose LLM multi-agent systems with significant advantages in complex reasoning scenarios. Code is available at https://github.com/MikeGu721/AgentGroupChat-V2.

Yin Cai, Zhouhong Gu, Zhaohan Du et al.

Large Language Models (LLMs) have shown remarkable capabilities in environmental perception, reasoning-based decision-making, and simulating complex human behaviors, particularly in interactive role-playing contexts. This paper introduces the Multiverse Interactive Role-play Ability General Evaluation (MIRAGE), a comprehensive framework designed to assess LLMs' proficiency in portraying advanced human behaviors through murder mystery games. MIRAGE features eight intricately crafted scripts encompassing diverse themes and styles, providing a rich simulation. To evaluate LLMs' performance, MIRAGE employs four distinct methods: the Trust Inclination Index (TII) to measure dynamics of trust and suspicion, the Clue Investigation Capability (CIC) to measure LLMs' capability of conducting information, the Interactivity Capability Index (ICI) to assess role-playing capabilities and the Script Compliance Index (SCI) to assess LLMs' capability of understanding and following instructions. Our experiments indicate that even popular models like GPT-4 face significant challenges in navigating the complexities presented by the MIRAGE. The datasets and simulation codes are available in \href{https://github.com/lime728/MIRAGE}{github}.

Jinghao Zhang, Sihang Jiang, Shiwei Guo et al.

As large language models (LLMs) are increasingly deployed in diverse cultural environments, evaluating their cultural understanding capability has become essential for ensuring trustworthy and culturally aligned applications. However, most existing benchmarks lack comprehensiveness and are challenging to scale and adapt across different cultural contexts, because their frameworks often lack guidance from well-established cultural theories and tend to rely on expert-driven manual annotations. To address these issues, we propose CultureScope, the most comprehensive evaluation framework to date for assessing cultural understanding in LLMs. Inspired by the cultural iceberg theory, we design a novel dimensional schema for cultural knowledge classification, comprising 3 layers and 140 dimensions, which guides the automated construction of culture-specific knowledge bases and corresponding evaluation datasets for any given languages and cultures. Experimental results demonstrate that our method can effectively evaluate cultural understanding. They also reveal that existing large language models lack comprehensive cultural competence, and merely incorporating multilingual data does not necessarily enhance cultural understanding. All code and data files are available at https://github.com/HoganZinger/Culture

Zhouhong Gu, Xingzhou Chen, Xiaoran Shi et al.

Recent advances in large language models have highlighted the critical need for precise control over model outputs through predefined constraints. While existing methods attempt to achieve this through either direct instruction-response synthesis or preferential response optimization, they often struggle with constraint understanding and adaptation. This limitation becomes particularly evident when handling fine-grained constraints, leading to either hallucination or brittle performance. We introduce Generative Adversarial Policy Optimization (GAPO), a novel framework that combines GAN-based training dynamics with an encoder-only reward model to progressively learn and adapt to increasingly complex constraints. GAPO leverages adversarial training to automatically generate training samples of varying difficulty while utilizing the encoder-only architecture to better capture prompt-response relationships. Extensive experiments demonstrate GAPO's superior performance across multiple benchmarks, particularly in scenarios requiring fine-grained constraint handling, where it significantly outperforms existing methods like PPO, DPO, and KTO. Our results suggest that GAPO's unique approach to preferential prompt learning offers a more robust and effective solution for controlling LLM outputs. Code is avaliable in https://github.com/MikeGu721/GAPO.

Lin Zhang, Zhouhong Gu, Suhang Zheng et al.

This paper presents LITE, an LLM-based evaluation method designed for efficient and flexible assessment of taxonomy quality. To address challenges in large-scale taxonomy evaluation, such as efficiency, fairness, and consistency, LITE adopts a top-down hierarchical evaluation strategy, breaking down the taxonomy into manageable substructures and ensuring result reliability through cross-validation and standardized input formats. LITE also introduces a penalty mechanism to handle extreme cases and provides both quantitative performance analysis and qualitative insights by integrating evaluation metrics closely aligned with task objectives. Experimental results show that LITE demonstrates high reliability in complex evaluation tasks, effectively identifying semantic errors, logical contradictions, and structural flaws in taxonomies, while offering directions for improvement. Code is available at https://github.com/Zhang-l-i-n/TAXONOMY_DETECT .

Lin Zhang, Zhouhong Gu, Xiaoran Shi et al.

As large language models (LLMs) advance, efficient knowledge evaluation becomes crucial to verifying their capabilities. Traditional methods, relying on benchmarks, face limitations such as high resource costs and information loss. We propose the Large-scale Reference-based Efficient Knowledge Evaluation for Large Language Model (RECKON), which directly uses reference data to evaluate models. RECKON organizes unstructured data into manageable units and generates targeted questions for each cluster, improving evaluation accuracy and efficiency. Experimental results show that RECKON reduces resource consumption by 56.5% compared to traditional methods while achieving over 97% accuracy across various domains, including world knowledge, code, legal, and biomedical datasets. Code is available at https://github.com/MikeGu721/reckon

Zhouhong Gu, Haoning Ye, Xingzhou Chen et al.

The effective utilization of structured data, integral to corporate data strategies, has been challenged by the rise of large language models (LLMs) capable of processing unstructured information. This shift prompts the question: can LLMs interpret structured data directly in its unstructured form? We propose an automatic evaluation data generation method for assessing LLMs' reasoning capabilities on structure-rich text to explore this. Our approach supports 8 structured languages and 29 tasks, generating data with adjustable complexity through controllable nesting and structural width. We introduce StrucText-Eval, a benchmark containing 5,800 pre-generated and annotated samples designed to evaluate how well LLMs understand and reason through structured text. StrucText-Eval is divided into two suites: a regular Test suite (3,712 samples) and a Test-Hard suite (2,088 samples), the latter emphasizing the gap between human and model performance on more complex tasks. Experimental results show that while open-source LLMs achieve a maximum accuracy of 74.9\% on the standard dataset, their performance drops significantly to 45.8\% on the harder dataset. In contrast, human participants reach an accuracy of 92.6\% on StrucText-Eval-Hard, highlighting LLMs' current limitations in handling intricate structural information. The benchmark and generation codes are open sourced in \url{https://github.com/MikeGu721/StrucText-Eval}

Xiaoxuan Zhu, Zhouhong Gu, Baiqian Wu et al.

Pre-training large language models (LLMs) necessitates enormous diverse textual corpora, making effective data selection a key challenge for balancing computational resources and model performance. Current methodologies primarily emphasize data quality metrics and mixing proportions, yet they fail to adequately capture the underlying semantic connections between training samples and quality disparities within individual domains. We introduce ToReMi (Topic-based Reweighting for Model improvement), a novel two-stage framework that dynamically adjusts training sample weights according to their topical associations and observed learning patterns. Our comprehensive experiments reveal that ToReMi variants consistently achieve superior performance over conventional pre-training approaches, demonstrating accelerated perplexity reduction across multiple domains and enhanced capabilities on downstream evaluation tasks. Code is available at https://github.com/zxx000728/ToReMi.

Zhouhong Gu, Xiaoxuan Zhu, Haoran Guo et al.

Language significantly influences the formation and evolution of Human emergent behavior, which is crucial in understanding collective intelligence within human societies. Considering that the study of how language affects human behavior needs to put it into the dynamic scenarios in which it is used, we introduce AgentGroupChat in this paper, a simulation that delves into the complex role of language in shaping collective behavior through interactive debate scenarios. Central to this simulation are characters engaging in dynamic conversation interactions. To enable simulation, we introduce the Verbal Strategist Agent, utilizing large language models to enhance interaction strategies by incorporating elements of persona and action. We set four narrative scenarios based on AgentGroupChat to demonstrate the simulation's capacity to mimic complex language use in group dynamics. Evaluations focus on aligning agent behaviors with human expectations and the emergence of collective behaviors within the simulation. Results reveal that emergent behaviors materialize from a confluence of factors: a conducive environment for extensive information exchange, characters with diverse traits, high linguistic comprehension, and strategic adaptability. During discussions on ``the impact of AI on humanity'' in AgentGroupChat simulation, philosophers commonly agreed that ``AI could enhance societal welfare with judicious limitations'' and even come to a conclusion that ``the essence of true intelligence encompasses understanding the necessity to constrain self abilities''. Additionally, in the competitive domain of casting for primary roles in films in AgentGroupChat, certain actors were ready to reduce their remuneration or accept lesser roles, motivated by their deep-seated desire to contribute to the project.

Jianchen Wang, Zhouhong Gu, Xiaoxuan Zhu et al.

Large Language Models have revolutionized numerous tasks with their remarkable efficacy. However, editing these models, crucial for rectifying outdated or erroneous information, often leads to a complex issue known as the ripple effect in the hidden space. While difficult to detect, this effect can significantly impede the efficacy of model editing tasks and deteriorate model performance. This paper addresses this scientific challenge by proposing a novel evaluation methodology, Graphical Impact Evaluation(GIE), which quantitatively evaluates the adaptations of the model and the subsequent impact of editing. Furthermore, we introduce the Selective Impact Revision(SIR), a model editing method designed to mitigate this ripple effect. Our comprehensive evaluations reveal that the ripple effect in the hidden space is a significant issue in all current model editing methods. However, our proposed methods, GIE and SIR, effectively identify and alleviate this issue, contributing to the advancement of LLM editing techniques.

Zhouhong Gu, Lin Zhang, Xiaoxuan Zhu et al.

Detecting evidence within the context is a key step in the process of reasoning task. Evaluating and enhancing the capabilities of LLMs in evidence detection will strengthen context-based reasoning performance. This paper proposes a benchmark called DetectBench for verifying the ability to detect and piece together implicit evidence within a long context. DetectBench contains 3,928 multiple-choice questions, with an average of 994 tokens per question. Each question contains an average of 4.55 pieces of implicit evidence, and solving the problem typically requires 7.62 logical jumps to find the correct answer. To enhance the performance of LLMs in evidence detection, this paper proposes Detective Reasoning Prompt and Finetune. Experiments demonstrate that the existing LLMs' abilities to detect evidence in long contexts are far inferior to humans. However, the Detective Reasoning Prompt effectively enhances the capability of powerful LLMs in evidence detection, while the Finetuning method shows significant effects in enhancing the performance of weaker LLMs. Moreover, when the abilities of LLMs in evidence detection are improved, their final reasoning performance is also enhanced accordingly.

Xiaoxuan Zhu, Zhouhong Gu, Sihang Jiang et al.

Online courses have significantly lowered the barrier to accessing education, yet the varying content quality of these videos poses challenges. In this work, we focus on the task of automatically evaluating the quality of video course content. We have constructed a dataset with a substantial collection of video courses and teaching materials. We propose three evaluation principles and design a new evaluation framework, \textit{VCEval}, based on these principles. The task is modeled as a multiple-choice question-answering task, with a language model serving as the evaluator. Our method effectively distinguishes video courses of different content quality and produces a range of interpretable results.