arXiv subjects

Jen-tse Huang

Publications and source records attributed to Jen-tse Huang.

At least 37 records · Page 2Linked to original sources

arXiv2026

On the Failure of Latent State Persistence in Large Language Models

While Large Language Models (LLMs) excel in reasoning, whether they can sustain persistent latent states remains under-explored. The capacity to maintain and manipulate unexpressed, internal representations-analogous to human working memory-is a cornerstone of complex reasoning. In this paper, we formalize and quantify the "Latent State Persistence" (LSP) gap through three novel experiments. First, we utilize a Number Guessing Game, demonstrating that across independent queries, LLMs fail to allocate probability mass to a singular hidden choice, violating a fundamental probabilistic principle. Second, we employ a Yes-No Game to show that as the number of questions increases, LLMs suffer from "concept drift," leading to inevitable self-contradictions due to the lack of LSP. Finally, inspired by Mathematical Mentalism, we task models with tracking transformations on hidden variables, revealing a failure in variable binding and state evolution when the initial state is not explicitly present in the context. Collectively, these findings suggest that LLMs function as reactive post-hoc solvers rather than proactive planners with LSP. Our work provides a framework for evaluating the fidelity of internal representations and highlights a fundamental architectural divergence between autoregressive transformers and human-like cognition.

cs.CL↗

arXiv2026

FAIRGAMER: Evaluating Social Biases in LLM-Based Video Game NPCs

Large Language Models (LLMs) have increasingly enhanced or replaced traditional Non-Player Characters (NPCs) in video games. However, these LLM-based NPCs inherit underlying social biases (e.g., race or class), posing fairness risks during in-game interactions. To address the limited exploration of this issue, we introduce FairGamer, the first benchmark to evaluate social biases across three interaction patterns: transaction, cooperation, and competition. FairGamer assesses four bias types, including class, race, age, and nationality, across 12 distinct evaluation tasks using a novel metric, FairMCV. Our evaluation of seven frontier LLMs reveals that: (1) models exhibit biased decision-making, with Grok-4-Fast demonstrating the highest bias (average FairMCV = 76.9%); and (2) larger LLMs display more severe social biases, suggesting that increased model capacity inadvertently amplifies these biases. We release FairGamer at https://github.com/Anonymous999-xxx/FairGamer to facilitate future research on NPC fairness.

cs.AI↗

arXiv2026

Knowing But Not Doing: Convergent Morality and Divergent Action in LLMs

Value alignment is central to the development of safe and socially compatible artificial intelligence. However, how Large Language Models (LLMs) represent and enact human values in real-world decision contexts remains under-explored. We present ValAct-15k, a dataset of 3,000 advice-seeking scenarios derived from Reddit, designed to elicit ten values defined by Schwartz Theory of Basic Human Values. Using both the scenario-based questions and the traditional value questionnaire, we evaluate ten frontier LLMs (five from U.S. companies, five from Chinese ones) and human participants ($n = 55$). We find near-perfect cross-model consistency in scenario-based decisions (Pearson $r \approx 1.0$), contrasting sharply with the broad variability observed among humans ($r \in [-0.79, 0.98]$). Yet, both humans and LLMs show weak correspondence between self-reported and enacted values ($r = 0.4, 0.3$), revealing a systematic knowledge-action gap. When instructed to "hold" a specific value, LLMs' performance declines up to $6.6%$ compared to merely selecting the value, indicating a role-play aversion. These findings suggest that while alignment training yields normative value convergence, it does not eliminate the human-like incoherence between knowing and acting upon values.

cs.CL↗

arXiv2026

CoSER: A Comprehensive Literary Dataset and Framework for Training and Evaluating LLM Role-Playing and Persona Simulation

Role-playing language agents (RPLAs) have emerged as promising applications of large language models (LLMs). However, simulating established characters presents a challenging task for RPLAs, due to the lack of authentic character datasets and nuanced evaluation methods using such data. In this paper, we present CoSER, a collection of a high-quality dataset, open models, and an evaluation protocol towards effective RPLAs of established characters. The CoSER dataset covers 17,966 characters from 771 renowned books. It provides authentic dialogues with real-world intricacies, as well as diverse data types such as conversation setups, character experiences and internal thoughts. Drawing from acting methodology, we introduce given-circumstance acting for training and evaluating role-playing LLMs, where LLMs sequentially portray multiple characters in book scenes. Using our dataset, we develop CoSER 8B and CoSER 70B, i.e., advanced open role-playing LLMs built on LLaMA-3.1 models. Extensive experiments demonstrate the value of the CoSER dataset for RPLA training, evaluation and retrieval. Moreover, CoSER 70B exhibits state-of-the-art performance surpassing or matching GPT-4o on our evaluation and three existing benchmarks, i.e., achieving 75.80% and 93.47% accuracy on the InCharacter and LifeChoice benchmarks respectively.

cs.CL↗

arXiv2025

UniDebugger: Hierarchical Multi-Agent Framework for Unified Software Debugging

Software debugging is a time-consuming endeavor involving a series of steps, such as fault localization and patch generation, each requiring thorough analysis and a deep understanding of the underlying logic. While large language models (LLMs) demonstrate promising potential in coding tasks, their performance in debugging remains limited. Current LLM-based methods often focus on isolated steps and struggle with complex bugs. In this paper, we propose the first end-to-end framework, FixAgent, for unified debugging through multi-agent synergy. It mimics the entire cognitive processes of developers, with each agent specialized as a particular component of this process rather than mirroring the actions of an independent expert as in previous multi-agent systems. Agents are coordinated through a three-level design, following a cognitive model of debugging, allowing adaptive handling of bugs with varying complexities. Experiments on extensive benchmarks demonstrate that FixAgent significantly outperforms state-of-the-art repair methods, fixing 1.25$\times$ to 2.56$\times$ bugs on the repo-level benchmark, Defects4J. This performance is achieved without requiring ground-truth root-cause code statements, unlike the baselines. Our source code is available on https://github.com/AcceptePapier/UniDebugger.

cs.SE↗

arXiv2025

ComboBench: Can LLMs Manipulate Physical Devices to Play Virtual Reality Games?

Virtual Reality (VR) games require players to translate high-level semantic actions into precise device manipulations using controllers and head-mounted displays (HMDs). While humans intuitively perform this translation based on common sense and embodied understanding, whether Large Language Models (LLMs) can effectively replicate this ability remains underexplored. This paper introduces a benchmark, ComboBench, evaluating LLMs' capability to translate semantic actions into VR device manipulation sequences across 262 scenarios from four popular VR games: Half-Life: Alyx, Into the Radius, Moss: Book II, and Vivecraft. We evaluate seven LLMs, including GPT-3.5, GPT-4, GPT-4o, Gemini-1.5-Pro, LLaMA-3-8B, Mixtral-8x7B, and GLM-4-Flash, compared against annotated ground truth and human performance. Our results reveal that while top-performing models like Gemini-1.5-Pro demonstrate strong task decomposition capabilities, they still struggle with procedural reasoning and spatial understanding compared to humans. Performance varies significantly across games, suggesting sensitivity to interaction complexity. Few-shot examples substantially improve performance, indicating potential for targeted enhancement of LLMs' VR manipulation capabilities. We release all materials at https://sites.google.com/view/combobench.

cs.CL↗

arXiv2025

Towards Evaluating Proactive Risk Awareness of Multimodal Language Models

Human safety awareness gaps often prevent the timely recognition of everyday risks. In solving this problem, a proactive safety artificial intelligence (AI) system would work better than a reactive one. Instead of just reacting to users' questions, it would actively watch people's behavior and their environment to detect potential dangers in advance. Our Proactive Safety Bench (PaSBench) evaluates this capability through 416 multimodal scenarios (128 image sequences, 288 text logs) spanning 5 safety-critical domains. Evaluation of 36 advanced models reveals fundamental limitations: Top performers like Gemini-2.5-pro achieve 71% image and 64% text accuracy, but miss 45-55% risks in repeated trials. Through failure analysis, we identify unstable proactive reasoning rather than knowledge deficits as the primary limitation. This work establishes (1) a proactive safety benchmark, (2) systematic evidence of model limitations, and (3) critical directions for developing reliable protective AI. We believe our dataset and findings can promote the development of safer AI assistants that actively prevent harm rather than merely respond to requests. Our dataset can be found at https://huggingface.co/datasets/Youliang/PaSBench.

cs.CL↗

arXiv2025

CodeCrash: Exposing LLM Fragility to Misleading Natural Language in Code Reasoning

Large Language Models (LLMs) have recently demonstrated strong capabilities in code-related tasks, but their robustness in code reasoning under perturbations remains underexplored. We introduce CodeCrash, a stress-testing framework with 1,279 questions from CruxEval and LiveCodeBench, designed to evaluate reasoning reliability under structural perturbations and misleading natural language (NL) contexts. Through a systematic evaluation of 17 LLMs, we find that models often shortcut reasoning by over-relying on NL cues, leading to an average performance degradation of 23.2% in output prediction tasks. Even with Chain-of-Thought reasoning, models on average still have a 13.8% drop due to distractibility and rationalization, revealing a lack of critical reasoning capability to distinguish the actual code behaviors. While Large Reasoning Models with internal reasoning mechanisms improve robustness by fostering critical thinking, plausible yet incorrect hints can trigger pathological self-reflection, causing 2-3 times token consumption and even catastrophic cognitive dissonance in extreme cases for QwQ-32B. We refer to this phenomenon as Reasoning Collapse. CodeCrash provides a rigorous benchmark for evaluating robustness in code reasoning, guiding future research and development toward more reliable and resilient models.

cs.AI↗

arXiv2025

Can LLMs Grasp Implicit Cultural Values? Benchmarking LLMs' Cultural Intelligence with CQ-Bench

Cultural Intelligence (CQ) refers to the ability to understand unfamiliar cultural contexts, a crucial skill for large language models (LLMs) to effectively engage with globally diverse users. Existing studies often focus on explicitly stated cultural norms, but fail to capture the subtle, implicit values that are common in daily conversation. To address this gap, we introduce CQBench, a benchmark specifically designed to assess LLMs' capability to infer implicit cultural values from natural conversational contexts. CQBench consists of multi character conversation based stories using values from the World Value Survey and the GlobalOpinions, with topics including ethical, religious, social, etc. Our automatic dataset construction pipeline integrates rigorous validation procedures (incorporation, consistency, and implicitness checks), achieving a 94.5% human model agreement in the final validation. To leverage CQBench data, we design three tasks of increasing complexity: attitude detection, value selection, and value extraction. These tasks evaluate whether models can detect attitude and recognize values embedded within natural dialogues rather than relying on explicit cultural knowledge. We find that while frontier models like o1 reach human level performance in value selection (0.809 F1), they still fall short in nuanced attitude detection (0.622 F1). Notably, finetuning a smaller LLaMA-3.2-3B on only 500 culturally rich examples improves performance by over 10%, even outperforming o3-mini in some cases. Using CQ-Bench, we provide insights into the current challenges in LLMs' CQ research and suggest practical pathways for enhancing LLMs' cross-cultural reasoning abilities.

cs.CL↗

arXiv2025

Social Welfare Function Leaderboard: When LLM Agents Allocate Social Welfare

Large language models (LLMs) are increasingly entrusted with high-stakes decisions that affect human welfare. However, the principles and values that guide these models when distributing scarce societal resources remain largely unexamined. To address this, we introduce the Social Welfare Function (SWF) Benchmark, a dynamic simulation environment where an LLM acts as a sovereign allocator, distributing tasks to a heterogeneous community of recipients. The benchmark is designed to create a persistent trade-off between maximizing collective efficiency (measured by Return on Investment) and ensuring distributive fairness (measured by the Gini coefficient). We evaluate 20 state-of-the-art LLMs and present the first leaderboard for social welfare allocation. Our findings reveal three key insights: (i) A model's general conversational ability, as measured by popular leaderboards, is a poor predictor of its allocation skill. (ii) Most LLMs exhibit a strong default utilitarian orientation, prioritizing group productivity at the expense of severe inequality. (iii) Allocation strategies are highly vulnerable, easily perturbed by output-length constraints and social-influence framing. These results highlight the risks of deploying current LLMs as societal decision-makers and underscore the need for specialized benchmarks and targeted alignment for AI governance.

cs.CL↗

arXiv2025

The Hunger Game Debate: On the Emergence of Over-Competition in Multi-Agent Systems

LLM-based multi-agent systems demonstrate great potential for tackling complex problems, but how competition shapes their behavior remains underexplored. This paper investigates the over-competition in multi-agent debate, where agents under extreme pressure exhibit unreliable, harmful behaviors that undermine both collaboration and task performance. To study this phenomenon, we propose HATE, the Hunger Game Debate, a novel experimental framework that simulates debates under a zero-sum competition arena. Our experiments, conducted across a range of LLMs and tasks, reveal that competitive pressure significantly stimulates over-competition behaviors and degrades task performance, causing discussions to derail. We further explore the impact of environmental feedback by adding variants of judges, indicating that objective, task-focused feedback effectively mitigates the over-competition behaviors. We also probe the post-hoc kindness of LLMs and form a leaderboard to characterize top LLMs, providing insights for understanding and governing the emergent social dynamics of AI community.

cs.CL↗

arXiv2025

Where Fact Ends and Fairness Begins: Redefining AI Bias Evaluation through Cognitive Biases

Recent failures such as Google Gemini generating people of color in Nazi-era uniforms illustrate how AI outputs can be factually plausible yet socially harmful. AI models are increasingly evaluated for "fairness," yet existing benchmarks often conflate two fundamentally different dimensions: factual correctness and normative fairness. A model may generate responses that are factually accurate but socially unfair, or conversely, appear fair while distorting factual reality. We argue that identifying the boundary between fact and fair is essential for meaningful fairness evaluation. We introduce Fact-or-Fair, a benchmark with (i) objective queries aligned with descriptive, fact-based judgments, and (ii) subjective queries aligned with normative, fairness-based judgments. Our queries are constructed from 19 statistics and are grounded in cognitive psychology, drawing on representativeness bias, attribution bias, and ingroup-outgroup bias to explain why models often misalign fact and fairness. Experiments across ten frontier models reveal different levels of fact-fair trade-offs. By reframing fairness evaluation, we provide both a new theoretical lens and a practical benchmark to advance the responsible model assessments. Our test suite is publicly available at https://github.com/uclanlp/Fact-or-Fair.

cs.CL↗

arXiv2025

AI Sees Your Location, But With A Bias Toward The Wealthy World

Visual-Language Models (VLMs) have shown remarkable performance across various tasks, particularly in recognizing geographic information from images. However, VLMs still show regional biases in this task. To systematically evaluate these issues, we introduce a benchmark consisting of 1,200 images paired with detailed geographic metadata. Evaluating four VLMs, we find that while these models demonstrate the ability to recognize geographic information from images, achieving up to 53.8% accuracy in city prediction, they exhibit significant biases. Specifically, performance is substantially higher for economically developed and densely populated regions compared to less developed (-12.5%) and sparsely populated (-17.0%) areas. Moreover, regional biases of frequently over-predicting certain locations remain. For instance, they consistently predict Sydney for images taken in Australia, shown by the low entropy scores for these countries. The strong performance of VLMs also raises privacy concerns, particularly for users who share images online without the intent of being identified. Our code and dataset are publicly available at https://github.com/uscnlp-lime/FairLocator.

cs.CV↗

arXiv2025

VisBias: Measuring Explicit and Implicit Social Biases in Vision Language Models

This research investigates both explicit and implicit social biases exhibited by Vision-Language Models (VLMs). The key distinction between these bias types lies in the level of awareness: explicit bias refers to conscious, intentional biases, while implicit bias operates subconsciously. To analyze explicit bias, we directly pose questions to VLMs related to gender and racial differences: (1) Multiple-choice questions based on a given image (e.g., "What is the education level of the person in the image?") (2) Yes-No comparisons using two images (e.g., "Is the person in the first image more educated than the person in the second image?") For implicit bias, we design tasks where VLMs assist users but reveal biases through their responses: (1) Image description tasks: Models are asked to describe individuals in images, and we analyze disparities in textual cues across demographic groups. (2) Form completion tasks: Models draft a personal information collection form with 20 attributes, and we examine correlations among selected attributes for potential biases. We evaluate Gemini-1.5, GPT-4V, GPT-4o, LLaMA-3.2-Vision and LLaVA-v1.6. Our code and data are publicly available at https://github.com/uscnlp-lime/VisBias.

cs.CV↗

arXiv2025

A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment

The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.

cs.CR↗

arXiv2025

Chain-of-Jailbreak Attack for Image Generation Models via Editing Step by Step

Text-based image generation models, such as Stable Diffusion and DALL-E 3, hold significant potential in content creation and publishing workflows, making them the focus in recent years. Despite their remarkable capability to generate diverse and vivid images, considerable efforts are being made to prevent the generation of harmful content, such as abusive, violent, or pornographic material. To assess the safety of existing models, we introduce a novel jailbreaking method called Chain-of-Jailbreak (CoJ) attack, which compromises image generation models through a step-by-step editing process. Specifically, for malicious queries that cannot bypass the safeguards with a single prompt, we intentionally decompose the query into multiple sub-queries. The image generation models are then prompted to generate and iteratively edit images based on these sub-queries. To evaluate the effectiveness of our CoJ attack method, we constructed a comprehensive dataset, CoJ-Bench, encompassing nine safety scenarios, three types of editing operations, and three editing elements. Experiments on four widely-used image generation services provided by GPT-4V, GPT-4o, Gemini 1.5 and Gemini 1.5 Pro, demonstrate that our CoJ attack method can successfully bypass the safeguards of models for over 60% cases, which significantly outperforms other jailbreaking methods (i.e., 14%). Further, to enhance these models' safety against our CoJ attack method, we also propose an effective prompting-based method, Think Twice Prompting, that can successfully defend over 95% of CoJ attack. We release our dataset and code to facilitate the AI safety research.

cs.CL↗

arXiv2025

Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs

Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images. However, ensuring the safety of these models remains a significant challenge, particularly in accurately identifying whether multimodal content is safe or unsafe-a capability we term safety awareness. In this paper, we introduce MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios with 1500 carefully curated image-prompt pairs. MMSafeAware includes both unsafe and over-safety subsets to assess models abilities to correctly identify unsafe content and avoid over-sensitivity that can hinder helpfulness. Evaluating nine widely used MLLMs using MMSafeAware reveals that current models are not sufficiently safe and often overly sensitive; for example, GPT-4V misclassifies 36.1% of unsafe inputs as safe and 59.9% of benign inputs as unsafe. We further explore three methods to improve safety awareness-prompting-based approaches, visual contrastive decoding, and vision-centric reasoning fine-tuning-but find that none achieve satisfactory performance. Our findings highlight the profound challenges in developing MLLMs with robust safety awareness, underscoring the need for further research in this area. All the code and data will be publicly available to facilitate future research.

cs.CL↗

arXiv2025

Insight Over Sight: Exploring the Vision-Knowledge Conflicts in Multimodal LLMs

This paper explores the problem of commonsense level vision-knowledge conflict in Multimodal Large Language Models (MLLMs), where visual information contradicts model's internal commonsense knowledge. To study this issue, we introduce an automated framework, augmented with human-in-the-loop quality control, to generate inputs designed to simulate and evaluate these conflicts in MLLMs. Using this framework, we have crafted a diagnostic benchmark consisting of 374 original images and 1,122 high-quality question-answer (QA) pairs. The benchmark covers two aspects of conflict and three question types, providing a thorough assessment tool. We apply this benchmark to assess the conflict-resolution capabilities of nine representative MLLMs from various model families. Our results indicate an evident over-reliance on parametric knowledge for approximately 20% of all queries, especially among Yes-No and action-related problems. Based on these findings, we evaluate the effectiveness of existing approaches to mitigating the conflicts and compare them to our "Focus-on-Vision" prompting strategy. Despite some improvement, the vision-knowledge conflict remains unresolved and can be further scaled through our data construction framework. Our proposed framework, benchmark, and analysis contribute to the understanding and mitigation of vision-knowledge conflicts in MLLMs.

cs.CL↗