Zekun Wu

Rishi Kalra, Zekun Wu, Ayesha Gulley et al.

Large Language Models (LLMs) face limitations in AI legal and policy applications due to outdated knowledge, hallucinations, and poor reasoning in complex contexts. Retrieval-Augmented Generation (RAG) systems address these issues by incorporating external knowledge, but suffer from retrieval errors, ineffective context integration, and high operational costs. This paper presents the Hybrid Parameter-Adaptive RAG (HyPA-RAG) system, designed for the AI legal domain, with NYC Local Law 144 (LL144) as the test case. HyPA-RAG integrates a query complexity classifier for adaptive parameter tuning, a hybrid retrieval approach combining dense, sparse, and knowledge graph methods, and a comprehensive evaluation framework with tailored question types and metrics. Testing on LL144 demonstrates that HyPA-RAG enhances retrieval accuracy, response fidelity, and contextual precision, offering a robust and adaptable solution for high-stakes legal and policy applications.

Theo King, Zekun Wu, Adriano Koshiyama et al.

Stereotypes are generalised assumptions about societal groups, and even state-of-the-art LLMs using in-context learning struggle to identify them accurately. Due to the subjective nature of stereotypes, where what constitutes a stereotype can vary widely depending on cultural, social, and individual perspectives, robust explainability is crucial. Explainable models ensure that these nuanced judgments can be understood and validated by human users, promoting trust and accountability. We address these challenges by introducing HEARTS (Holistic Framework for Explainable, Sustainable, and Robust Text Stereotype Detection), a framework that enhances model performance, minimises carbon footprint, and provides transparent, interpretable explanations. We establish the Expanded Multi-Grain Stereotype Dataset (EMGSD), comprising 57,201 labelled texts across six groups, including under-represented demographics like LGBTQ+ and regional stereotypes. Ablation studies confirm that BERT models fine-tuned on EMGSD outperform those trained on individual components. We then analyse a fine-tuned, carbon-efficient ALBERT-V2 model using SHAP to generate token-level importance values, ensuring alignment with human understanding, and calculate explainability confidence scores by comparing SHAP and LIME outputs...

Mengfei Liang, Archish Arun, Zekun Wu et al.

Hallucination, the generation of factually incorrect content, is a growing challenge in Large Language Models (LLMs). Existing detection and mitigation methods are often isolated and insufficient for domain-specific needs, lacking a standardized pipeline. This paper introduces THaMES (Tool for Hallucination Mitigations and EvaluationS), an integrated framework and library addressing this gap. THaMES offers an end-to-end solution for evaluating and mitigating hallucinations in LLMs, featuring automated test set generation, multifaceted benchmarking, and adaptable mitigation strategies. It automates test set creation from any corpus, ensuring high data quality, diversity, and cost-efficiency through techniques like batch processing, weighted sampling, and counterfactual validation. THaMES assesses a model's ability to detect and reduce hallucinations across various tasks, including text generation and binary classification, applying optimal mitigation strategies like In-Context Learning (ICL), Retrieval Augmented Generation (RAG), and Parameter-Efficient Fine-tuning (PEFT). Evaluations of state-of-the-art LLMs using a knowledge base of academic papers, political news, and Wikipedia reveal that commercial models like GPT-4o benefit more from RAG than ICL, while open-weight models like Llama-3.1-8B-Instruct and Mistral-Nemo gain more from ICL. Additionally, PEFT significantly enhances the performance of Llama-3.1-8B-Instruct in both evaluation tasks.

Zekun Wu, Ze Wang, Seonglae Cho et al.

When a tool-calling agent picks the wrong tool, the failure is invisible until execution: the email gets sent, the meeting gets missed. Probing 12 instruction-tuned models across Gemma 3, Qwen 3, Qwen 2.5, and Llama 3.1 (270M to 27B), we find the identity of the chosen tool is linearly readable and steerable inside the model. Adding the mean-difference between two tools' average internal activations switches which tool the model selects at 77-100% accuracy on name-only single-turn prompts (93-100% at 4B+), and the JSON arguments that follow autoregressively match the new tool's schema, so flipping the name is enough. The same per-tool means also flag likely errors before they happen: on Gemma 3 12B and 27B, queries where the gap between the top-1 and top-2 tool is smallest produce 14-21x more wrong calls than queries with the largest gap. The causal effect concentrates along one direction, the row of the output layer that produces the target tool's first token: a unit vector along it at matched magnitude already reaches 93-100%, while what is left over leaves the choice almost untouched. Activation patching localises this to a small set of mid- and late-layer attention heads, and a within-topic probe across 14 same-domain $τ$-bench airline tools reaches top-1 61-89% across five 4B-14B models, ruling out the reading that we are just moving the model along a topic axis. Even base models encode the right tool before they can emit it: cosine readout from the internal state recovers 69-82% on BFCL while base generation reaches only 2-10%, suggesting pretraining forms the representation and instruction tuning later wires it to the output. We measure tool identity selection and JSON schema correctness in single-turn fixed-menu settings; multi-turn agentic transfer is more fragile and is discussed in Limitations.

Xin Guan, Ze Wang, Nathaniel Demchak et al.

The development of unbiased large language models is widely recognized as crucial, yet existing benchmarks fall short in detecting biases due to limited scope, contamination, and lack of a fairness baseline. SAGED(bias) is the first holistic benchmarking pipeline to address these problems. The pipeline encompasses five core stages: scraping materials, assembling benchmarks, generating responses, extracting numeric features, and diagnosing with disparity metrics. SAGED includes metrics for max disparity, such as impact ratio, and bias concentration, such as Max Z-scores. Noticing that metric tool bias and contextual bias in prompts can distort evaluation, SAGED implements counterfactual branching and baseline calibration for mitigation. For demonstration, we use SAGED on G20 Countries with popular 8b-level models including Gemma2, Llama3.1, Mistral, and Qwen2. With sentiment analysis, we find that while Mistral and Qwen2 show lower max disparity and higher bias concentration than Gemma2 and Llama3.1, all models are notably biased against countries like Russia and (except for Qwen2) China. With further experiments to have models role-playing U.S. presidents, we see bias amplifies and shifts in heterogeneous directions. Moreover, we see Qwen2 and Mistral not engage in role-playing, while Llama3.1 and Gemma2 role-play Trump notably more intensively than Biden and Harris, indicating role-playing performance bias in these models.

Navya Jain, Zekun Wu, Cristian Munoz et al.

The manipulation of the personality traits of large language models (LLMs) has emerged as a key area of research. Methods like prompt-based In-Context Knowledge Editing (IKE) and gradient-based Model Editor Networks (MEND) have been explored but show irregularity and variability; IKE depends on the prompt, leading to variability and sensitivity, while MEND yields inconsistent and gibberish outputs. To address this, we employed Opinion QA Based Parameter-Efficient Fine-Tuning (PEFT), specifically Quantized Low-Rank Adaptation (QLoRA), to manipulate the Big Five personality traits: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. After PEFT, models such as Mistral-7B-Instruct and LLaMA-2-7B-chat showed a latent behaviour by generating emojis for certain traits, despite no emojis being present in the PEFT data. For instance, LLaMA-2-7B-chat generated emojis in 99.5\% of extraversion-related test instances, while Mistral-7B-Instruct did so in 92.5\% of openness-related test instances. ICL Explainability analysis indicated that the LLMs used emojis intentionally to express these traits. Mechanistic Interpretability analysis showed that this latent behaviour of LLMs could be traced to specific neurons that became activated or amplified after PEFT. This paper provides a number of novel contributions. First, introducing an Opinion QA dataset for PEFT-driven personality manipulation; second, developing metric models to benchmark LLM personality traits; third, demonstrating PEFT's superiority over IKE in personality manipulation; and finally, analysing and validating emoji usage through explainability methods such as Mechanistic Interpretability and In-context learning Explainability methods.

Zekun Wu, Seonglae Cho, Umar Mohammed et al.

Open-source AI libraries are foundational to modern AI systems, yet they present significant, underexamined risks spanning security, licensing, maintenance, supply chain integrity, and regulatory compliance. We introduce LibVulnWatch, a system that leverages recent advances in large language models and agentic workflows to perform deep, evidence-based evaluations of these libraries. Built on a graph-based orchestration of specialized agents, the framework extracts, verifies, and quantifies risk using information from repositories, documentation, and vulnerability databases. LibVulnWatch produces reproducible, governance-aligned scores across five critical domains, publishing results to a public leaderboard for ongoing ecosystem monitoring. Applied to 20 widely used libraries, including ML frameworks, LLM inference engines, and agent orchestration tools, our approach covers up to 88% of OpenSSF Scorecard checks while surfacing up to 19 additional risks per library, such as critical RCE vulnerabilities, missing SBOMs, and regulatory gaps. By integrating advanced language technologies with the practical demands of software risk assessment, this work demonstrates a scalable, transparent mechanism for continuous supply chain evaluation and informed library selection.

Airlie Hilliard, Cristian Munoz, Zekun Wu et al.

Large Language Models (LLMs) are increasingly being utilized by both candidates and employers in the recruitment context. However, with this comes numerous ethical concerns, particularly related to the lack of transparency in these "black-box" models. Although previous studies have sought to increase the transparency of these models by investigating the personality traits of LLMs, many of the previous studies have provided them with personality assessments to complete. On the other hand, this study seeks to obtain a better understanding of such models by examining their output variations based on different input prompts. Specifically, we use a novel elicitation approach using prompts derived from common interview questions, as well as prompts designed to elicit particular Big Five personality traits to examine whether the models were susceptible to trait-activation like humans are, to measure their personality based on the language used in their outputs. To do so, we repeatedly prompted multiple LMs with different parameter sizes, including Llama-2, Falcon, Mistral, Bloom, GPT, OPT, and XLNet (base and fine tuned versions) and examined their personality using classifiers trained on the myPersonality dataset. Our results reveal that, generally, all LLMs demonstrate high openness and low extraversion. However, whereas LMs with fewer parameters exhibit similar behaviour in personality traits, newer and LMs with more parameters exhibit a broader range of personality traits, with increased agreeableness, emotional stability, and openness. Furthermore, a greater number of parameters is positively associated with openness and conscientiousness. Moreover, fine-tuned models exhibit minor modulations in their personality traits, contingent on the dataset. Implications and directions for future research are discussed.

Ze Wang, Zekun Wu, Jeremy Zhang et al.

Model collapse, a phenomenon characterized by performance degradation due to iterative training on synthetic data, has been widely studied. However, its implications for bias amplification, the progressive intensification of pre-existing societal biases in Large Language Models (LLMs), remain significantly underexplored, despite the growing influence of LLMs in shaping online discourse. In this paper, we introduce a open, generational, and long-context benchmark specifically designed to measure political bias amplification in LLMs, leveraging sentence continuation tasks derived from a comprehensive dataset of U.S. political news. Our empirical study using GPT-2 reveals consistent and substantial political bias intensification (e.g., right-leaning amplification) over iterative synthetic training cycles. We evaluate three mitigation strategies, Overfitting, Preservation, and Accumulation, and demonstrate that bias amplification persists independently of model collapse, even when the latter is effectively controlled. Furthermore, we propose a mechanistic analysis approach that identifies neurons correlated with specific phenomena during inference through regression and statistical tests. This analysis uncovers largely distinct neuron populations driving bias amplification and model collapse, underscoring fundamentally different underlying mechanisms. Finally, we supplement our empirical findings with theoretical intuition that explains the separate origins of these phenomena, guiding targeted strategies for bias mitigation.

Nathaniel Demchak, Xin Guan, Zekun Wu et al.

Open-generation bias benchmarks evaluate social biases in Large Language Models (LLMs) by analyzing their outputs. However, the classifiers used in analysis often have inherent biases, leading to unfair conclusions. This study examines such biases in open-generation benchmarks like BOLD and SAGED. Using the MGSD dataset, we conduct two experiments. The first uses counterfactuals to measure prediction variations across demographic groups by altering stereotype-related prefixes. The second applies explainability tools (SHAP) to validate that the observed biases stem from these counterfactuals. Results reveal unequal treatment of demographic descriptors, calling for more robust bias metric models.

Zekun Wu, Sahan Bulathwela, Maria Perez-Ortiz et al.

Stereotype detection is a challenging and subjective task, as certain statements, such as "Black people like to play basketball," may not appear overtly toxic but still reinforce racial stereotypes. With the increasing prevalence of large language models (LLMs) in human-facing artificial intelligence (AI) applications, detecting these types of biases is essential. However, LLMs risk perpetuating and amplifying stereotypical outputs derived from their training data. A reliable stereotype detector is crucial for benchmarking bias, monitoring model input and output, filtering training data, and ensuring fairer model behavior in downstream applications. This paper introduces the Multi-Grain Stereotype (MGS) dataset, consisting of 51,867 instances across gender, race, profession, religion, and other stereotypes, curated from multiple existing datasets. We evaluate various machine learning approaches to establish baselines and fine-tune language models of different architectures and sizes, presenting a suite of stereotype multiclass classifiers trained on the MGS dataset. Given the subjectivity of stereotypes, explainability is essential to align model learning with human understanding of stereotypes. We employ explainable AI (XAI) tools, including SHAP, LIME, and BertViz, to assess whether the model's learned patterns align with human intuitions about stereotypes.Additionally, we develop stereotype elicitation prompts and benchmark the presence of stereotypes in text generation tasks using popular LLMs, employing the best-performing stereotype classifiers.

Yuxuan Luo, Zekun Wu, Zhouhui Lian · pku

Human-like planning skills and dexterous manipulation have long posed challenges in the fields of robotics and artificial intelligence (AI). The task of reinterpreting calligraphy presents a formidable challenge, as it involves the decomposition of strokes and dexterous utensil control. Previous efforts have primarily focused on supervised learning of a single instrument, limiting the performance of robots in the realm of cross-domain text replication. To address these challenges, we propose CalliRewrite: a coarse-to-fine approach for robot arms to discover and recover plausible writing orders from diverse calligraphy images without requiring labeled demonstrations. Our model achieves fine-grained control of various writing utensils. Specifically, an unsupervised image-to-sequence model decomposes a given calligraphy glyph to obtain a coarse stroke sequence. Using an RL algorithm, a simulated brush is fine-tuned to generate stylized trajectories for robotic arm control. Evaluation in simulation and physical robot scenarios reveals that our method successfully replicates unseen fonts and styles while achieving integrity in unknown characters.

Xingrui Gu, Zhixuan Wang, Irisa Jin et al.

This research presents a novel multimodal data fusion methodology for pain behavior recognition, integrating statistical correlation analysis with human-centered insights. Our approach introduces two key innovations: 1) integrating data-driven statistical relevance weights into the fusion strategy to effectively utilize complementary information from heterogeneous modalities, and 2) incorporating human-centric movement characteristics into multimodal representation learning for detailed modeling of pain behaviors. Validated across various deep learning architectures, our method demonstrates superior performance and broad applicability. We propose a customizable framework that aligns each modality with a suitable classifier based on statistical significance, advancing personalized and effective multimodal fusion. Furthermore, our methodology provides explainable analysis of multimodal data, contributing to interpretable and explainable AI in healthcare. By highlighting the importance of data diversity and modality-specific representations, we enhance traditional fusion techniques and set new standards for recognizing complex pain behaviors. Our findings have significant implications for promoting patient-centered healthcare interventions and supporting explainable clinical decision-making.

Ze Wang, Zekun Wu, Xin Guan et al.

The use of Large Language Models (LLMs) in hiring has led to legislative actions to protect vulnerable demographic groups. This paper presents a novel framework for benchmarking hierarchical gender hiring bias in Large Language Models (LLMs) for resume scoring, revealing significant issues of reverse gender hiring bias and overdebiasing. Our contributions are fourfold: Firstly, we introduce a new construct grounded in labour economics, legal principles, and critiques of current bias benchmarks: hiring bias can be categorized into two types: Level bias (difference in the average outcomes between demographic counterfactual groups) and Spread bias (difference in the variance of outcomes between demographic counterfactual groups); Level bias can be further subdivided into statistical bias (i.e. changing with non-demographic content) and taste-based bias (i.e. consistent regardless of non-demographic content). Secondly, the framework includes rigorous statistical and computational hiring bias metrics, such as Rank After Scoring (RAS), Rank-based Impact Ratio, Permutation Test, and Fixed Effects Model. Thirdly, we analyze gender hiring biases in ten state-of-the-art LLMs. Seven out of ten LLMs show significant biases against males in at least one industry. An industry-effect regression reveals that the healthcare industry is the most biased against males. Moreover, we found that the bias performance remains invariant with resume content for eight out of ten LLMs. This indicates that the bias performance measured in this paper might apply to other resume datasets with different resume qualities. Fourthly, we provide a user-friendly demo and resume dataset to support the adoption and practical use of the framework, which can be generalized to other social traits and tasks.