Papers by Xuezhi Cao
Leveraging Dual Process Theory in Language Agent Framework for Real-time Simultaneous Human-AI Collaboration (2025.acl-long)
Copied to clipboard
Shao Zhang, Xihuai Wang, Wenhao Zhang, Chaoran Li, Junru Song, Tingyu Li, Lin Qiu, Xuezhi Cao, Xunliang Cai, Wen Yao, Weinan Zhang, Xinbing Wang, Ying Wen
| Challenge: | Large language models (LLMs) excel in turn-by-turn human-AI collaboration but struggle with simultaneous tasks requiring real-time interaction. |
| Approach: | They propose a language agent framework that integrates *System 1* and *System 2* for efficient real-time simultaneous human-AI collaboration. |
| Outcome: | The proposed framework improves on existing LLM-based agents and human collaborators by integrating Theory of Mind and asynchronous reflection to infer human intentions and perform reasoning-based autonomous decisions. |
AMO-Bench: Large Language Models Still Struggle in High School Math Competitions (2026.findings-acl)
Copied to clipboard
Junlin Liu, Shengnan An, Shuang Zhou, Dan Ma, Yehao Lin, Xinxuan Lv, Xuanlin Wang, Xiaoyu Li, Ziwen Wang, Xuezhi Cao, Xunliang Cai
| Challenge: | Existing benchmarks for mathematical reasoning are becoming less effective due to performance saturation. |
| Approach: | They propose to use a mathematical reasoning benchmark with Olympiad difficulty to evaluate top-tier LLMs. |
| Outcome: | The proposed benchmarks are cross-validated by experts to meet IMO difficulty standards and entirely original problems to prevent performance leakages from data memorization. |
Instance-level Randomization: Toward More Stable LLM Evaluations (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Evaluations of large language models suffer from instability, where small changes of random factors can lead to drastic fluctuations of scores and even model rankings. |
| Approach: | They propose an instance-level randomization method to reduce variance and improve fairness in evaluations by randomizing all factors that affect evaluation scores for every single instance. |
| Outcome: | The proposed method reduces variance and improves fairness in model comparisons by using instance-level randomization. |
Attribution-Based Analysis and Optimization of Modular Agentic Workflows (2026.findings-acl)
Copied to clipboard
Yingxuan Yang, Bo Huang, Siyuan Qi, Chao Feng, Haoyi Hu, Yuxuan Zhu, Jinbo Hu, Haoran Zhao, Ziyi He, Xiao Liu, ZongYu Wang, Muning Wen, Lin Qiu, Xuezhi Cao, Xunliang Cai, Yong Yu, Weinan Zhang
| Challenge: | Large Language Models (LLMs) have driven the rise of agentic workflows . yet, how can we attribute performance gains to individual upgrades and their interactions? |
| Approach: | They propose a game-theoretic framework that models component upgrades as players and evaluates component coalitions to compute Shapley values. |
| Outcome: | The proposed framework provides interaction-aware attribution and recommendation for model allocation under a fixed workflow structure. |
Why Not Act on What You Know? Unleashing Safety Potential of LLMs via Self-Aware Guard Enhancement (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated impressive capabilities across various tasks but are vulnerable to meticulously crafted jailbreak attacks. |
| Approach: | They propose a training-free defense strategy to align LLMs’ strong safety discrimination performance with their relatively weaker safety generation ability. |
| Outcome: | The proposed strategy achieves an average 99% success rate against numerous complex and covert jailbreak methods while maintaining helpfulness on general benchmarks. |
Meeseeks: A Feedback-Driven, Iterative Self-Correction Benchmark evaluating LLMs’ Instruction Following Capability (2026.findings-acl)
Copied to clipboard
Jiaming Wang, Yunke Zhao, Peng Ding, Jun Kuang, Yibin Shen, Zhe Tang, Yilin Jin, ZongYu Wang, Xiaoyu Li, Xuezhi Cao
| Challenge: | Existing models lack the ability to adhere to instructions, resulting in suboptimal performance. |
| Approach: | They propose an automated iterative instruction-following benchmark with integrated feedback mechanism. |
| Outcome: | The proposed benchmark identifies erroneous components in model responses and provides feedback accurately. |
SOP-Maze: Evaluating Large Language Models on Complicated Business Standard Operating Procedures (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are widely deployed as domain-specific agents, but evaluation of their capabilities in such contexts has not been fully explored. |
| Approach: | They propose a benchmark to evaluate LLMs' ability to follow instructions and make decisions in real-world scenarios. |
| Outcome: | The proposed benchmark is constructed from real-world business data and adapted into 23 complex SOP scenarios. |
A Wolf in Sheep’s Clothing: Generalized Nested Jailbreak Prompts can Fool Large Language Models Easily (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for generating 'jailbreaks' suffer from manual design or require optimization on other white-box models, which compromises either generalization or efficiency. |
| Approach: | They propose a framework that leverages LLMs to generate effective jailbreak prompts and a generalized framework that can be used to generate prompts. |
| Outcome: | The proposed framework improves the attack success rate while reducing the time cost compared to baselines. |
MUSE: MCTS-Driven Red Teaming Framework for Enhanced Multi-Turn Dialogue Safety in Large Language Models (2025.emnlp-main)
Copied to clipboard
Siyu Yan, Long Zeng, Xuecheng Wu, Chengcheng Han, Kongcheng Zhang, Chong Peng, Xuezhi Cao, Xunliang Cai, Chenjuan Guo
| Challenge: | Existing defenses target single-turn attacks, but real-world usage involves multi-turn dialogues, exposing models to attacks that exploit conversational context to bypass safety measures. |
| Approach: | They propose a framework that tackles multi-turn jailbreaks from both attack and defense angles. |
| Outcome: | Experiments on large language models show that MUSE effectively mitigates multi-turn jailbreaks. |
CoreCodeBench: Decoupling Code Intelligence via Fine-Grained Repository-Level Tasks (2026.acl-long)
Copied to clipboard
Lingyue Fu, Hao Guan, Bolun Zhang, Haowei Yuan, Yaoming Zhu, Lin Qiu, ZongYu Wang, Xuezhi Cao, Xunliang Cai, Weiwen Liu, Weinan Zhang, Yong Yu
| Challenge: | Existing large language models for software engineering rely on coarse-grained pass rates obscuring specific cognitive bottlenecks. |
| Approach: | They propose a repository-level benchmark that dissects coding capabilities through atomized tasks. |
| Outcome: | The proposed framework achieves a 78.55% validity yield, surpassing the 31.7% retention rate of SWE-bench-Verified. |
Conjoin after Decompose: Improving Few-Shot Performance of Named Entity Recognition (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing prompt-based NER models fail to detect entity boundaries, causing performance degradation. |
| Approach: | They propose a model which consists of a BART encoder and a parabiotic decoder and propose ' boundary expansion strategy' to enhance the model's capability in entity type classification. |
| Outcome: | The proposed model can achieve significant performance gains over state-of-the-art models. |
Transferable and Efficient: Unifying Dynamic Multi-Domain Product Categorization (2023.acl-industry)
Copied to clipboard
| Challenge: | e-commerce platforms are encountering increasingly complex product categorization scenarios . multiple business domains correspond to different category taxonomies, with different depths and distinct literal expressions of category names. |
| Approach: | They propose a taxonomy-agnostic framework that calculates semantic relatedness between product titles and category names in the vector space. |
| Outcome: | The proposed framework outperforms strong baselineson three dynamic multi-domain product categorization tasks. |
Table Fact Verification with Structure-Aware Transformer (2020.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained models cannot be used to encode semi-structured data because of their nature. |
| Approach: | They propose a Structure-Aware Transformer which injects table structural information into mask . method could combine symbolic and linguistic reasoning, they propose . |
| Outcome: | The proposed method outperforms baseline on a large scale table verification dataset. |