Zhaowei Zhang (张钊为)

Ph.D. Student
School of Intelligence Science and Technology
Peking University
Email: zwzhang [at] stu (dot) pku (dot) edu (dot) cn

[Google Scholar] [Github] [Twitter] [LinkedIn]

Research Interests
  • Recursive Self Improvment
  • AI Agent & Agentic Reinforcement Learning
  • Omni MLLM & World Models
  • Latent Space Reasoning

Zhaowei is pronounced as "Ju" (as in judge) + "ou" (as in out) + "Way"; Zhang, or Cheung in Hong Kong, is "Ju" (as in judge) + "on" | audio ([International Phonetic Alphabet, IPA]) here: [tʂɑuwei][tʂɑŋ].

I am currently a third-year Ph.D. candidate at Institute for AI, School of Intelligence Science and Technology, Peking University. Specifically, I am in the team of PAIR-Lab led by Prof. Yaodong Yang. The long-term goal of my research is to explore the edge of intelligence in the scaling process and how such intelligence can better serve human society. I have previously focused on problems in RL, online convex learning, gerative AI, AI alignment, and LLM reasoning. Currently, the question I care most about is how to build a strong and efficient agentic system that can make the most use of computational resources to achieve the greatest automation and intellgence. I welcome more friends to discuss these topics with me ☺️.

Selected Publications (* indicates equal contribution.)

    2026
  • GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning
    Zhaoxin Yu, Qi Shen, Hengli Li, Zhaowei Zhang, Song-Chun Zhu, Chi Zhang, Zilong Zheng
    Preprint
    [Paper]
  • PILA: Plug-and-Play Insertion for LLM-native Advertising
    Zhaowei Zhang, Yuhan Fu, Yihang Zhang, Xiaohan Liu, Ceyao Zhang, Xiaoyuan Zhang, Yipeng Kang, Tonghan Wang, Yaodong Yang
    Preprint
    [Paper]
  • VINE: Taming Generative Control Policies for Reinforcement Learning
    AgiBot Team
    Preprint
    [Paper]
  • Agora: Toward Autonomous Bug Detection in Production-Level Consensus Protocols with LLM Agents
    Xiang Liu *, Sa Song *, Zhaowei Zhang *, Huiying Lan, Jason Zeng, Ming Wu, Michael Heinrich, Yong Sun, Ceyao Zhang
    ICML 2026
    [Paper]
  • From human videos to robot manipulation: A survey on scalable vision-language-action learning with human-centric data
    MSRA Embodied AI Team
    IJCAI 2026
    [Paper]
  • Does LLM Alignment Really Need Diversity? An Empirical Study of Adapting RLVR Methods for Moral Reasoning
    Zhaowei Zhang, Xiaohan Liu, Xuekai Zhu, Junchao Huang, Ceyao Zhang, Zhiyuan Feng, Yaodong Yang, Xiaoyuan Yi, Xing Xie
    ICML 2026
    [Paper]
    • 2025
  • Make an Offer They Can't Refuse: Grounding Bayesian Persuasion in Real-World Dialogues without Pre-Commitment
    Buwei He, Yang Liu, Zhaowei Zhang, Zixia Jia, Huijia Wu, Zhaofeng He, Zilong Zheng, Yipeng Kang
    Preprint
    [Paper]
  • Aegis: Automated Error Generation and Identification for Multi-Agent Systems
    Fanqi Kong, Ruijie Zhang, Huaxiao Yin, Guibin Zhang, Xiaofei Zhang, Ziang Chen, Zhaowei Zhang, Xiaoyuan Zhang, Song-Chun Zhu, Xue Feng
    ICLR 2026
    [Paper] [Website] [Dataset]
  • PoliCon: Evaluating LLMs on Achieving Diverse Political Consensus Objectives
    Zhaowei Zhang, Xiaobo Wang, Minghua Yi, Mengmeng Wang, Fengshuo Bai, Zilong Zheng, Yipeng Kang, Yaodong Yang
    ICLR 2026
    [Paper] [Website]
  • Evaluating Generalization Capabilities of LLM-Based Agents in Mixed-Motive Scenarios Using Concordia
    Cooperate with the DeepMind Concordia Team
    NeurIPS DB Track 2025
    [Paper]
  • Amulet: ReAlignment During Test Time for Personalized Preference Adaptation of LLMs
    Zhaowei Zhang, Fengshuo Bai, Qizhi Chen, Chengdong Ma, Mingzhi Wang, Haoran Sun, Zilong Zheng, Yaodong Yang
    ICLR 2025
    [Paper] [Website] [Code]
    • 2024
  • ValueDCG: Measuring Comprehensive Human Value Understanding Ability of Language Models
    Zhaowei Zhang, Fengshuo Bai, Jun Gao, Yaodong Yang
    NeurIPS 2025 Workshop on Regulatable ML
    [Paper] [Blog] [Chinese Blog]
  • Foundational Challenges in Assuring Alignment and Safety of Large Language Models
    Core author
    TMLR
    [Paper] [Website]
  • Roadmap on Incentive Compatibility for AI Alignment and Governance in Sociotechnical Systems
    Zhaowei Zhang, Fengshuo Bai, Mingzhi Wang, Haoyang Ye, Chengdong Ma, Yaodong Yang
    AGI 2025 (Oral)
    [Paper] [Chinese Blog]
    • 2023
  • AI Alignment: A Comprehensive Survey
    Core author
    ACM Computing Surveys
    [Paper] [Website]
  • ProAgent: Building Proactive Cooperative AI with Large Language Models
    Ceyao Zhang, Kaijie Yang, Siyi Hu, Zihao Wang, Guanghe Li, Yihang Sun, Cheng Zhang, Zhaowei Zhang, Anji Liu, Song-Chun Zhu, Xiaojun Chang, Junge Zhang, Feng Yin, Yitao Liang, Yaodong Yang
    AAAI 2024 (Oral)
    [Paper]
  • Heterogeneous Value Alignment Evaluation for Large Language Models
    Zhaowei Zhang, Nian Liu, Siyuan Qi, Ceyao Zhang, Ziqi Rong, Shuguang Cui, Song-Chun Zhu, Yaodong Yang
    AGI 2025 & AAAI 2024 Workshop: Public Sector LLMs (Oral)
    [Paper]
  • STAS: Spatial-Temporal Return Decomposition for Solving Sparse Rewards Problems in Multi-agent Reinforcement Learning
    Sirui Chen *, Zhaowei Zhang *, Yali Du, Yaodong Yang
    AAAI 2024
    [Paper] [Code]

Projects

PaperMaster
An AI native paper reading and autoresearch system on MacOS. [Project Page]
Project owner.
May 2026
LightX2V
A 2.7k+ star video inference infra framework. [Project Page]
Contributor for building world action model (WAM) acceleration.
March 2026

Blogs & Tutorials

  • AI for Helping People to Find Consensus
    Abstract: This tutorial explains how to use LLMs as scalable "AI mediators" that infer preferences from large amounts of free-text opinions and generate consensus statements representing different groups, helping address information bandwidth limits, anchoring effects, and inefficiencies in real-world deliberation. It covers group-level generative social choice and stakeholder-level consensus generation and evaluation, and discusses challenges such as bias, dynamic negotiation, and interaction costs.
    [Tutorial Slides]
  • The Three-Layer Paradigm for Implementing Sociotechnical AI Alignment: A Top-Down-Top Outlook
    Abstract: This blog clarifies what socio-technical systems (STS) mean in the context of AI alignment, resolving inconsistent definitions across scales. It presents a computable, multi-scale view of STS alignment problems and outlines possible research directions.
    [English Version] [Chinese Version]

Internship

LongCat Foundation Model Team, Meituan
Beidou Project Intern, working on RSI and Agentic Post Training.
July 2026
Kling AI, Kuaishou Tech.
Research Intern, working on video generation with Xintao Wang
February 2026 - July 2026
Microsoft Research (MSRA)
Research Intern, working on LLM reasoning diversity with Xing Xie & Xiaoyuan Yi
September 2025 - January 2026

Selected Awards

  • Century Frontier Young Scholar Award, 2026.
  • Huawei Spark Award, 2025. (the only student recipient) [News]

Services

  • Reviewer for AI conferences represented by ICLR, NeurIPS, and ICML.
  • Program Committee Member for AAAI 2026 AIA Track.
  • Program Committee Member for AAAI 2026.
  • Program Committee Member for DAI 2024.