PeopleofSuperintelligence

Researcher

Zijun Liu

Zijun Liu is an AI researcher at Tsinghua University who works on large language model agents and reward models and is among the authors of the DeepSeek-V3 and DeepSeek-R1 papers. As a visiting student at Stanford University in 2023, he first-authored the DyLAN framework, in which a team of agents collaborates through a dynamic communication structure. After starting a PhD in Yang Liu's group at Tsinghua in 2024, he first-authored AIGS in November 2024 and was among the authors of the DeepSeek-V3 Technical Report in December 2024 and of DeepSeek-R1, published in Nature in September 2025. In April 2025, his paper on inference-time scaling for generalist reward modeling, written with DeepSeek-AI and Tsinghua colleagues, introduced Self-Principled Critique Tuning and the DeepSeek-GRM models. He is a PhD student in Yang Liu's group at Tsinghua University, where he first-authored the ExtAgents framework in May 2025 and coauthored a test-time environment method for LLM web agents in August 2026.

Badges

Timeline

  1. 2023

    The DyLAN agent network

    In October 2023, he first-authors DyLAN, a Dynamic LLM-Powered Agent Network in which a team of agents is selected from candidates and collaborates through a dynamic communication structure. The work is done while he is a UGVR visiting student at Stanford University, with an affiliation in the Department of Computer Science and Technology at Tsinghua University, and the paper is published at COLM 2024 with Yanzhe Zhang, Peng Li, Yang Liu and Diyi Yang.

  2. 2024

    Self-distillation and agent alignment

    He coauthors two 2024 conference papers with Peng Li and Yang Liu. At ACL 2024, a self-distillation method called SDRRL improves the multilingual performance of LLMs by drawing on their capabilities in resource-rich languages; at ICML 2024, a position paper proposes Unified Alignment for Agents, which aligns agents with human intentions, environmental dynamics and self-constraints such as monetary budgets.

  3. 2024

    PhD in Yang Liu's group

    In 2024, he starts a PhD in Yang Liu's group at Tsinghua University. In November 2024, he first-authors AIGS with Kaiming Liu, which studies AI-Generated Science, where agents independently and autonomously complete the entire research process and discover scientific laws, and presents the Baby-AIGS demonstration system.

  4. December 2024

    Author of DeepSeek-V3

    He is listed among the authors of the DeepSeek-V3 Technical Report, posted in December 2024. The report presents a Mixture-of-Experts language model with 671B total parameters, 37B of them activated for each token.

  5. January 2025–September 2025

    Author of DeepSeek-R1

    He is among the authors of DeepSeek-R1, posted to arXiv in January 2025 and published in Nature in September 2025. The paper shows that the reasoning abilities of LLMs can be incentivized through pure reinforcement learning, obviating the need for human-labeled reasoning trajectories.

  6. April 2025

    Inference-time scaling for reward models

    In April 2025, he is first author of "Inference-Time Scaling for Generalist Reward Modeling," listed with a DeepSeek-AI affiliation alongside Peiyi Wang, Runxin Xu, Shirong Ma, Chong Ruan, Peng Li, Yang Liu and Yu Wu. The paper proposes Self-Principled Critique Tuning, which trains generative reward models through online reinforcement learning to generate principles and critiques, resulting in the DeepSeek-GRM models.

  7. May 2025

    Multi-agent scaling with ExtAgents

    In May 2025, he first-authors ExtAgents with Zhennan Wan, a multi-agent collaboration framework for scaling the external knowledge input of LLMs beyond their context windows. The paper is accepted to ACL 2026.

  8. August 2026

    Test-time environments at Tsinghua

    In August 2026, he coauthors a paper with Junxuan Li and Ziyi Huang on Test-Time Environment Decomposition (TTED), a label-free method that lets LLM web agents adapt with experience during inference. He is a PhD student in Yang Liu's group at Tsinghua University.

Career

  1. NowTsinghua UniversityAI researcher
  2. From 2025DeepSeek
  3. 2023Stanford UniversityUGVR visiting student

Education

  1. From 2024Tsinghua UniversityPhD

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Zijun Liu. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. 16% of AI career moves in the 2020s went from a university to a companyIncludes Zijun Liu. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  3. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes Tsinghua University. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.
  4. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Tsinghua University. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Zijun Liu yet. Send us a free-licensed photo.

  1. A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration arxiv.org. Checked 22 October 2026.
  2. A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration (arXiv HTML) arxiv.org. Checked 22 October 2026.
  3. Enhancing Multilingual Capabilities of Large Language Models through Self-Distillation from Resource-Rich Languages - ACL Anthology aclanthology.org. Checked 22 October 2026.
  4. Position: Towards Unified Alignment Between Agents, Humans, and Environment proceedings.mlr.press. Checked 22 October 2026.
  5. Yang Liu's Homepage at Tsinghua University nlp.csai.tsinghua.edu.cn. Checked 22 October 2026.
  6. AIGS: Generating Science from AI-Powered Automated Falsification arxiv.org. Checked 22 October 2026.
  7. DeepSeek-V3 Technical Report arxiv.org. Checked 22 October 2026.
  8. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org. Checked 22 October 2026.
  9. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning | Nature nature.com. Checked 22 October 2026.
  10. Inference-Time Scaling for Generalist Reward Modeling arxiv.org. Checked 22 October 2026.
  11. Inference-Time Scaling for Generalist Reward Modeling (arXiv HTML) arxiv.org. Checked 22 October 2026.
  12. Scaling External Knowledge Input Beyond Context Windows of LLMs via Multi-Agent Collaboration arxiv.org. Checked 22 October 2026.
  13. Learning Simple Test-Time Environments for LLM Web Agents arxiv.org. Checked 22 October 2026.