PeopleofSuperintelligence

Researcher

Zhiyuan Zeng

Also known as 曾致远

Zhiyuan Zeng is a PhD student at the University of Washington who works on how language models are evaluated, trained with reinforcement learning and built in the open. A gold medal at the 2019 China National Olympiad in Informatics led to Tsinghua University, where he earned a computer science degree in 2024 after a summer as a visiting student in Danqi Chen's group at Princeton in 2023. He coauthored LLMBar and Sheared LLaMA in 2023, began his PhD at Washington in September 2024 under Hannaneh Hajishirzi and Pang Wei Koh, and in 2025 introduced EvalTree, wrote RLVE as first author and contributed to the Olmo 3 technical report. He holds an Amazon AI PhD Fellowship for 2025 to 2027, and his RLVE paper appears at ICML 2026.

Timeline

  1. 2019–2024

    Computer science at Tsinghua

    A gold medal at the China National Olympiad in Informatics in July 2019, a top-50 result nationwide, brings him direct admission to Tsinghua University. He receives a Bachelor of Engineering from the Department of Computer Science and Technology there in June 2024.

  2. 2023

    LLMBar and Sheared LLaMA

    He spends the summer of 2023 as a visiting undergraduate in Danqi Chen's group at Princeton. In October 2023, he is first author of LLMBar, a benchmark of 419 manually curated output pairs that tests how well LLM evaluators judge instruction following; the paper appears at ICLR 2024. Sheared LLaMA, posted in October 2023 with Mengzhou Xia, Tianyu Gao and Danqi Chen, prunes LLaMA2-7B down to 1.3B and 2.7B parameters.

  3. 2024

    PhD at the Allen School

    In September 2024, he begins a PhD at the Paul G. Allen School of Computer Science & Engineering at the University of Washington. Hannaneh Hajishirzi and Pang Wei Koh are his advisors.

  4. 2024–2027

    Allen School and Amazon fellowships

    The Allen School's Computer Science & Engineering Research Fellowship supports him from September 2024 to June 2025. An Amazon AI PhD Fellowship runs from September 2025 to September 2027.

  5. March 2025

    EvalTree weakness profiles

    In March 2025, he posts EvalTree with Yizhong Wang, Hannaneh Hajishirzi and Pang Wei Koh, and the paper appears at COLM 2025. EvalTree builds a capability tree described in natural language from a benchmark and extracts the nodes where a language model performs poorly as a weakness profile. On the MATH and WildChat benchmarks, it identifies weaknesses more precisely and comprehensively than baseline methods.

  6. November 2025

    RLVE and ThetaEvolve

    In November 2025, he is first author of RLVE, which trains language models with reinforcement learning on adaptive verifiable environments. Its RLVE-Gym suite has 400 environments, and joint training lifts a 1.5B reasoning model by 3.37% on average across six reasoning benchmarks. He also coauthors ThetaEvolve, an open-source framework in which a small model such as DeepSeek-R1-0528-Qwen3-8B reaches new best-known bounds on open problems from AlphaEvolve.

  7. December 2025

    A contributor to Olmo 3

    In December 2025, he is among the authors of the Olmo 3 technical report, which introduces a family of fully-open language models at the 7B and 32B parameter scales. The release includes every stage, checkpoint, data point and dependency used to build the models.

  8. As of October 2026

    Allen School doctoral student

    He is a PhD student at the Paul G. Allen School of Computer Science & Engineering at the University of Washington, where he began in 2024. RLVE, on which he is first author, appears at ICML 2026.

Career

  1. 2023Princeton UniversityVisiting undergraduate

Education

  1. 2024Tsinghua UniversityBachelor of Engineering
  2. From 2024University of WashingtonPhD

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Princeton 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.
  2. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorateIncludes Princeton University. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorate. Across 2,116 people with both an undergraduate degree and a doctorate recorded, we trace which undergraduate schools feed which doctoral programs, how often students stay at the same school, and how paths differ by birth-country region.
  3. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes Princeton 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 median h-index is 48 at universities and 14 at frontier AI labsIncludes Princeton University. 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.

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 Zhiyuan Zeng yet. Send us a free-licensed photo.

  1. Zhiyuan Zeng zhiyuan-zeng.github.io. Checked 21 October 2026.
  2. Lab · Danqi Chen cs.princeton.edu. Checked 21 October 2026.
  3. [2310.07641] Evaluating Large Language Models at Evaluating Instruction Following arxiv.org. Checked 21 October 2026.
  4. [2310.06694] Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning arxiv.org. Checked 21 October 2026.
  5. [2503.08893] EvalTree: Profiling Language Model Weaknesses via Hierarchical Capability Trees arxiv.org. Checked 21 October 2026.
  6. [2511.07317] RLVE: Scaling Up Reinforcement Learning for Language Models with Adaptive Verifiable Environments arxiv.org. Checked 21 October 2026.
  7. [2511.23473] ThetaEvolve: Test-time Learning on Open Problems arxiv.org. Checked 21 October 2026.
  8. [2512.13961] Olmo 3 arxiv.org. Checked 21 October 2026.