THINK, ACT, EVOLVE
Haoyang Zou 邹皓阳
Ph.D. student at Shanghai Jiao Tong University.
I study how strong reasoning becomes reliable agent behavior, and I like building the full loop from training to real-world systems.
About Me
I work on models that can think, act, and hold up once they leave controlled environments and enter real work.
I am a first-year Ph.D. student at Shanghai Jiao Tong University, advised by Prof. Pengfei Liu, and a research intern at ByteDance Seed, where I work closely with Yujia Qin. My research centers on reasoning, reinforcement learning, and agents, with a focus on training foundation models for capable and reliable agent behavior. Before that, I received my B.S. in Artificial Intelligence from Fudan University.
Research Interests
- Reasoning and Reinforcement Learning for Agents Studying how agents can plan, adapt, and remain reliable in complex environments.
- Coding Agents in Practice Designing coding agents that optimize for real-world usefulness rather than benchmark performance.
- Self-Improving Agent Systems Exploring systems whose capability grows through use and interaction.
News
- 2026.02 Released Seed 2.0 as a core contributor.
- 2025.12 Released Seed 1.8 as a core contributor.
- 2025.09 Released the UI-TARS-2 technical report.
- 2025.09 Started my Ph.D. at Shanghai Jiao Tong University.
- 2025.06 Joined ByteDance Seed as a research intern.
- 2025.06 Graduated from Fudan University.
Selected Work
Seed 2.0
Product Launch, Feb 2026
As a core contributor, I am responsible for the coding agent ability of Seed 2.0, a multimodal foundation model series built for strong agentic performance.
Seed 1.8
Product Launch, Dec 2025
As a core contributor, I am responsible for the coding agent ability of Seed 1.8, a generalized agentic model unifying search, code, and GUI abilities in one system.
UI-TARS-2: Advancing GUI Agent with Multi-Turn Reinforcement Learning
Tech Report, Sep 2025
A GUI-centered agent trained with multi-turn RL and a data flywheel, reaching SOTA on OSWorld, Mind2Web, and AndroidWorld, and supporting the Doubao Mobile Assistant.
ToRL: Scaling Tool-Integrated RL
Preprint, Mar 2025
Models
A tool-integrated RL framework that lets base models acquire tool use through training rather than prompting alone.
LIMR: Less is More for RL Scaling
Preprint, Feb 2025
A data selection method showing that better-chosen trajectories can match larger RL runs at a fraction of the scale.
O1 Replication Journey -- Part 2: Surpassing O1-preview through Simple Distillation, Big Progress or Bitter Lesson?
Preprint, Nov 2024
A distillation study showing that simple pipelines can push open models past o1-preview on targeted reasoning evaluations.
O1 Replication Journey: A Strategic Progress Report -- Part 1
Preprint, Oct 2024
An early progress report documenting what did and did not work while replicating o1-style reasoning behavior.
Generative AI for Math: Abel
Project 2024
A mathematical reasoning model that reached state-of-the-art results on GSM8K and MATH at the time.
Generative AI Act II: Test Time Scaling Drives Cognition Engineering
Preprint, Apr 2025
A survey framing test-time scaling as an emerging engineering discipline for reasoning systems.
O1 Replication Journey -- Part 3: Inference-time Scaling for Medical Reasoning
Preprint, Jan 2025
An inference-time scaling study for medical reasoning, focused on where extra computation actually improves clinical problem solving.
PC Agent: While You Sleep, AI Works -- A Cognitive Journey into Digital World
Preprint, Dec 2024
A GUI agent project inspired by human cognitive workflows for completing computer tasks end to end.
OlympicArena: Benchmarking Multi-discipline Cognitive Reasoning for Superintelligent AI
NeurIPS 2024 Datasets & Benchmarks
A multimodal benchmark designed to test cross-discipline reasoning at olympiad difficulty.
Experiences
-
2025.06 - Present
Research Intern, ByteDance SeedFocused on coding and GUI agents, working closely with Yujia Qin, and contributing to UI-TARS-2, Seed 1.8, and Seed 2.0 as a core contributor.
-
2025.09 - Present
Ph.D. Student, Shanghai Jiao Tong UniversityComputer Science, advised by Prof. Pengfei Liu, with a focus on reasoning, reinforcement learning, and agent systems.
-
2021.09 - 2025.06
B.S., Fudan UniversityArtificial Intelligence