张洋 ☕️
张洋 Zhang Yang

Undergraduate Student

Peking University, Intern @ UCSD Roselab

About Me

Aligning today’s highly-capable AI systems with social and intellectual progress. Also does funky stuff on the guitar.

Download CV
Interests
  • Value Alignment
  • Equitable AI
  • Domain Agents
Education

  • Research Intern

    University of California San Diego

  • Exchange Student

    National University of Singapore

  • BSc Intelligence Science and Technology

    Peking University

Thanks for Scrolling Down!

I am a senior undergraduate in the Zhi Class Honors Program at Peking University, currently conducting research internships at Prof. Rose Yu’s Lab @ UC San Diego. Previously, I was fortunate enough to be supervised by Prof. Yaodong Yang @ Peking University. Please see ’experience’ page for details.

My research interests lie in understanding, grounding and building large language models for impactful application. This would include fileds such as alignment, interpretability and agent systems. My ultimate goal is to build AI systems that boosts moral progress and equitable access, and AI agents that advance us in high-value domains such as science. Please see ‘writings’ page for what I’ve been tinkering.

Recent News
  • [2026.01.07] I’ve joined ByteDance as an intern, working with the amazing Lark (Feishu) Search/AI team.
  • I’m actively seeking for PhD opportunities starting in 2026 Fall. Feel free to reach out!
  • [2025.12.19] I was awarded the “Qiusuo” Scholarship by School of Electric Engineering and Computer Science, Peking University (10 recipients).
  • [2025.12.04] I was awarded the “Taotian” Scholarship jointly by Alibaba and School of Intelligence Science and Technology of Peking University (4 undergraduate recipients).
  • [2025.12.01] Welcome to my website!
Featured Publications
The Emergence of Abstract Thought in Large Language Models Beyond Any Language
ProgressGym: Alignment with a Millennium of Moral Progress
Blogs & Manuscripts

Things I’ve been tinkering.

Machine Learning Course Project

Complementing OpenEvolve with experience replay to find solution for a problem that’s on par with human-expert SOTA (which no code-generation agent has accomplished before).

Multi-Modal Learning Project

Inference-time scaling for multimodal reasoning, improving accuracy and efficiency under tight compute.

Capstone Project Proposal

Proposal outlining the background, goals, methodology, and expected impact of my capstone project.