Siyuan Zhuang

Today, I am Tech Lead of AI at SAIR Foundation. I work on the scientific foundations of artificial general intelligence and the virtuous cycle between AI and science: using scientific methods to better understand intelligence, then using better AI to accelerate discovery.

My broader goal is to understand what distinguishes humans from AI, what AI can teach us about ourselves and intelligence, and how humans and AI should relate in the future. My current work approaches these questions through three capabilities that I believe are still missing:

• Continual learning: acquiring knowledge and skills over time without forgetting what came before.
• Learning from experience: moving beyond fitting human-generated data toward creative adaptation through interaction with the world.
• Eliminating plausible hallucinations: understanding why models produce convincing falsehoods and making reliable truth-seeking a scientific objective.

I earned my PhD in Computer Science from UC Berkeley, where I was advised by Ion Stoica and Dawn Song. During my PhD, I was affiliated with the Sky Computing Lab and RISELab, and contributed to open-source community projects including Ray, SkyPilot, vLLM, and Chatbot Arena. These projects grew into widely used infrastructure for the AI community and industry.

Before Berkeley, I completed the Computer Science Honors Program at the University of Science and Technology of China. Across each chapter, I have returned to the same underlying question: what can the intelligent systems we build teach us about intelligence, ourselves, and the future relationship between humans and AI?