I am a heliophysicist at UCLA. I study the Alfvénic structures that fill the solar wind, from the magnetic switchbacks Parker Solar Probe flies through near the Sun to the Alfvénon: a three-dimensional solitary Alfvén wave, which I construct as a nonlinear solution of ideal MHD and simulate on GPUs.

I also build infrastructure for AI-driven science. LingTai is an open-source runtime for lifelong research agents. ScienceIDE turns real scientific codebases into environments where agents are evaluated and trained. Since October 2026 I have led the astronomy track of the Agent Last Exam, which maps the open problems of astronomy into tasks for AI agents.

I received my Ph.D. in Geophysics and Space Physics from UCLA in 2024, and my dissertation was awarded the 2024 IAU PhD Prize of Division E (Sun and Heliosphere).

自述

黄泽森，字澍之，号玉兰居士，广州东山人也。蒙祖荫，生长于广州城。幼而好格物，耽读天文理数之书。束发，就学于执信；继而北上，入庐阳。及冠，渡海赴美，求学于洛城，徘徊至今。初探荧惑之天际，后究太阳之风息，辗转数载，终膺列邦司天会博士旌表。今感智械之威，慨然有志，遂辟灵台，育器灵焉。

In my own words

Huang Zesen, courtesy name Shuzhi, literary name the Magnolia Recluse, is from Dongshan in Guangzhou. Blessed by his forebears, he grew up in the city of Guangzhou. As a boy he loved to work out how things are, and lost himself in books of astronomy, physics and mathematics. In his teens he studied at Zhixin; then he went north, to Luyang. Grown, he crossed the ocean to America to study in Los Angeles, and has lingered there ever since. He first explored the sky of Mars, then the wind of the Sun, and after some winding years was honoured with the PhD Prize of the International Astronomical Union. Now, struck by the power of artificial intelligence, he has resolved to open LingTai and raise its agents there.

The 自述 (self-account) of lingtai.ai, in classical Chinese; English rendering on this site.

Current focus

Alfvénons

Solitary Alfvén waves in 3D: construction, propagation, and now head-on collisions on GPUs.

LingTai 灵台

An open-source runtime for lifelong digital scientists that remember, grow, and work in networks.

ScienceIDE

64 executable environments from 27 scientific codes, graded by the physics, for agent evaluation and RL.

Agent Last Exam: Astronomy

1,779 open problems of astronomy, sorted into tasks where AI agents can keep improving.

News