Welcome

Hello, traveler from afar, who has stumbled upon this humble corner of the internet. Whether your arrival is by chance or by intention, you’re warmly welcomed. May you find exactly what you’re seeking—or perhaps uncover something delightful and unexpected.

I’m Xiang Fu

Broadly, my research interests center on a loop of model behavior, evaluation, and alignment. I treat language models as systems whose abilities emerge only through interaction with users and environments, and I aim to characterize how they learn, generalize, and respond to changing context. Across learning, representation, interaction, and verification, I design predictive measures such as distributional fidelity, reasoning transparency, and coordination efficiency that reveal where models are reliable, where they fail, and how interventions in training, decoding, or safety policies propagate through the system. In doing so I aim to build a taxonomy of model behaviors, analogous in spirit to a periodic table: not a complete explanation of intelligence, but a stable set of components and interactions on which richer theories and safer applications can be built.

Should you choose to explore further, you might start with some projects

There’s also my CV