Jingxu Xie

I am a quantitative researcher at Voleon. I completed my PhD at UC Berkeley, where I worked with Prof. Feng Wang on experimental condensed-matter physics.

My research spans machine learning and automated experimental systems, with recent work on AI agents, evaluation, and scaling laws. I am currently collaborating with Prof. Dawn Song on Agents’ Last Exam as a core contributor.

I am also exploring robot learning and sim-to-real transfer through hands-on manipulation experiments. I am interested in how learning systems generalize beyond their training data and interact with the physical world.

Jingxu Xie

Selected work

Machine learning & robotics

ScienceIDE scientific-domain coverage and agent evaluation results on ScienceIDE-Hard.

Selected workshop papers

Cross-document linkage attacks and the LinkGuard corpus-aware defense.

COLM Re-Data workshop 2026

Pseudonymized but Linkable: Cross-Document Re-identification Risks in LLM-Ready Sensitive Text Corpora

Jingxu Xie

Studying how repeated details can link identities across pseudonymized documents, and how corpus-aware generalization can reduce that risk.

Clarify-to-Act framework comparing the expected utility of asking a question and taking an action.

COLM LSEI workshop 2026

Ask, Don't Guess: Learning When to Clarify in Situated Instruction Following

Jingxu Xie

Learning when a clarification question is worth asking by balancing useful information against interruption and error costs.

Experimental Quantum Physics

* Equal contribution