Rice (Xilin) Wang

I’m a second-year computer science PhD student at Northeastern University, advised by Professor Byron C. Wallace. I completed my undergraduate studies at Brown University, where I received an Sc.B. in Applied Mathematics–Computer Science.

My research focuses on the interpretability of large language models (LLMs), with a particular focus on clinical and biomedical applications. The question I would like to study during my PhD is: how can we use mechanistic interpretability in realistic, high-stakes settings in a way that is actionable?

Recently, I have been working on model organisms, models trained with deliberately installed behaviors, and using them as testbeds for a wide range of interpretability techniques.

news

Oct 07, 2026	New preprint released to arXiv: How to train your model organism.
Sep 03, 2025	Started my PhD in Computer Science at Khoury College, Northeastern University, advised by Byron C. Wallace.
May 25, 2025	Graduated from Brown University with an Sc.B. in Applied Mathematics–Computer Science and an honors thesis on circuit discovery in biomedical LLMs.

selected publications

arXiv

How to train your model organism

Xilin Wang, David Bau, and Byron C. Wallace

arXiv preprint arXiv:2610.10203, Oct 2026

Abs arXiv

Model organisms of alignment-relevant behaviors (e.g., backdoors, sycophancy, spurious correlations) have emerged as a key tool for evaluating whitebox interpretability techniques. We argue that the prevailing practice of training model organisms to a single objective of installing the target behavior is insufficient and propose validating model organisms with respect to three objectives with associated metrics: target-behavior installation, general-capability preservation (i.e., parametric knowledge, chat quality), and output naturalness (i.e., CoT and activations). We re-visit two publicly released organism suites using this validation framework and show that (1) chat quality and CoT naturalness degrade substantially across training recipes, and (2) validation metrics predict how well interpretability methods recover the installed behavior, e.g., a logit lens readout covaries with an organism’s general capabilities. We introduce a multi-objective training approach based on model merging to train more realistic model organisms. Finally, on a new suite of model organisms targeting demographic biases in clinical reasoning, we compare training recipes and find that DPO training stays closer to the base model than supervised finetuning, and the proposed model optimization approach better preserves capabilities and naturalness. Auditing this suite with an investigator agent, we again observe validation metrics tracking bias recovery. In sum, training methods shape the interpretability conclusions an organism supports, and we argue that one should consider multiple objectives to draw generalizable conclusions about interpretability methods using (realistic) model organisms.