Math & CS @ HCMUS · Learning Mechanistic Interpretability
I'm a second-year student in Mathematics & Computer Science at Ho Chi Minh City University of Science (HCMUS), Vietnam National University.
My main interest is mechanistic interpretability: figuring out what is actually happening inside large language models. Right now I'm working through the ARENA curriculum, running small experiments with TransformerLens, and building my first research repo on safety probes. This blog is my lab notebook: I write down what I try, what breaks, and what I learn, including the parts I don't understand yet.
My near-term goal is to join MATS and do real AI safety research. Longer term, I want to become a Research Engineer at a frontier AI lab, and one day bring that experience back home to help Vietnam build AI at the core-technology level.
Ho Chi Minh City University of Science (HCMUS)
Year 1 GPA: 9.67/10 | Focus: Interpretability & AI Safety
Reverse-engineering the circuits and features inside transformers, e.g. how a model decides between "a" and "an".
Probing and evaluating model behavior, and checking whether safety signals still hold when the data distribution shifts.
Linear algebra, calculus and probability: the tools I lean on to read papers and reason about what models compute.
I'm always happy to talk about interpretability, AI safety, or research ideas, and I'd love to hear from anyone further along this path. Feel free to reach out:
Follow along as I go from "what is a residual stream?" to doing real interpretability research, one lab notebook entry at a time.
Read the Lab Notebook