Manqing Liu

My current research develops methods to detect reasoning pathologies in large language models, including post-hoc rationalization, encoded reasoning, and internalized reasoning. I construct model organisms exhibiting each pathology and build end-to-end evaluation pipelines for training-time monitoring and scalable oversight. I’ve also worked on causal machine learning research in collaboration with Dr. Andrew Beam and Dr. James Robins. I am seeking Research Scientist or Research Engineer roles in model evaluation, post-training, alignment, and safety.

In my free time, I enjoy reading philosophy and drawing connections between classical ideas on concepts, reasoning, knowledge, and understanding—from Kant, Wittgenstein, and Schopenhauer—to modern LLM research.