Biography

I am a CS PhD candidate at the Center for AI Safety at Stanford University, under the guidance of Prof. Clark Barrett. I am interested in training AIs to be human-like both in terms of the concepts they contain, as well as the values they hold. I think the key to this lies in training the language model to produce its own memory during training and to influence its future training data. In this vein, I have been optimizing pre-trained models to produce Chain-of-Thought reasoning which is informative to itself and to humans. In the future, I am interested in using a similar blend of unsupervised learning and reinforcement learning towards active learning, in the hopes that this sets the preconditions to raise the AI like a child, and have that result in something child-like.

During my PhD, my focus has evolved from formal verification and programming languages to AI alignment. Prior to this, I majored in computer science and electrical engineering at MIT, contributing to AI and robotics research. After MIT, I explored interactive theorem proving at CMU with Simon Dedeo, publishing research on abduction in mathematics in the Cognition journal.

My primary character trait is curiosity, and I really love math.