I’m a machine learning researcher working towards artificial agents that can reason like humans. I am interested in developing neural networks capable of system-2 or cognitive processing in order to generalize better than current models can.
My research is focused on what I see as the most direct paths to system-2 deep learning: reinforcement learning for discrete search, compositional representation learning, and emergent abilities of large language models.
Right now, I work at NVIDIA on aligning large language models with human preferences. I’ve also used language models for automated theorem proving and designed arithmetic circuits with reinforcement learning.
In the past, I’ve worked in Anima Anandkumar’s lab at Caltech studying the differences between how transformers and tree-structured neural networks learn representations. Before that, I was at Descartes Labs building better unsupervised learning algorithms for satellite imagery and finding trees. I recently graduated from Caltech, where I taught a class on deep learning.
For more on my background, you can check out my resume.