About

Hi, I'm Dylan! I'm a final year PhD student in the Machine Learning Department at CMU, where I am fortunate to be advised by Zico Kolter. My work is supported by a NSF Graduate Research Fellowship. I'm also currently working with Gray Swan AI.

My research focuses on making language models behave in safe, controllable, and predictable ways. I'm currently working on curating better/safer training data and on monitoring for harmful behavior. In the past, I worked on generalization and on training with weak supervision or interpretable forms of domain knowledge.

Previously, I studied math and computer science at Brown University, where I was advised by Stephen Bach. I've also worked as as a student researcher at Google Research, a research intern at Amazon AWS, the Bosch Center for AI, and at NASA JPL. If you are interested in my work, feel free to get in touch. I am always looking to chat about research and form new collaborations!

News

• [May 2025]: I gave a talk at UMD on Safety Pretraining, where we create natively safe language models during pretraining!
• [Feb 2025]: Excited to share my Google internship work on analyzing similarity metrics in language model pretraining data curation!
• [Jan 2025]: Just released new work on monitoring the performance and behaviors of LLMs in a black-box setting!
• [May 2024]: Spending the summer at Google Research in NYC, where I'll be working on data curation for LLM pretraining.
• [Apr 2024]: I'll be traveling to AISTATS to present "Auditing Fairness under Unobserved Confounding" and to ICLR to present work on providing generalization bounds for prompt engineering VLMs.
• [Jul 2023]: I'm giving a talk on learning data-driven priors for BNNs that incorporates interpretable domain knowledge at the KLR workshop @ ICML 2023!