About Me

I’m an Assistant Professor at Princeton University with appointments in the Department of Computer Science and School of Public and International Affairs. I will also be affiliated with the Princeton Center for Information Technology Policy (CITP) and the Center for Statistics and Machine Learning (CSML).

My research focuses on aligning machine learning, law, and policy for responsible real-world deployments. This includes work on AI safety, methods to improve reasoning capabilities in foundation models, interdisciplinary methods in law and AI, as well as core work on legal doctrine and policy (particularly around AI governance). My group also engages with external organizations to ground our research in real-world machine learning and public policy challenges. Some of my research has received coverage by TechCrunch, Science, New York Times, The Wall Street Journal, Bloomberg, and more.

Previously, as a joint degree candidate, I received a JD from Stanford Law School where I was advised by Dan Ho and Mariano-Florentino Cuéllar, as well as worked with many other wonderful colleagues. There, I co-led the Domestic Violence Pro Bono Project, worked on client representation with the Three Strikes Project, and contributed to the Stanford Native Law Pro Bono Project. I also received my PhD in Computer Science (AI) at Stanford University where I am lucky enough to be advised by Dan Jurafsky. Before that, I was lucky enough to be advised by David Meger and Joelle Pineau for my M.Sc. at McGill University and the Montréal Institute for Learning Algorithms. I also spent time as a Software Engineer and Applied Scientist at Amazon AWS/Alexa, worked with Justice Cuéllar at the California Supreme Court, was a part-time researcher with the Internal Revenue Service Research, Applied Analytics and Statistics Division, and more.

Latest News

• “A Safe Harbor for AI Evaluation and Red Teaming” is online. We advocate for legal and technical protections ensuring access to AI models for safety assessments. You can also sign our open letter.
• Our paper “Rethinking Machine Learning Benchmarks in the Context of Professional Codes of Conduct” is online and was presented at the Symposium on Computer Science and Law.
• Our paper “On the Societal Impact of Open Foundation Models” is online. We design a risk assessment framework for analyzing the marginal risk of open foundation models relative to pre-existing technologies.
• Check out our new work, “Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications.” We find that pruning ~2-3% of neurons is all it takes to remove safety guardrails in a 7B parameter aligned model. This may help explain some of our recent findings that “Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!”. In that work showed that jailbreaking gpt-3.5-turbo via the fine-tuning API takes only 10 examples and costs $0.20. And even typical benign fine-tuning datasets like Alpaca can degrade safety to some extent. This lays the land for a big part of my agenda: ensuring safety even after fine-tuning or continuous learning, particularly with open access models. If you’re interested in collaborating on this reach out! You might also be interested in Self-Destructing Models: Increasing the Costs of Harmful Dual Uses in Foundation Models which we presented at AIES 2023!
• We have two new papers out in the Journal of Free Speech Law: Where’s the liability in harmful AI speech? (with Tatsunori Hashimoto and Mark Lemley) and Freedom of Speech and AI Output (with Eugene Volokh and Mark Lemley).
• Our paper Foundation Models and Fair Use is online and forthcoming at JMLR!
• Our piece “Algorithmic Rulemaking v. Algorithmic Guidance” will be published in the Harvard Journal of Law and Technology (2024). Draft available on request and will be online soon!