I am a computer science Ph.D. candidate at Princeton University's Center for Information Technology Policy and a Porter Ogden
Jacobus Fellow. Previously, I was a senior fellow at Mozilla
and a Laurance
S. Rockefeller Graduate Prize Fellow at Princeton University.
I co-authored AI Snake Oil with Arvind Narayanan, named one of Nature's 10 best books of 2024. I
am a recipient of the Privacy Papers
for Policymakers Award, a best paper award at ACM FAccT, an impact recognition award at ACM
CSCW, and was included in TIME's inaugural list of the 100 most
influential people in AI.
Developing the science of AI evaluation
I study how to evaluate AI agents and frontier systems by building large-scale systems to conduct evaluations at scale.
- Reliability: Towards a Science of AI Agent Reliability (ICML 2026) · Log analysis for credible agent evaluation (preprint 2026)
- HAL: Holistic Agent Leaderboard (ICLR 2026) · Project page
- CRUX: Open-world evaluations for frontier AI capabilities (preprint 2026) · Benchmark
AI as Normal Technology
With Arvind Narayanan, I write the AI as Normal Technology newsletter, and I am working on my next book on the topic with him.
- Framework: AI as Normal Technology · Guide to AI as normal technology
- AGI: AGI is not a milestone
- Government response: Government responses to AI's transformative impacts
Evidence-based AI policy
I work on AI policy grounded in evidence about model openness, evaluation access, transparency, and accountability.
-
Open foundation models:
Science (2024): governance considerations for open foundation models.
ICML oral (2024): societal impacts of open foundation models. - FMTI: A longitudinal effort to measure transparency across leading foundation model developers: 2025 (TMLR 2026) · 2024 (TMLR 2025) · 2023 (TMLR 2025)
-
Testimonies and input:
ASAP RFI on AI-based science.
New Jersey Assembly: reducing harm from deepfakes.
Congressional Forum: accountability in predictive AI.
AI-based science
I study how AI-based science fails to reproduce and what practices, benchmarks, and reporting standards can make computational research more credible.
- Leakage: Leakage and the reproducibility crisis in AI-based science (Patterns 2023).
- REFORMS: REFORMS: Consensus-based Recommendations for Machine-learning-based Science (Science Advances 2024).
- CORE-Bench: A benchmark for evaluating AI's ability to reproduce computational research (CORE-Bench, TMLR 2025).
- Essays: Could AI slow science? (AI as Normal Technology 2025) · Scientists should use AI as a tool, not an oracle (AI as Normal Technology 2024).