I recently completed my PhD in the ECE Department at the University of Washington, advised by Prof. Maryam Fazel and Prof. Lillian Ratliff. During my PhD, I also collaborated closely with Prof. Jeff Bilmes and Prof. Kevin Jamieson. I am on the industry job market.
I study how different forms of user data shape LLM behavior, with a focus on user feedback in RLHF and adversarially constructed fine-tuning data. I investigate when training on these signals produces manipulation, deception, or other hidden misalignment. I combine empirical LLM research with reinforcement learning, game theory, and optimization to characterize these failures and develop mitigations. See the Publications tab for the full list.
Prior to joining UW, I completed my master’s degree in 2020 in the BLISS Lab in the EECS Department at UC Berkeley, where I was advised by Prof. Anant Sahai. I completed my undergraduate studies at UC Berkeley with a major in EECS and a minor in Theater, Dance & Performance Studies.
A. Narang*, A. Tajdini*, C. Zhang, J. Morgenstern. “Inference-Time Consensus for Mitigating Hidden Behaviors from LLM Fine-Tuning.” Preprint, 2026; accepted to the AdvML-Frontiers × CoTMA Workshop at COLM 2026 (paper) (code)
A. Narang, S. Dean, L. Ratliff, M. Fazel. “Dynamics of Learning under User Choice: Overspecialization and Peer-Model Probing.” UAI 2026 (arXiv)
M. Williams*, M. Carroll*, A. Narang, C. Weisser, B. Murphy, A. Dragan. “On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback.” ICLR 2025 (arXiv) (Coverage in the New York Times and Washington Post)
A. Narang, A. Wagenmaker, L. Ratliff, K. Jamieson. “Sample Complexity Reduction via Policy Difference Estimation in Tabular Reinforcement Learning.” NeurIPS 2024 (Spotlight) (arXiv)
A. Narang, E. Faulkner, D. Drusvyatskiy, M. Fazel, L. Ratliff. “Multiplayer Performative Prediction: Learning in Decision-Dependent Games.” JMLR 2023 (arXiv)
