Yulia Tsvetkov
University of Washington associate professor working on reliable, socially aware, and multilingual NLP
- Associate Professor, Paul G. Allen School of Computer Science & Engineering, University of Washington1
- Adjunct Professor, Language Technologies Institute (previously Assistant Professor), Carnegie Mellon University1
Yulia Tsvetkov is an associate professor in the Paul G. Allen School of Computer Science & Engineering at the University of Washington and an adjunct professor at Carnegie Mellon University's Language Technologies Institute, where Tsvetkov was previously an assistant professor; before that, Tsvetkov was a postdoctoral researcher in the Stanford NLP Group. Tsvetkov's group studies the capabilities and limitations of large language models, including reasoning and personalization, AI ethics and safety, multilingual and culturally inclusive modeling, and high-stakes applications in health and science. Tsvetkov is a co-author of Spurious Rewards: Rethinking Training Signals in RLVR, which examines why reinforcement learning with verifiable rewards can improve math reasoning in some models even with uninformative rewards.12
Work
Spurious Rewards: Rethinking Training Signals in RLVR (external site: arxiv.org)Published research
Co-author of the 2025 paper showing that RLVR training with GRPO can raise math benchmark scores for certain models even with random or incorrect rewards, and tracing the effect to a clipping bias that amplifies behaviors learned in pretraining.2
Earlier roles
- Postdoctoral researcher, Stanford NLP Group, Stanford University1
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.