Mengyu Ye

I am a 2nd-year Ph.D. student in NLP at the [Fundamental AI lab] (https://www.fai.cds.tohoku.ac.jp/)(member of the Tohoku NLP Group) at Tohoku University, advised by Prof. Jun Suzuki. I am also a Google PhD Fellow. My research examines the behavior of large language models through precise, controlled experimentation. I design synthetic tasks and datasets, build training, inference, and evaluation pipelines, and use causal methods to characterize how models reason and where they fail.

My work focuses on uncovering overlooked failure modes and validating their underlying mechanisms, with resulting publications at NeurIPS, ACL, and EMNLP. A central aim of my research is to develop tools and methodologies that make model capabilities more predictable, reliable, and controllable.

I am currently extending this approach to diffusion language models, studying how post-training and inference-time algorithms shape coherence and reasoning. I also explore agentic systems, applying the same experimental discipline; our deep-research agent received the Best Static Evaluation Prize in the MMU-RAG competition at NeurIPS 2025.