Zachary Horvitz

Research Interests

I am excited about eliciting and estimating rare behaviors in generative models, inference-time control + adaptive reasoning, and building robust verifiers for text generation tasks! I am currently supported by a Google PhD Fellowship.

Selected Publications

• Estimating Tail Risks in Language Model Output Distributions (ICML 2026, Spotlight)
  Rico Angell*, Raghav Singhal*, Zachary Horvitz*, et al. [arXiv]
• No Compute Left Behind: Rethinking Reasoning and Sampling with Masked Diffusion Models (Preprint)
  Zachary Horvitz*, Raghav Singhal*, et al. [arXiv]
• A General Framework for Inference-time Scaling and Steering of Diffusion Models (ICML 2025)
  Raghav Singhal*, Zachary Horvitz*, et al. [arXiv]
• StyleDistance: Stronger Content-Independent Style Embeddings with Synthetic Parallel Examples (NAACL 2025)
  Ajay Patel, Jiacheng Zhu, Justin Qiu, Zachary Horvitz, et al. [Semantic Scholar]
• TinyStyler: Efficient Few-Shot Text Style Transfer with Authorship Embeddings (EMNLP 2024)
  Zachary Horvitz, et al. [ACL Anthology]
• ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer (AAAI 2024)
  Zachary Horvitz, et al. [AAAI]
• Getting Serious about Humor: Crafting Humor Datasets with Unfunny Large Language Models (ACL 2024, Outstanding Paper Award)
  Zachary Horvitz, et al. [ACL Anthology]

Experience & Education

Prior to my PhD, I was a Machine Learning Engineer and then Research Lead at Rad AI.

I completed my ScM and AB at Brown University, where I studied Computer Science and Anthropology. During that time, I wrote for The Brown Noser, Brown's satirical newspaper.