AI Researcher // PhD Student

Zayne
Sprague

Scholar GitHub LinkedIn Semantic Scholar CV
Zayne Sprague

About

I'm a PhD student at NYU, advised by Greg Durrett. My passion is understanding and instilling reasoning into AI models — think you know what "reasoning" is? Reach out, I'd love to hear your take.

Currently interning at Google in Seattle, working with Adam Fisch, Jacob Eisenstein, and Fantine Huot.

TAUR Lab

Research Codex

Selected papers · Full list on Scholar

ICLR 2026

SkillFactory: Self-Distillation for Learning Cognitive Behaviors

Zayne Sprague, Jack Lu, Manya Wadhwa, Sedrick Keh, Mengye Ren, Greg Durrett
ICLR 2026 Oral

OpenThoughts: Data Recipes for Reasoning Models

Etash Guha, Ryan Marten, Sedrick Keh, Negin Raoof, Georgios Smyrnis, Hritik Bansal, Marianna Nezhurina, Jean Mercat, Trung Vu, Zayne Sprague, +40 authors
COLM 2025

EvalAgent: Discovering Implicit Evaluation Criteria from the Web

Manya Wadhwa, Zayne Sprague, Chaitanya Malaviya, Philippe Laban, Junyi Jessy Li, Greg Durrett
ICLR 2025

To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning

Zayne Sprague, Fangcong Yin, Juan Diego Rodriguez, Dongwei Jiang, Manya Wadhwa, Prasann Singhal, Xinyu Zhao, Xi Ye, Kyle Mahowald, Greg Durrett
ICLR 2024 Spotlight

MuSR: Testing the Limits of Chain-of-thought with Multistep Soft Reasoning

Zayne Sprague, Xi Ye, Kaj Bostrom, Swarat Chaudhuri, Greg Durrett
NLRSE @ ACL 2023

Deductive Additivity for Planning of Natural Language Proofs

Zayne Sprague, Kaj Bostrom, Swarat Chaudhuri, Greg Durrett
AAAI 2023 Demo

SocialGym 2.0: Simulator for Multi-Agent Social Robot Navigation in Shared Human Spaces

Zayne Sprague*, Rohan Chandra*, Joydeep Biswas
EMNLP 2022

Natural Language Deduction with Incomplete Information

Zayne Sprague, Kaj Bostrom, Swarat Chaudhuri, Greg Durrett
EMNLP Findings 2022

Natural Language Deduction through Search over Statement Compositions

Kaj Bostrom, Zayne Sprague, Swarat Chaudhuri, Greg Durrett

// Blogs & Projects

Open-source tools & write-ups

Open SourceTool

RACA: Research Assistant Coding Agent

One Claude Code session: design experiments, run them on your compute, analyze the results. RACA bridges Claude Code and your compute infrastructure—Slurm clusters, Runpod, or local GPUs—so one conversation handles the full research lifecycle: design, red-team review, job submission, monitoring, artifact management, and a live results dashboard.

Code Blog Post

Contact

Interested in reasoning, language models, or just want to debate what "reasoning" really means? I'd love to hear from you.

> zrs2020@nyu.edu (nyu)
> zaynesprague@gmail.com (personal)