Ken Gu


ken-gu.jpeg


I work on the evaluation and development of language models and agents, with a focus on scientific and technical domains. I care about building benchmarks that measure reasoning, and agents that can plan and reliably carry out open-ended tasks.

I am currently a Member of Technical Staff at xAI working on language model reasoning, and a PhD student in Computer Science at the University of Washington (on leave), advised by Tim Althoff. I have also conducted research at Google, Salesforce, and Microsoft Research.


Selected Publications

2026

  1. synthworlds-teaser.png
    SynthWorlds: Controlled Parallel Worlds for Disentangling Reasoning and Knowledge in Language Models
    Ken Gu, Advait Bhat, Mike A. Merrill, Robert West, Xin Liu, and 2 more authors
    ICLR 2026
    arXiv PDF Code
  2. collaborative-scaling.png
    Completion ≠Collaboration: Scaling Collaborative Effort with Agents
    Shannon Zejiang Shen*, Valerie Chen*,  Ken Gu, Alexis Ross, Zixian Ma, and 9 more authors
    ACL Findings 2026
    🏅 NeurIPS Workshop on Responsible Foundation Models 2025 Best Paper
    arXiv PDF
  3. scaling-agent-systems.png
    Towards a Science of Scaling Agent Systems
    Yubin Kim,  Ken Gu, Chanwoo Park, Chunjong Park, Samuel Schmidgall, and 14 more authors
    Preprint 2026
    arXiv PDF Blog

2025

  1. radar-teaser.png
    RADAR: Benchmarking Language Models on Imperfect Tabular Data
    Ken Gu, Zhihan Zhang, Kate Lin, Yuwei Zhang, Akshay Paruchuri, and 16 more authors
    NeurIPS 2025
    Integrated in Gemini
    arXiv PDF Code

2024

  1. blade.png
    BLADE: Benchmarking Language Model Agents for Data-Driven Science
    Ken Gu, Ruoxi Shang, Ruien Jiang, Keying Kuang, Richard-John Lin, and 11 more authors
    EMNLP 2024
    arXiv PDF Code Website

Updates

Jan 2026 Two papers from my PhD are accepted to ICLR 🇧🇷 and CHI 🇪🇸!
Nov 2025 I joined xAI as a Member of Technical Staff.
Sep 2025 Update from my 1 year at Google 🚀 I led two exciting projects: RADAR, a dataset advancing Gemini’s tabular and data science reasoning, recently accepted at NeurIPS, and our 145-page Personal Health Agent paper, introducing a multi-agent framework that integrates data from wearables and personal health records to drive personalized health insights.
Oct 2024 🎙️ Gave an invited talk on BLADE at AI2. I enjoyed the insightful discussions that followed, especially on how we can approach evaluation for data-driven science and open-ended tasks.
Sep 2024 🍂 Thrilled to start an internship at Google Research, focusing on agents for personal health data and building upon insights from our BLADE benchmark!
Jan 2024 🧙 Excited to share that two papers on understanding human-AI collaboration in data science have been accepted to CHI 2024!! One stems from my internship with Microsoft Research last summer, and the other is a Wizard-of-Oz study conducted with collaborators at UW, where we acted as LLM data analysis assistants.
Jun 2023 🏔 Started my internship at Microsoft Research with Chenglong Wang and Steven Drucker!
Apr 2023 🇩🇪 Attended CHI 2023 in Hamburg, Germany! This was my first in-person conference and my first time in Europe!