Sumeet Motwani

UC Berkeley

prof_pic.jpg

Hi! I’m Sumeet, an ML Researcher studying Computer Science at UC Berkeley. I’m working on Multiagent Systems, Security, Reward Learning, and Foundation Model Safety.

My current research focus is on Language Agents at Berkeley Artificial Intelligence and on Agentic Collusion with research groups in the UK. Previously, I’ve done internships/residencies at Redwood Research, Solana Labs, Convergent Finance, and GEn1E Lifesciences.

This website is under construction.

LinkedIn, Google Scholar, Future of Life Institute

updates

Jan 20, 2024 STARC has been accepted to ICLR 2024!
Dec 15, 2023 Check out Christian’s talk on our ongoing work at NOLA 2023
Oct 27, 2023 Perfect Collusion has been accepted to NeurIPS MASEC Workshop
Oct 7, 2023 Goals
Sep 26, 2023 STARC released and under review
Aug 15, 2023 Completed an internship working on DL for Drug Discovery and clinical trial site identification at GEn1E, a YC/Khosla Ventures company
Jul 22, 2023 Accepted to Berkeley’s EECS Honors Program
Feb 10, 2023 Completed a Research Residency at Redwood Research, focusing on mechanistic interpretability
Sep 10, 2021 Started at Berkeley AI Research under Professor Dawn Song. Currently under Professor Avideh Zakhor

papers

  1. Under Review
    Secret Collusion Among Generative AI Agents
    Sumeet Ramesh Motwani, Mikhail Baranchuk, Martin Strohmeier, Vijay Bolina, Philip H. S. Torr, Lewis Hammond, and Christian Schroeder Witt
    2024
  2. ICLR 2024
    STARC: A General Framework For Quantifying Differences Between Reward Functions
    J. Skalse, L. Farnik, S. R. Motwani, E. Jenner, A. Gleave, and A. Abate
    The Twelfth International Conference on Learning Representations, Sep 2023
  3. NeurIPS MASEC
    A Perfect Collusion Benchmark: How can AI agents be prevented from colluding with information-theoretic undetectability?
    S. R. Motwani, M. Baranchuk, L. Hammond, and C. S. Witt
    In Multi-Agent Security Workshop, NeurIPS’23, Oct 2023