Hello!

I am a Research Scientist at Google DeepMind on the Open-Endedness team. I am interested in developing autonomous agents that are safe, curious, and capable of endless open-ended learning and discovery! Our recent work represents a powerful convergence of world generation and embodied intelligence: Genie 3, a world model that can generate an unprecedented diversity of interactive environments in real-time (featured as one of TIME's 100 Best Inventions of 2025), and SIMA 2, a Gemini-powered agent that can play, reason, and learn in virtual 3D worlds. Together, these systems demonstrate how AI can both create rich environments and automatically self-improve in them—a crucial step toward general embodied intelligence.

Previously, I was a postdoctoral research and teaching fellow at the University of British Columbia and the Vector Institute, supervised by Prof. Jeff Clune. During this time, we developed The AI Scientist, the first agent to automate the entire scientific process (from forming hypotheses and conducting experiments to visualizing results, writing a paper, and reviewing it). Recently, The AI Scientist-v2 achieved another milestone by generating the first fully AI-written paper to pass peer review at an ICLR workshop. Our work has been featured in the State of AI Report for two consecutive years and in the press by Science News, Nature News, VentureBeat, Ars Technica, WIRED, IEEE Spectrum (and here), Forbes, Air Street Press, TechCrunch, MIT Technology Review, and Fortune. I have also been interviewed on how AI is transforming science on CBC's Quirks & Quarks, and the prospect of Self-Improving AI on ML Street Talk!

I previously received my PhD at the University of Oxford under the supervision of Prof. Michael A. Osborne and Prof. Yee Whye Teh. During my PhD, I focused on offline reinforcement learning—exploring topics such as generalization to unseen tasks, uncertainty quantification for offline world models, learning from pixels, and diffusion synthetic data for reinforcement learning. Please feel free to reach out!

You can find my PhD thesis here.

Recent News

• [11/2025] Excited to share SIMA 2, the new frontier for Gemini-powered agents that can reason, learn, and self-improve in virtual 3D worlds, representing a significant step toward general embodied intelligence.
• [10/2025] Thrilled that Genie 3 has been recognized as one of TIME's 100 Best Inventions of 2025! This recognition highlights the groundbreaking potential of world models to create interactive digital environments from text prompts.
• [10/2025] Honored that The AI Scientist has been featured in the State of AI Report for the second consecutive year, highlighting its progress in AI-assisted scientific research!
• [8/2025] Excited to share Genie 3, the most powerful foundation world model yet that can generate an unprecedented diversity of interactive environments in real-time. This represents a significant advancement in world models and embodied AI research!
• [5/2025] We released our paper, Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents, on self-improving AI, which was subsequently covered by Fortune and IEEE Spectrum.
• [3/2025] Gave a keynote talk at the AI4Research Workshop at AAAI 2025 on knowledge-grounded autonomous scientific research.
• [2/2025] Excited to be joining Google DeepMind on the Open-Endedness team! See you in London!
• [2/2025] We introduce Automated Capability Discovery (ACD)! ACD automatically identifies surprising new capabilities and failure modes in foundation models via self-exploration.
• [1/2025] Delighted that two papers were accepted to ICLR 2025: Intelligent Go-Explore and Automated Design of Agentic Systems. Thank you to my incredible collaborators Shengran and Jeff!
• [12/2024] Excited to be presenting at NeurIPS 2024 with The Edge-of-Reach Problem in Offline MBRL and Stable Control Representations.
• [10/2024] Excited to join as a visiting research scientist at Sakana AI!
• [9/2024] I am starting as a teaching fellow for 3rd year undergraduate and graduate machine learning. See you in CS340!
• [8/2024] We propose Automated Design of Agentic Systems, a new paradigm for automatically designing LLM agents.
• [8/2024] We introduce The AI Scientist, the first fully autonomous agent that can complete the entire scientific process!
• [8/2024] Our work was covered in Science News!
• [5/2024] Excited to introduce Intelligent Go-Explore, empowering foundation model agents to robustly explore in complex environments!
• [5/2024] Check out our new survey on video diffusion models!
• [4/2024] Our work on language-guided control won an outstanding paper award at the GenAI4DM Workshop!
• [1/2024] Excited to begin my postdoc at the University of British Columbia working on open-endedness!

Teaching Experience

• Course Instructor:
  • Fall 2024: CS340/540 Machine Learning and Data Mining @ UBC. Taught 36 lectures for a cross-listed undergrad/grad course of 258 students (8 TAs).
• Teaching Assistant:
  • 2022: Advanced Simulation (Statistics, Oxford)
  • 2021: Imperative Programming (Computer Science, Oxford)
  • 2021: Probability, Measure and Martingales (Mathematics, Oxford)

Academic Service

• Reviewing:
  • Journals: Nature Machine Intelligence
  • Conferences: AISTATS 2021, ICML 2022-24, NeurIPS 2022-24 (top 8% reviewer in 2022), ICLR 2024-25
  • Other: ICLR Tiny Papers 2023, Reincarnating RL Workshop @ ICLR 2023, NeurIPS MINT Workshop 2024
• Program Committee:
  • Foundation Models for Decision Making Workshop @ NeurIPS 2022-23
  • RL for Real Life Workshop @ NeurIPS 2022
  • Agent Learning in Open-Endedness Workshop @ ICLR 2022, NeurIPS 2023