Terry Zhang

Terry is a DPhil student working with Adam Mahdi, Christian Schroeder de Witt and Philip Torr. He previously obtained a BSc and MSc in Interdisciplinary Sciences at ETH Zurich, with a dual focus on computer science and physics. His research seeks to de-risk the rapid scaling of multi-agent systems that contain increasingly heterogeneous agents and operate over increasingly long horizons. His vertical focus is on stress-testing the reliability and generalisability of frontier evaluations, and on understanding transfer dynamics from a data-centric perspective; his horizontal focus spans formal verification (Lean), multimodal reasoning and actionable interpretability methods. Together, this work aims to provide worst-case assurance for frontier AI systems and best-case prediction for their use in accelerating research.