Harry Mayne

Digital Social Sciences (2022 cohort)

Harry is a DPhil student at the University of Oxford, specialising in explainability and interpretability techniques for large language models (LLMs). His research focuses on determining whether LLMs can generate reliable natural language explanations of their own decision-making, a key requirement for transparent and safe artificial general intelligence. He is also exploring novel training methods to improve the trustworthiness of these explanations.

In addition to his primary research, Harry has been involved with several projects to measure the capabilities of LLMs. These have included benchmarks that test whether LLMs can truly reason in ways we typically attribute to biological intelligence and explore whether LLM-based agentic systems might begin to automate parts of the human academic research pipeline. This work has been featured at leading AI conferences.

Prior to his DPhil, Harry completed a BA in Economics from the University of Cambridge and an MSc in Data Science from the University of Oxford. He is currently supervised by Dr Adam Mahdi (Oxford Internet Institute) and Associate Professor Jakob Foerster (Department of Engineering Science). He is a member of the Reasoning with Machines AI Lab (RML) at the Oxford Internet Institute.