Harry is a second-year DPhil at the OII, specialising in explainability and interpretability techniques for large language models (LLMs). His research investigates whether LLMs can produce reliable natural language explanations of their own decision-making, a key requirement for trustworthy AI and effective human-computer interaction. At the centre of his work, he is motivated by the challenge of mitigating the risks of advanced AI.
Alongside his primary research, Harry has contributed to several projects evaluating the capabilities of frontier LLMs. These include the LingOly and LingOly-TOO benchmarks, which assess whether LLMs can truly reason in ways typically attributed to biological intelligence, and forthcoming work examining whether LLM-based agents can autonomously conduct academic research. His work has featured at leading AI conferences.
Prior to his DPhil, Harry completed a BA in Economics at the University of Cambridge and an MSc in Data Science at the University of Oxford, where he was awarded the prize for best overall thesis. He is supervised by Dr Adam Mahdi (OII) and Associate Professor Jakob Foerster (Department of Engineering Science), and is a member of the OII’s Reasoning with Machines Lab (RML).
LLMs, explainability, interpretability, LLM evaluations and benchmarks, AI safety, alignment.
6 December 2024
Several researchers and DPhil students from the Oxford Internet Institute, University of Oxford, will head to Vancouver for the Thirty-Eighth annual Conference on Neural Information Processing Systems (NeurIPS) from 10-15 December 2024.
19 February 2024
We are delighted to announce the winners of the 2023 OII MSc Thesis Prizes.
