Patrick Leask

About

I'm a PhD student at Durham University supervised by Noura Al Moubayed. Previously, I completed the Constellation Astra Fellowship with Charlie Griffin from UK AISI and MATS 6.0 with Neel Nanda, and worked as a software engineer in industry.

Links

• Curriculum Vitae
• GitHub
• Google Scholar
• LessWrong
• LinkedIn

Publications

• Inference-Time Decomposition of Activations (ITDA): A Scalable Approach to Interpreting Large Language Models.
  P. Leask, N. Nanda, N. Al Moubayed. ICML 2025.
• Sparse Autoencoders Do Not Find Canonical Units of Analysis.
  P. Leask, B. Bussmann, M. Pearce, J. Bloom, C. Tigges, N. Al Moubayed, L. Sharkey, N. Nanda. ICLR 2025.
• BatchTopK Sparse Autoencoders.
  B. Bussmann, P. Leask, N. Nanda. NeurIPS 2024 Workshop on Scientific Methods for Understanding Deep Learning , 2024.