Miscellaneous
Talks & Media
The Emergence of Intelligence: From Bitter Lessons to Universal Minds
Talk at Oxford University on the emergence of capabilities and behaviour in AI systems and their links to classical emergence.
June 2025
Structured World Representations in Maze-Solving Transformers
Podcast discussion about our work on understanding how transformers learn to represent and solve maze navigation tasks.
June 2024
Sparse Relational Reasoning with Object-Centric Representations
Conference talk from ICML 2022 DyNN Workshop on the benefits of sparsity and structured representations for neural relational reasoning.
July 2022
Projects
Curated and automated collection of AI Safety papers and resources
A tool for running machine learning experiments across different compute environments (SSH, Local, etc.)
haiku_probe ↗
2022A set of lightweight tools for mechanistic interpretability, probing ANNs built in Haiku. Implements probing/interventio...
1st Place Winner at AIHack2022 - Microfluidic Droplet Coalescence Prediction using Hamiltonian Neural Networks.
Analysis of salient neurons representing grid cell activity using neuropixel data from mice, developed for the Neuroscie...