People

Visiting researcher, Meridian Cambridge

Mechanistic interpretability researcher with grants on hidden fine-tuned behaviours and eval awareness. Wrote the Tokenized SAEs paper and the ezinterp library. Grad school in Computation and Neural Systems; has performed in over 150 improv shows in LA.

Iliad Fellow

MSc in theoretical physics from the Netherlands and MSt in philosophy of physics from Oxford. AI safety projects with SPAR and Dovetail; April 2026 Iliad Intensive alum. Also worked on building a satellite that is now in orbit.

PhD student, Australian National University; Head of Teaching, ARENA

Works on universal AI with Marcus Hutter, and co-wrote the textbook An Introduction to Universal Artificial Intelligence.

Visiting researcher, Meridian Cambridge

Works on encoded reasoning, steganography and chain-of-thought monitoring. Co-author of Steering Llama 2 via Contrastive Activation Addition. Wrote and taught Iliad's ML engineering and mechanistic interpretability days.

MATS extension fellow; Iliad Fellow

Physicist by training. Researches, with Richard Ngo, how and when neural networks learn structured computation; see A Spectral Theory of Computation. Speaks Latin, Attic Greek and Mandarin.

Iliad Fellow

Physicist; PhD from UC Santa Barbara on traversable wormholes and entanglement. Won Apart's AI Safety x Physics challenge applying computational mechanics to transformers. April 2026 Iliad Intensive alum.

Chief Scientist, Beneficial AI Foundation

PhD on the algebraic-geometric aspects of singular learning theory. Works on deductive program synthesis and on the information theory of spiking networks.

Miller Postdoctoral Fellow, UC Berkeley

PhD from Stanford. Works on deep learning theory: training dynamics, symmetries and conservation laws, feature learning. Known for the SynFlow pruning method.

Professor of Geometry and Physics, University of Edinburgh

Works on the mathematics of quantum field theory and string theory, and on bridging the two. Iliad Fellow, interested in a scientific theory of deep learning.

Co-founder and Chief Scientist, Simplex AI Safety

Theoretical physicist (PhD, UC Davis). Applies computational mechanics to what neural networks represent internally. Research lead at the Astera Institute and a MATS mentor.

XOR Labs, with Fernando Rosas

Earlier work on the thermodynamics of computation: Maxwell's demon and information ratchets. With Rosas, wrote AI in a Vat on the limits of world modelling.

Teaching and Delivery Lead, Iliad; PhD candidate, University of Amsterdam

Started in mechanistic interpretability, with a review of the field and a measure of superposition; now works on human agency among AI agents. Co-founded the AI Safety Initiative Amsterdam.

Research scientist, Astera Institute / Simplex

PhD from MIT in Max Tegmark's group. Known for The Quantization Model of Neural Scaling and for work on scaling sparse autoencoders.

Iliad Fellow

Maths graduate of the Australian National University, from Canberra. Works on agent foundations and abstractions, and on black-box methods for studying AI systems. April 2026 Iliad Intensive alum.

Alignment Research Center (ARC)

Independent researcher

Works on pluralistic alignment: how agents coordinate, abstain or ask for more information when their incentives differ. Background in computational cognitive neuroscience and reinforcement learning, with research roles at Caltech, UC Berkeley, CHAI and MATS.