writing

Notes & deep dives

Long-form pieces where I take an idea apart until it makes sense, usually about machine learning, interpretability, and how models represent the world. Written to be read by anyone, not just researchers.

interpretability · superposition 11 min read

How a 10-neuron network serves 100 features

A tiny network shouldn't be able to do this. It has ten neurons and a hundred things to remember. Here's the clever trick it learns instead, and why that trick is the reason neurons inside real AI models are so hard to read.

read the piece →