I have many thoughts. This is where I share some of them.

2026

What Would Non-Linear Features Actually Look Like?

2025

Favourite Papers of 2025
paper Adversarial Examples Are Not Bugs, They Are Superposition
Adversarial Examples Aren't Bugs, They're... Superposition?

2024

paper Group Crosscoders for Mechanistic Analysis of Symmetry
Interpretable Features and Circuits in InceptionV1's Mixed5b
Curve Detector Manifolds in InceptionV1
[Poster] The Missing Curve Detectors of InceptionV1
The Missing Curve Detectors of InceptionV1
Hello!