A blog about math, science, and gremlins.


Sparse autoencoding and the Tanner-Donoho crossing

March 8, 2026. Sparse autoencoders (SAEs) are a crucial mechanism for extracting interpretable information from a neural network. In this post, we explore the decay of recoverability with the size of the SAE, guided by work from compressed sensing.

Reflections on Bell Labs: Part 1

February 6, 2024. This essay reflects on what made Bell Labs so successful as an β€œinstitute of creative technology”, why that success has been so hard to replicate, and how one might attempt to replicate it in a distributed, self-organized, and rebellious fashion.

Six Impossible Things: Issue 1

January 17, 2024. In the first issue of 6IT: the mathematics of supervising children; unreliably clairvoyant demons; why the sky is blue (and why it isn’t); a Poisson process for primes; sychronizing fireflies; and quantum error correction from Penrose tilings.

Hello world

December 2, 2023. Introduction to the site and its author.