website

Transformer Circuits Thread

by Anthropic

Anthropic's interpretability team publishes its research here as long, interactive papers — A Mathematical Framework for Transformer Circuits, Toy Models of Superposition, Scaling Monosemanticity, On the Biology of a Large Language Model — which together form the core literature of mechanistic interpretability. Hard going, but this is where the attempt to open the black box is actually happening, with monthly Circuits Updates on work in progress.