2026-01-08 | ARBOx Day 4

Goal: Day 4 of ARBOx: Superposition and SAEs

Summary: ARENA 1.3 Notebook: Introduction to Superposition and using SAEs

Work sessions

In	Out
10:00	18:00

While I have an intuition of the Toy Models of Superposition and how feature directions often have entangled concepts, reading the original Toy Models of Superposition will help me improve my theoretical understanding of MechInterp.

Added to todos

Also added Towards Monosemanticity: Decomposing Language Models With Dictionary Learning