Why Linear Interpretability Works: Invariant Subspaces as a Result of Architectural Constraints

Add code
Feb 10, 2026

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: