[Filler] Sparse Autoencoder Feature Geometry in Small Transformers
How SAE features arrange themselves across layers, and what that implies for circuit-level interpretability.
How SAE features arrange themselves across layers, and what that implies for circuit-level interpretability.