Skip to main content

Tiny SAE

Crack a layer's tangled activations into thousands of single-meaning features.

Key Insight​

This project trains a small sparse autoencoder (SAE) on the residual stream of a small LLM and visualizes a handful of the recovered features — directions in activation space that fire for exactly one concept ("Golden Gate Bridge," "is a JSON key," "negation in a clause").

Why This Matters​

SAEs decompose a layer's dense, polysemantic activations into a much larger but mostly-zero set of monosemantic features, giving the most promising current vocabulary for "what is the model thinking" — the working dictionary that drives interpretability research in 2024–2025.