← back to paper
arxiv: 2509.02565 · 2 revisions
Understanding sparse autoencoder scaling in the presence of feature manifolds