Sparse Autoencoder
from
Prism: mapping interpretable concepts and features in a latent space of language#6756948baff09e0000c2ea13