InterpretabilityYou trained a sparse autoencoder with thousands of latents on the 512 MLP neurons of a one-layer transformer, and want to label one latent, say 'Arabic text'.
InterpretabilityInterpretabilityInterpretabilityYou trained a sparse autoencoder with thousands of latents on the 512 MLP neurons of a one-layer transformer, and want to label one latent, say 'Arabic text'.