Fetching the paper…

Can sparse autoencoders be used to decompose and interpret steering vectors? · Around