leogao comments on Improving Dictionary Learning with Gated Sparse Autoencoders

[leogao]

Makes sense that the shift would be helpful