L lesswrong.com Negative Results for Sparse AutoEncoders on Downstream Tasks 2 points ottaborra 1 year ago 0 comments