“The concept of scaling laws has been successfully applied to interpretability research to project the training requirements for large-scale sparse autoencoders.”