Nathan Helm-Burger comments on Do Sparse Autoencoders (SAEs) transfer across base and finetuned language models?