Review Bot comments on Sparse Autoencoders Find Highly Interpretable Directions in Language Models