Rohin Shah comments on Showing SAE Latents Are Not Atomic Using Meta-SAEs