How LLMs Learn: Tracing Internal Representations with Sparse Autoencoders

Open in new window