Has anyone tried using Sparse Autoencoders for automated ontology induction? Train a task-specific model (such as embeddings), use SAEs to expand its representations and find composite features, then distill that into a taxonomy. The model already learned something valuable.