How would you describe them?
High-Performance Hyperbolic Sparse Autoencoders for Mechanistic Interpretability
