Show HN: LLMs think in hierarchies. Their interpretability tools should too