Paper outlining three categories of understanding in LLMs (simple, contingent, principled).
AI Explained