Behavioral and mechanistic definitions (often confuse AI alignment discussions) — LessWrong