LLM training data mixture optimization breaks when training pools shift — every prior proxy experiment becomes stale.

Think saying um and uh makes you sound unprepared? These tiny hesitation words play an important role in how we think, speak, and communicate.