"AI actors usually look hollow because they don't listen. Using
@Meta's annotated 4,000-hour dataset of natural human face-to-face interactions, we’re training AI performances built on genuine behavioral dynamics: reactive eye-lines, conversational latency, and physical subtext."