I don't think HN understands how important model distillation still is for federated learning. Hype >> substance ITT