The Mathematics of Large Language Models II: Concentration — A Depth Lower Bound for Robust Classification | Synapse