Theoretical Bottlenecks of Million-Token-Per-Second Inference

A thought experiment explores the implications of million-token-per-second throughput, noting that extreme speed shifts the bottleneck from content generation to human decision-making and evaluation. Increased velocity does not improve reasoning capabilities but necessitates faster judgment of rapid-fire options.