How do you think about the trade-off between accuracy, latency, and computational cost in an AI system?
