504. Meta's AI Assistant can respond to users via voice. How would you balance the metrics tradeoff between voice quality (which requires more processing power and latency) and response speed (which users expect to be near- instant)?