Methods for improving output reliability such as Chain-of-Thought reasoning, Self-Consistency checks, and multi-sample voting all require additional computational steps that directly increase response time