Technique: Self-Consistency Sampling
Self-consistency takes the chain-of-thought technique further by generating multiple independent reasoning paths for the same problem. Each path may use different approaches, assumptions, or methods — but if they converge on the same answer, confidence is high. If they diverge, the majority answer is typically most reliable.
This mirrors how critical decisions are made in engineering and science: multiple independent analyses of the same problem provide higher confidence than a single analysis, no matter how thorough. In production AI systems, this is implemented by calling the API multiple times with higher temperature and taking the most common answer (majority voting).
When to use: High-stakes decisions, complex math problems, diagnostic tasks, ambiguous questions with multiple valid interpretations, and any scenario where you need confidence in the answer. Particularly valuable when the cost of being wrong is high.