Self‑Consistency
Chain‑of‑Thought is powerful, but a single reasoning path might be wrong. Self‑consistency runs the same CoT prompt multiple times (with a bit of randomness) and takes the majority answer. It’s like asking five experts and trusting the most common opinion.
Self‑consistency = run CoT several times, pick the answer that appears most often.
How It Works
1. Ask a question with a CoT prompt (e.g., "Let's think step by step").
2. Run the same prompt 3‑5 times (use temperature > 0, e.g., 0.5).
3. Collect all final answers.
4. Pick the answer that occurs most frequently.
Example: Arithmetic Reasoning
Run 1: 11
Run 2: 11
Run 3: 10
Majority vote → 11 (correct).Why This Works
LLMs are probabilistic. For many problems, the correct reasoning path is more probable than wrong ones. By sampling multiple times, the correct answer emerges as the mode. It’s like averaging out the noise.
When to Use Self‑Consistency
- Problems where one wrong step can derail the answer
- When you cannot afford a mistake (but can afford multiple API calls)
- Tasks with a single correct numeric or categorical answer
Note on Cost
Self‑consistency uses more API calls and tokens, so it costs more and takes longer. Use only when accuracy is critical.
Two Minute Drill
- Self‑consistency runs CoT multiple times and takes the majority answer.
- Requires temperature > 0 to get variety.
- Improves reliability for hard reasoning tasks.
- Costs more; use only when needed.
Need more clarification?
Drop us an email at career@quipoinfotech.com