How does the self-consistency method improve chain-of-thought answers?
Sample many reasoning paths and keep the most common final answer.
- Ask the model if it is sure
- Sample many reasoning paths, take the majority answer ✔
- Make the model reason only once, greedily
Why: Hard problems can be solved many ways that all land on the same right answer. Self-consistency samples several chains of thought and picks the answer they agree on most, which beat a single greedy chain.
Huh, didn't know that: Self-consistency raised accuracy on the GSM8K math problems by 17.9 percent over plain chain-of-thought.
Self-consistency is how I pick a restaurant. Ask the family five times and go with the most common answer.
References
Answer: Sample many reasoning paths, take the majority answer. Hard problems can be solved many ways that all land on the same right answer. Self-consistency samples several chains of thought and picks the answer they agree on most, which beat a single greedy chain.
- Wang et al., Self-Consistency Improves Chain of Thought Reasoning in Language Models (arXiv 2203.11171) arxiv.org/abs/2203.11171
It first samples a diverse set of reasoning paths instead of only taking the greedy one, and then selects the most consistent answer
Checked 2026-10-10. - Kojima et al., Large Language Models are Zero-Shot Reasoners (arXiv 2205.11916) arxiv.org/abs/2205.11916
LLMs are decent zero-shot reasoners by simply adding "Let's think step by step" before each answer
Checked 2026-10-10.
Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.
More stuff everyone should know
That's stuff everyone should know.