Self-Consistency Improves Chain of Thought Reasoning in Language Models

• 类型：arxiv
• 标识：2203.11171
• 链接：https://arxiv.org/abs/2203.11171
• 主题：evaluation
• 主分类：llm-infra
• 形态：method
• 被引：7499
• 被引来源：Semantic Scholar
• S2被引：7499
• OpenAlex被引：704
• 影响力被引：942
• TLDR：This paper proposes a new decoding strategy, self-consistency, to replace the naive greedy decoding used in chain-of-thought prompting that first samples a diverse set of reasoning paths instead of only taking the greedy one, and then selects the most consistent answer by marginalizing out the sampled reasoning paths.
• OpenAlex ID：W4221161695
• OpenAlex DOI：10.48550/arxiv.2203.11171
• DOI：10.48550/arxiv.2203.11171
• DOI来源：OpenAlex
• 开放获取：green
• 开放获取链接：https://arxiv.org/pdf/2203.11171
• OpenAlex更新：2026-07-24
• 待LLM分类：否
• 成熟度：production
• 场景：chain-of-thought、reasoning、self-consistency
• 标题中文：Self-Consistency Improves Chain of Thought Reasoning in Language Models
• TLDR中文：本文提出了一种新的解码策略——self-consistency，用于替代思维链 prompt 中使用的朴素贪心解码：首先采样一组多样化的推理路径，而非仅取贪心路径，然后通过对采样路径进行边缘化来选择最一致的答案。
• 来源文件：
• [OpenAlex discover]
• [S2 enrich]