Unverified50% confidenceBenchmarkExact time
2022年Wei等人的论文《Chain-of-Thought Prompting Elicits Reasoning in Large Language Models》证明在few-shot示例中加入推理步骤可将PaLM 540B在GSM8K上的准确率从17.9%提升至58.1%
1
Sources
50%
Confidence
Long-term
Relevance
8/22/2026
First Seen
Sources
Related Claims
Unverified使用包含引导语的Chain-of-Thought提示，可使大语言模型在复杂推理任务上准确率提升40%以上71% similarUnverified过程奖励模型（PRM）在GSM8K等数学推理基准测试中显著提升了模型的推理准确率70% similarUnverified在MATH数学推理数据集上，采用元提示策略的模型表现比标准提示方法提升了约17%69% similarUnverified思维链提示与少样本示例可提升7B至14B参数量级模型在复杂编码任务上的成功率69% similarVerifiedChain-of-Thought提示技术由Google Brain团队在2022年的论文《Chain-of-Thought Prompting Elicits Reasoning in Large Language Models》中正式提出69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/785306API
curl https://kongchang.com/api/v1/knowledge/claims/785306MCP
get_claim(id=785306)