Unverified50% confidenceBenchmarkExact time
Reflexion框架在HotpotQA、HumanEval等基准上相比单次生成基线提升了10%~20%的准确率
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
Verification Loops Quadruple DeepSeek's Intelligence at One-Seventh the Cost of Claude Opus
hackernewshackernews7/7/2026
Related Claims
Unverified自我反思（self-reflection）由Shinn等人在2023年的Reflexion论文（NeurIPS 2023）中系统化，在HotpotQA、HumanEval等基准上相比单次生成基线提升了10%~20%的准确率78% similarUnverified在HotpotQA等多跳推理基准上，ReAct相比Act-only方案的任务完成率提升约10-15个百分点，同时幻觉率显著降低68% similarVerifiedThrough experiments on multi-hop reasoning benchmarks such as HotpotQA and FEVER, the ReAct paper demonstrated that compared to pure Chain-of-Thought reasoning, ReAct improved accuracy by approximately 10%–30% on tasks requiring external information retrieval.67% similarUnverifiedExperimental data shows that when key information is located at the beginning and end of the input sequence, the model's retrieval accuracy can be as high as over 90%, but when located in the middle, it may drop sharply to below 60%.63% similarUnverifiedFew-shot提示通过提供2-5个示例让模型理解模式，比纯文字描述准确率通常高出20%-40%62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/284408API
curl https://kongchang.com/api/v1/knowledge/claims/284408MCP
get_claim(id=284408)