Unverified50% confidenceFactExact time
Self-Refine（Madaan等，2023）让模型生成初始输出后以批评者身份评估并给出改进建议再重新生成
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedSelf-reflection is achieved by having the model evaluate the quality of its own output and iteratively improve it; the Reflexion framework demonstrates that an agent can learn from failed attempts and adjust its strategy in the next iteration.69% similarUnverifiedDesigners can annotate specific parts of the prototype and propose revisions, and the model will iterate and update accordingly.67% similarUnverified2024年中期推理模型首次上线时用户反馈简单问题等待过久，催生了自动路由机制65% similarUnverifiedActivation patching was widely popularized by Meng et al. in the 2022 ROME paper; its principle is to transplant the activation values of a certain layer from one run into another run to observe output changes, thereby performing causal inference.65% similarUnverified宪法AI框架让模型依据预定义的原则列表对自身输出进行批判与修订，再通过RLHF进一步优化65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/572478API
curl https://kongchang.com/api/v1/knowledge/claims/572478MCP
get_claim(id=572478)