Unverified50% confidenceTradeoffExact time

单次实体抽取容易受幻觉影响导致图谱中出现虚构的节点或错误的边，多步验证策略通过增加中间可审查环节压缩错误传播空间

1

Sources

50%

Confidence

Long-term

Relevance

9/16/2026

First Seen

Sources

可解释的智能体检索：让RAG不只是"猜对答案"

Related Entities

实体抽取concept Chain-of-Thoughtconcept

Related Claims

Verified对一张被神经网络正确分类的图像叠加人眼几乎不可见的微小扰动后，模型输出会以极高置信度发生错误74% similar Unverified对抗样本攻击可以在图像上添加人眼难以察觉的噪声,却能让模型产生错误分类73% similar UnverifiedOpik内置幻觉检测功能，可识别模型生成的与上下文不符的虚构内容73% similar Unverified后见之明偏差可细分为记忆扭曲、必然性错觉和可预见性错觉三种亚型72% similar Unverified针对特定检测器的对抗性攻击可以通过在生成内容中添加人眼不可见的微小扰动来欺骗检测模型70% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/927911

API

curl https://kongchang.com/api/v1/knowledge/claims/927911

MCP

get_claim(id=927911)