[KongchangAI]
Unverified50% confidenceFactExact time

自回归语言模型的OOD检测比传统机器学习更困难,因为语言组合空间庞大、语义OOD与表面形式OOD不对应,且对抗者可构造统计上看似in-distribution但语义触发OOD的输入

1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen

Sources

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/784179
API
curl https://kongchang.com/api/v1/knowledge/claims/784179
MCP
get_claim(id=784179)