Unverified50% confidenceFactExact time
能力退化现象（也称'模型坍缩'）在用模型自身生成的合成数据进行微调时已被多次观察到
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2023年牛津大学Ilia Shumailov团队在Nature上发表研究系统阐述了模型崩溃效应，指出模型在自身生成数据上迭代训练相当于对原始数据分布有损压缩74% similarVerified模型坍缩（Model Collapse）概念最早由牛津大学和剑桥大学研究团队在2023年论文《The Curse of Recursion: Training on Generated Data Makes Models Forget》中系统阐述71% similarUnverified评估与合成数据生成、模型蒸馏等场景高度重叠70% similarUnverifiedSynthetic data used to train large language models carries the risk of 'Model Collapse'.70% similarUnverified模式崩塌（Mode Collapse）术语源自生成对抗网络研究，指生成模型倾向于产生少数几种安全的输出模式69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/831916API
curl https://kongchang.com/api/v1/knowledge/claims/831916MCP
get_claim(id=831916)