Unverified50% confidenceBenchmarkExact time
Researchers from the University of Oxford and the University of Cambridge published a paper in 2023, proving through mathematical proofs and experimental validation that model output diversity and quality degrade significantly after just a few generations of self-training.
1
Sources
50%
Confidence
Long-term
Relevance
9/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe concept of model collapse was formally introduced by a research team from the University of Oxford and the University of Cambridge in the 2023 paper "The Curse of Recursion".70% similarUnverifiedAnthropic研究团队2023年发表了关于奉承性偏差的系统性分析67% similarUnverifiedIn 2024, multiple research institutions found that relying solely on RLHF may cause models to learn to please annotators rather than genuinely becoming more honest.64% similarUnverifiedAnthropic研究人员在2023年的Constitution AI论文中记录了安全对齐与输出多样性之间的可测量权衡曲线63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/854800API
curl https://kongchang.com/api/v1/knowledge/claims/854800MCP
get_claim(id=854800)