Verified80% confidenceFactExact time
Model collapse refers to the phenomenon whereby, after an AI model is trained on data generated by a previous-generation model, the output distribution gradually deviates from the original real data distribution across multiple generations of iteration, losing information in the tail of the distribution.
6
Sources
80%
Confidence
Long-term
Relevance
9/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen language models are repeatedly trained on data generated by themselves or similar models, systematic degradation occurs, with the diversity of model outputs decreasing with each generation and the tail information of the distribution gradually being lost.79% similarUnverified2023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78% similarUnverified不同AI模型在预训练数据分布和RLHF对齐策略上存在差异，导致对同一问题有不同的推理路径，因此换模型有时能解决单一模型反复失败的问题75% similarVerifiedModel drift refers to the decline in model prediction quality caused by changes in the distribution of input data in the production environment over time.75% similarUnverifiedThe capability degradation phenomenon (also known as 'model collapse') has been observed multiple times when fine-tuning with synthetic data generated by the model itself.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/838835API
curl https://kongchang.com/api/v1/knowledge/claims/838835MCP
get_claim(id=838835)