Verified70% confidenceFactExact time
A 2023 joint study by Oxford University, Cambridge University, and other institutions elaborated on the model collapse effect: when a model is iteratively trained on its own generated data, the statistical variance of outputs continuously decreases, rare long-tail knowledge disappears first, and the model ultimately converges to a cultural mean state.
4
Sources
70%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Regression to the Mean: How LLMs Are Quietly Killing Innovation and Diversity
hackernewshackernews7/6/2026
Related Claims
Unverified牛津大学2023年的一项研究量化了模型坍缩效应，仅经过五代自我训练，文本多样性指标下降了约40%82% similarUnverifiedIn 2023, researchers from Oxford University and other institutions published a study finding that after only 5–10 generations of recursive training, the degradation in model output quality became visible to the naked eye.75% similarUnverifiedThe study additionally introduced six deliberately weakened models with clearly inferior capabilities as positive controls.68% similarUnverifiedIn the large model field, first-mover advantage is being weakened: model capabilities may be matched or even overtaken every few months, and user migration cost mainly comes from data and workflows rather than the product itself.65% similarUnverifiedAnthropic's claimed 40% reduction is based on a comparison using default settings (the new model defaults to mid-tier, while the previous generation defaulted to high-tier); at the same tier, the new model thinks more, so actual consumption is not necessarily lower.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/165887API
curl https://kongchang.com/api/v1/knowledge/claims/165887MCP
get_claim(id=165887)