Unverified50% confidenceFactExact time
When a system continuously uses its own output as input, errors are continuously accumulated and amplified, and rare samples (long tails) in the original distribution are the first to disappear.
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified当生成模型反复在自身生成的数据上训练时，输出分布会逐渐退化，低概率事件被系统性遗忘76% similarUnverified当标注不一致存在时，增加更多数据只会成倍放大混乱，向模型提供自相矛盾的监督信号75% similarUnverifiedIn distributed systems, fault-tolerance mechanisms themselves can sometimes become amplifiers of fault propagation.73% similarUnverifiedLosing the prefill checkpoint at the end of the prompt under sparse sampling conditions can cause state corruption, output deviation, and even non-reproducible generation results.72% similarUnverifiedWhen the wording of a prompt changes slightly, the model's output distribution shifts accordingly, causing contradictions to emerge between successive conclusions.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/989855API
curl https://kongchang.com/api/v1/knowledge/claims/989855MCP
get_claim(id=989855)