Unverified50% confidencePredictionExact time
As the proportion of AI-generated content on the internet rises, models that crawl web data for pre-training in the future will find it increasingly difficult to avoid content pollution sources.
1
Sources
50%
Confidence
Long-term
Relevance
9/13/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedAI-generated content flooding the internet in large volumes will become training data for the next generation of models, forming a self-reinforcing feedback loop that causes the tails of the distribution to gradually disappear.77% similarUnverifiedIf AI-generated content is allowed to contaminate the data pool without restraint, the next generation of models may suffer from increasingly incomprehensible and increasingly homogenized outputs.75% similarUnverifiedAI倾向于过度工程化，这是因为训练数据大量来自GitHub开源项目、技术博客和Stack Overflow，这些内容天然偏向展示最佳实践和设计模式74% similarUnverified随着AI Agent能力规模扩大，类似被娱乐内容吸引的行为模式是否会从训练数据中涌现出来，目前仍是一个开放性问题73% similarUnverified敏感信息传入AI上下文后存在多个潜在风险面：日志留存、模型训练数据混入，以及若AI被授权网络访问时可能的外发行为72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/919473API
curl https://kongchang.com/api/v1/knowledge/claims/919473MCP
get_claim(id=919473)