Unverified50% confidenceCautionExact time
AI生成的代码有时会在不经意间对全量数据先做归一化再分割，产生隐性数据泄漏
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedAI在语境模糊时倾向于将本应用行内代码块呈现的内容硬拆成多行代码块，因训练语料中代码示例多以多行代码块形式出现80% similarUnverifiedAI tends to generate correct but inefficient code, and large-scale data processing scenarios require manual optimization intervention.78% similarUnverifiedAI code generation models are prone to errors in scenarios such as boundary condition handling, random seed setting, dataset splitting order, and mixed use of multi-class and binary classification APIs.78% similarVerifiedAI代码生成的幻觉率与上下文的歧义程度正相关，项目约定越模糊模型越倾向于发明看似合理的实现78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/533693API
curl https://kongchang.com/api/v1/knowledge/claims/533693MCP
get_claim(id=533693)