Unverified50% confidenceFactExact time
数据污染问题源于大模型在包含 GitHub、Stack Overflow、LeetCode 等平台公开代码题解的语料上进行预训练
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
Grok vs GPT vs Claude: Which AI is the Best Coding Assistant?
hackernewshackernews7/8/2026
Related Claims
UnverifiedData contamination refers to the possibility that a model's training data already includes bug fix solutions publicly available on GitHub, meaning inference during testing may actually be memory recall.75% similarUnverified大模型在预训练阶段接触了海量GitHub上的diff格式数据，对该格式有极高的理解与生成能力71% similarUnverifiedThe pre-training corpora of current mainstream large models are highly dependent on publicly available internet data, including CommonCrawl, Wikipedia, and the GitHub codebase.71% similarUnverified代码训练模型在构建训练集时会特别关注去重处理，因为GitHub上大量fork和复制代码会导致模型过度记忆特定代码片段而非学习通用模式70% similarUnverifiedMainstream benchmarks like SWE-Bench Pro heavily rely on issues and commits from public repositories, creating data contamination risk for models trained on public GitHub code.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/457375API
curl https://kongchang.com/api/v1/knowledge/claims/457375MCP
get_claim(id=457375)