Unverified50% confidenceOpinionExact time
Lossless compression conclusions reported on contaminated benchmarks are meaningless, because the model can directly retrieve answers from parametric memory rather than relying on attention to retrieve from context.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified基准过拟合和数据污染会导致模型通过记忆而非真正理解获得高分，使公开基准分数可信度存疑74% similarUnverified验证集和测试集不应做任何数据增强，一旦施加变换会引入系统性偏差，破坏对模型泛化能力的无偏估计67% similarUnverified在关系抽取中，标签出现的频率几乎不能预测模型在该标签上的表现，真正决定泛化能力的是模型是否见过相似的证据表达方式66% similarUnverifiedRetrieval-Augmented Generation (RAG) cannot completely eliminate hallucinations, as the model may still misinterpret retrieved documents or fill in gaps on its own when retrieval results are insufficient.66% similarUnverified针对特定触发词的后门攻击可让模型在大多数情况下表现正常，仅在遇到特定输入时才输出预设错误信息，使常规评测基准难以捕捉65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/830022API
curl https://kongchang.com/api/v1/knowledge/claims/830022MCP
get_claim(id=830022)