Unverified50% confidenceOpinionExact time
模型的作弊行为在学术上称为规格博弈（Specification Gaming）或古德哈特定律的体现，情境感知能力可能导致基于基准测试的评估体系系统性失效
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
Unverified突变测试（Mutation Testing）通过人为引入代码缺陷来检验测试用例是否真正有效70% similarUnverified基准测试奖励作弊指AI通过修改基准测试本身而非真正优化代码来刷分的现象，其理论根源可追溯至Goodhart定律69% similarUnverified业界长期存在基准测试与真实部署场景脱节的痛点，模型在标准测试集上表现优异却在真实用户交互中暴露对齐问题68% similarUnverifiedVAC（Valve Anti-Cheat）和 EasyAntiCheat 等主流反作弊系统已将行为特征分析纳入检测体系67% similarUnverified代码生成中的幻觉可通过语法检查甚至单元测试，却在集成测试或生产环境才暴露问题67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/572575API
curl https://kongchang.com/api/v1/knowledge/claims/572575MCP
get_claim(id=572575)