Unverified50% confidenceFactExact time
OpenAI研究员在2021年发表论文指出过度优化奖励信号的风险，被称为Goodhart定律在AI对齐中的体现
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
AI助手"表演性顺从"：当Claude学会敷衍而非真正解决问题
hackernewshackernews7/5/2026
Related Claims
Verified基准过拟合是Goodhart定律在AI评估领域的体现，即衡量指标成为优化目标时便失去衡量有效性69% similarUnverifiedAI trading assistants, based on statistical patterns in historical data, tend to systematically underestimate tail risks when confronted with structural breaks and extreme black swan events — a fundamental constraint of statistical learning methods that cannot be eliminated through engineering means.67% similarUnverifiedIn mature AI projects, swapping in an uncertain new model carries more risk than benefit, and stability and controllability are more important than using the latest model.66% similarUnverifiedAI对齐研究中，模型倾向回避高度不确定性的开放问题的现象被称为奖励模型过拟合或规范游戏66% similarUnverifiedWith AI support, researchers can attempt high-risk, high-reward research questions that were previously daunting due to the high cost of validation.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/254001API
curl https://kongchang.com/api/v1/knowledge/claims/254001MCP
get_claim(id=254001)