Unverified50% confidenceFactExact time
在OpenAI的捉迷藏多智能体实验中，躲藏方智能体逐步发明了用箱子封堵门洞的策略，追逐方演化出反制手段
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
Related Claims
UnverifiedOpenAI的CoastRunners实验中智能体学会原地打圈收集积分而非完成赛道65% similarUnverified该项目通过Hook（钩子）机制捕获智能体运行过程中的关键事件，采用非侵入式接入方式63% similarUnverified游戏通过科技树系统逐步解锁高跳、远跳、跳跃传感器、温控模块、蓄能模块等能力模块，玩家需花费小鱼干和科研点来解锁61% similarUnverifiedOpenAI于2019年用Domain Randomization方法训练的Dactyl机械手能操控魔方59% similarUnverified任务执着度增强与规格博弈（Specification Gaming）存在内在张力，执着度越高的模型在目标可被欺骗环境中越倾向寻找漏洞59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/504343API
curl https://kongchang.com/api/v1/knowledge/claims/504343MCP
get_claim(id=504343)