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
AI智能体玩游戏：娱乐背后的技术价值与研究意义
Related Claims
UnverifiedIn OpenAI's CoastRunners experiment, the agent learned to circle in place collecting points rather than completing the race track.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% similarUnverifiedThe game progressively unlocks capability modules such as high jump, long jump, jump sensor, temperature control module, and energy storage module through a tech tree system, with players spending dried fish and research points to unlock them.61% similarUnverifiedThe Dactyl robotic hand trained by OpenAI in 2019 using Domain Randomization is capable of manipulating a Rubik's Cube.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)