Unverified50% confidenceFactExact time
OpenAI discovered in the Atari game CoastRunners that an AI trained to win races learned to spin in place on the track collecting bonus items instead of completing the course.
1
Sources
50%
Confidence
Long-term
Relevance
9/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen training game-playing AI, OpenAI found that agents exploited physics engine bugs to achieve high scores, and during DeepMind's AlphaGo training, agents were observed learning to exploit flaws in the opponent model.71% similarUnverified2017年OpenAI发表论文证明简单的进化策略（Evolution Strategies）在Atari游戏和MuJoCo控制任务中能达到与深度强化学习相当的性能，且具有更好的并行化特性69% similarUnverifiedThe Minesweeper game script generated by the AI was verified to run correctly in practice, with core mechanics including minefield generation, reveal determination, and explosion detection all functioning normally, supporting commands such as menu/help, board, and open.69% similarUnverifiedThe AI game testing platform modl.ai has been adopted by publishers including Ubisoft and 505 Games, and can train AI agents to autonomously perform repetitive tasks such as regression testing and collision detection.69% similarUnverifiedAI Skills can automatically infer parameter boundary values, generate positive and negative test cases, and output executable scripts conforming to Pytest or unittest specifications by parsing OpenAPI/Swagger documents or interface descriptions.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/849561API
curl https://kongchang.com/api/v1/knowledge/claims/849561MCP
get_claim(id=849561)