开发者Keshav Saharia在Hacker News上分享了一个升级版本的Java程序，用于在益智游戏Drop7中寻找最优策略1。该程序最初编写于16年前，近期进行了重新研究和改进1。
Drop7是一款涉及搜索、隐藏信息、长期规划和随机性的复杂益智游戏，每回合只有七种选择1。当前最佳自动化策略的平均得分在30万至40万分之间，而经验丰富的人类玩家通过长期战略思维可以突破100万分1。为了推进研究，该项目邀请社区通过全球排行榜竞争并提交人类策略1，同时也是关于使用AI代理作为研究合作者的一次实验1。感兴趣的开发者和玩家可以访问https://drop7.dev/compete 参与排行榜竞争，或在https://github.com/keshavsaharia/drop7-bench 查看代码仓库1。
A developer has unveiled an upgraded version of a Java program originally written 16 years ago to search for optimal strategies in Drop7, a puzzle game that combines search, hidden information, long-term planning, and randomness 1. The current best automated strategy achieves an average score between 300,000 and 400,000 points, while experienced human players can reach over 1 million points through deliberate long-term thinking 1.
The game presents players with only seven choices per turn, making it a tractable yet complex domain for strategy optimization 1. The project invites the community to compete on a global leaderboard and submit human strategies, while also serving as an experiment in using AI agents as research collaborators 1. Interested participants can access the global leaderboard at https://drop7.dev/compete and view the code repository at https://github.com/keshavsaharia/drop7-bench 1.
评论
还没有评论，欢迎留下第一条。