devwithjevJev vs Random Maze Navigation Test
The author built a maze-navigation test comparing Jev with a pure random-number baseline. Jev received the rules, features, and Manhattan distance to the goal as a heuristic, with instructions to reduce that distance.
karminski-牙医@karminski3𝕏
为什么Jev有些时候不如随机数发生器? 整了个活, 突然想着Jev会不会某些时候还不如纯正态分布决策效果好, 直接用随机数发生器(甚至也可以叫它 System-1)的高 QPS 硬怼？毕竟猴子在打字机前也能敲出莎士比亚. 而且这事在数学上这其实是有正经定理撑腰的, 著名的波利亚随机游走定理(Pólya's Random Walk Theorem),在有限的二维连通网格里, 简单随机游走是常返的(Recurrent), 也就是说, 一个毫无思想的醉汉在迷宫里瞎晃, 以概率 1 最终必定能摸到终点. 神奇吧？理论有了, 于是我写了个测试环境, 拉来 Jev 和纯随机做了一波迷宫寻路对抗测试. 为了不亏待Jev, 我把规则和特征给足, 把曼哈顿距离作为启发式指标塞进输入, 并告诉它尽量减小与终点的距离. 然后用Rust + xoshiro256++ + bit reservoir + Tokio
The author implemented the test using Rust, xoshiro256++, a bit reservoir, and Tokio. The post cites Pólya’s random-walk theorem as motivation for testing a random baseline.
Also filed under Games & real time
- Mapping Children's Behavioral Intent in Games
Mores Research used Jev to map children’s behavioral intent in games and examine how it correlates with real-world behavior. It reports mapping intents across 45+ real sessions involving 20 child users over a long horizon.
- StarCraft Agent for Fast Tactical Decisions
Jev_Star is a StarCraft agent experiment that uses Jev for fast tactical decisions.
- AI-Town Character Action Simulation
JevTown is an AI-town simulation where Jev selects character actions and responses.
- Jev-Powered Drone Swarm Simulation
The project simulates a drone swarm using Jev for reflex decisions and collision avoidance.