Jev 驱动的实时 Doom 演示
发布线程展示游戏场景中的实时决策效果。Diogo 说明演示约每秒调用 10 次 Jev,并附有 2 分 43 秒视频。

作者构建了一个与 Jev 对弈国际象棋的 playground,每个局面把棋盘作为状态、所有合法走法作为 choice() 输入,并评价结果远未达到 AGI。
LLM's are absolutely terrible at chess. I wondered if Jev could do better so I built a playground to play against it. Every position, it gets the board as state, and all legal moves as a choice(). The result? Not quite AGI.