维基百科链接竞速:高选项数决策演示
让模型在数百到数千个链接中选择下一步,用游戏方式展示决策速度与约束输出的效果。原帖附有 4 分 04 秒视频。

作者用 Jev 做了一个 Karpo 实验,从 Google Maps 历史学习偏好,对 1,253 个纽约地点进行个性化排序。
I gave Jev 1,253 NYC places and one job: figure out which ones you’d actually love 👀 Made a little Karpo experiment that learns from your Google Maps history, then ranks the entire Google Maps list based on your taste. Jev from @typesafeai is kind of perfect for this: