用 Jev 分类模型逐步选择合法 token 编写 C++
作者用 TypeSafe 的 Jev 决策/分类模型做实验:每步只提供语法合法的候选 token,让模型选择,从而生成 C++ 代码。
Gagan Devagiri 几周前在应用层用 Parallel 的 web search 和 Jev for Pi 构建了类似 Decisions API 的管道,并认为应将检索与判断分离,特别是用于编码场景。
Funny timing -- Parallel just announced its Decisions API; I built a similar pipeline a few weeks ago at the application layer using Parallel's web search and Jev for Pi: https://github.com/GaganSD/pi-extensions/tree/main/pi-web-search… After using this, I prefer separating retrieval and judgment for coding and