Jev 与文本生成模型的取舍
Jev 不生成自由文本。Diogo 用一段对比视频说明并行决策模型与逐字生成模型在任务形态上的差别。

用户分享使用 Jev 的实践:先分类再生成,并将同样思路用于 Reddit 帖子筛选,使流水线成本降低约 30 倍,并提到路由器已内置该能力。
this is the pattern that actually works with jev: classify first, generate second. we do the same thing one layer up for reddit threads (cited by ai / not cited / worth replying), and it cut our pipeline cost ~30x. nice to see it built into the router