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。

Jev 接收一行文本,回答九个封闭问题(时代、发光或反光、颜色、字体、磨损、密度、情绪、布局、场景),再通过第二次调用对十五个参数进行评分,全程无需 LLM。
Jev composes and manipulates UI in real time one line of text goes in. Jev answers nine closed questions: era, glowing or reflective, colour, typeface, wear, density, mood, layout, scene. a second call scores fifteen dials against those answers. no LLM in the loop @typesafeai