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。

作者认为 JEV(System-1)将带来重大改变,当前 LLM 模拟的是 System-2,而 JEV 的实现是近期最新颖的概念之一,看似只是动态分类器却令人意外。
Models like JEV (System-1) will, in my opinion, change a great deal. Current LLM models simulate System-2. For me, the implementation of JEV is one of the freshest concepts of the recent months. It seems like nothing particularly special (dynamic classifier) yet it's surprising