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。

作者阅读 Jev 相关内容后指出,Jev 接受输入和固定选项并直接返回概率,引发对是否每个 AI 决策都需要 LLM 的思考。
Recently I was reading about Jev, and one idea stood out. We use LLMs for generation, but many AI systems just need to make a decision. Give Jev an input + fixed choices, and it returns probabilities. It got me thinking: Do we really need an LLM for every AI decision?