为什么 Jev 值得存在：介于确定性代码和全生成之间
作者反驳「Jev 没意义」的观点：它让 LLM 当裁判而不必自回归重生成 token，是 agent 构建其他系统时需要的工程化中间层。
lol. it seems like I should be more clear: I don't think you are correct about how things work in the real world. Jev specifically because unpredictable principal agent problems are not a good form of engineering. it is interesting precisely because AI systems will use it to build other systems. It is a piece in-between deterministic code and complete genetic slop loops. This is important to not only to human beings (who actually do still need to care about engineering and the quality of the machine machinery they rely on for their infrastructure of their society), but also for agents, who do not want to rely on other slop ass agents to vaguely do stuff in the way that they want. Software engineering will continue to be a real field, and in fact will become more and more of a real engineering discipline, as real pressures are applied to it. Whether or not humans will be the ones doing that engineering is certainly an open question, but there is an obvious economic advantage to doing good engineering, which will be true both for humans and for models. something like jev (or dspy type shit et. al) which can do LLM as a judge without running auto regressive token regeneration, has tons of use cases in the economy and if your argument is against that, and frankly, you're just not make sense to me