用 Jev 给交易 bot 补上「语境判断」这一环
作者实测 Jev 后认为它适合做量化风控难以覆盖的持仓语境判断，并设想让社区 agent 通过 Noesis 的 Jev API 询问某个 meme 币是否还该持有。
i’ve been experimenting with jev today. it reminds me a little of diffusion models: looking at the whole output instead of reasoning through it one word at a time. the trade-offs are fairly obvious. inputs and outputs are constrained, and local hallucinations can happen. JEV claims to be hallucination-free, which would be very useful if it holds up. what matters most to me in practice is speed. especially for trading. i saw @noesis_network get stopped out of $SHROOM again. it currently uses a traditional risk system because it needs to react faster and more cheaply than an LLM. the problem is that it only sees numbers, not the context around a position. JEV seems almost made for this gap. it can be used for the parts of position monitoring that are difficult to quantify. that’s quite exciting, because it actually expands what noesis can do. before, noesis used theses from community agents to make its own investment decisions. now those agents could also use noesis jev api to ask whether a meme should still be held, and why. noesis keeps developing by absorbing new data, test results and new AI capabilities like jev. idk anyone could have written its roadmap in advance.
这条是引用着另一条写的：「ChatGPT 联合发明人发布新模型 Jev」