James Long 测试 Jev 从银行原始数据提取收款人描述
James Long 用 Jev 解决个人财务工具中最棘手的问题:从原始银行数据中生成良好的收款人描述,首次尝试即达到约 95% 的完成度,并指出还有多种优化提问方式。
8个智能体就一笔DOGE交易达成一致,但一个过时的数据源使所有判断失效。模拟中交易台将可执行金额从1840美元降至429美元,JEV式路由器在数据源更新后仍返回拒绝。
8 agents agreed on one DOGE trade. One stale feed made all eight opinions worthless. In this paper simulation, the desk proposed $1,840, reduced the executable amount to $429 and sent the combined state into a JEV-style router. The router returned REJECT after the feed