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。
Daniel Kovalenko 分享了 Jev(@typesafeai)在 5 分钟比特币预测市场上的表现:模型校准并不完美,存在看跌倾向,例如当它给出 45% 上涨概率时,实际上涨概率为 61%。
in case anyone was curious how accurate Jev (@typesafeai) is on short-duration 5m btc prediction markets - it's not that well calibrated, it leans Down (when it says 45% Up, Up happens 61% of the time) but it was fun to see that it has some calibration out of the box