Jev 在邮件分类基准测试中不敌 Gemini,但仍考虑投产
作者在 1,565 封德语和英语工业供应商商务邮件的 10 类分类基准上测试了 Jev,发现其准确率不及 Gemini,但更关注错误出现的模式,并仍有意将其投入生产。
Twenty 表示,Jev 将 AI 从聊天窗口扩展到整个核心流程,邮件同步、工作流条件等原本靠手工规则的环节现在也能利用 AI。
Jev moves AI from the chat window into the core of Twenty. Until now, AI in Twenty meant the chat and agents. Everything automatic, like email sync or workflow conditions, still follows rules we wrote by hand, because an LLM on every synced email is too slow and too expensive.