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。
GTM工程师分享实用策略:让Claude找出关键数据源,再借助Jev模型以极低的成本评估每个来源的质量,并生成工件供人工审查。适用于大规模TAM清单构建。
Underrated GTME tactic when working with agents to map out large TAM lists: Ask Claude to find the key sources where it'll extract the data from. Have JEV score the quality of each at 0.0001 usd. Then have it build you an artifact to review the sources. Specially useful in