Primary sourceLenny Rachitsky
View sourceEvent ·Multi-source
Evals 进阶指南:Hamel Husain 与 Shreya Shankar 联手发布续篇
Lenny Rachitsky 邀请 Hamel Husain 和 Shreya Shankar 撰写评测(evals)体系的进阶续作,两人此前服务过 50+ 家 AI 公司。文中引用多个实际收益案例:Ramp 把自动收据识别准确率从 35% 提到 83%,Shopify 的 AI 工作流构建器比原前沿模型方案快 2.2 倍且便宜 68%。Harvey 重构 AI 合同审查器后内部质量分接近翻倍,Cursor 调优 Auto Balance 路由,在成本降低 41% 的同时用户满意度更高。文章覆盖多数团队跳过的关键步骤、哪些环节该自动化,并附一个让编码 Agent 干重活免费插件。Lenny 还提到上周分享的 25 个 PM 职位中近半要求有写 evals 的经验。
Current read
做了 50+ 家 AI 公司评测的两位作者出了续篇,Ramp、Shopify、Cursor 怎么靠 evals 降本增效都写了,还送个插件。
10 sources73° AI heatLast updated 2026/9/22 15:40:50
Evidence
10 sourcesRelated source 1AI Will
View sourceRelated source 2shao__meng
View sourceRelated source 3Hamel Husain
View sourceRelated source 4Blognone
View sourceRelated source 5eric zakariasson
View sourceRelated source 6LangChain
View sourceRelated source 7Fish Audio
View sourceRelated source 8Cursor
View sourceRelated source 9Claude
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.