Jev 正式发布:面向决策的 System One 模型
TypeSafe 创始人 Diogo Almeida 介绍 Jev、RLCD 训练方法,以及面向软件决策场景的模型定位。原帖附有 2 分 56 秒发布视频。

TypeSafe 推出 JEV 模型,不生成文本只做决策。作者在 80 个真实路由决策上测试,top-1 准确率 98.3%,置信度≥0.9 时达 100%,总成本 3 美分,深度阅读占比 25% 且模型自行标记。
TypeSafe just launched JEV: an AI model that never writes text, only decides. I ran it on 80 real routing decisions: 98.3% top-1, 100% at confidence ≥0.9, total cost 3 cents. Deep reading? 25%, and it flagged that itself. Write-up: https://linkedin.com/posts/activity-7506291059535126529-DM0_… #TypeSafe #JEV #AI