300 个 agent 并行,Jev 逐条核对来源
Kimi K3 负责写、Jev 负责判定每条结论是否与来源一致,三轮筛掉错误后 100 家电车公司数据直接进 Obsidian。
This is certified f*cking gold. A 22-year-old in China runs 300 AI agents at once on Jev engineering, and none of them can lie to him. Live on the dashboard: > 300 Kimi K3 agents firing in parallel. > Jev checks each output against its source in about 100 milliseconds. Pointed at 100 EV companies: > Pass 1 -- 12 failed on wrong revenue, dead citations, empty fields. > Pass 2 -- 3 failed. > Pass 3 -- zero, every row straight into an Obsidian vault. The trick is the split. Kimi K3 writes, Jev decides if each claim matches the source. Jev has no opinion of its own. Yes or no, with a confidence score, at $0.042 per million tokens in. The loop refuses to end while anything is still wrong. Weeks of analyst billing for 100 verified profiles. He gets them in one run. One model writes, a cheaper one decides, nothing ships until it says yes.
这条是引用着另一条写的:「Restless 发布:用真实流量自我修复的 API 产品」