🍀 Spring Appointment, Lucky Draw Gifts! Growth Value Issue 1️⃣7️⃣ Spring Lucky Draw Carnival Begins!
Seize Spring Luck! 👉 https://www.gate.com/activities/pointprize?now_period=17
🌟 How to Participate?
1️⃣ Enter [Plaza] personal homepage, click the points icon next to your avatar to enter [Community Center]
2️⃣ Complete plaza or hot chat tasks like posting, commenting, liking, and speaking to earn growth value
🎁 Every 300 points can draw once, 10g gold bars, Gate Red Bull gift boxes, VIP experience cards and more great prizes await you!
Details 👉 https://www.gate.com/announcements/article/
Simon Willison(Django 联合创始人,现在专注 AI 工程实践)发了一篇 subagent 工程模式指南,把 LLM 子代理拆成三种模式:Serial(串行探索)、Parallel(并行执行)、Specialist(专家角色)。核心问题是 context 窗口有限,subagent 用独立窗口分担负载。
实际用下来,Parallel 被严重低估。派一批轻量模型并行跑探索,主窗口等结果回来就行,体验和效率差距明显。但最大的反模式不是过度分解,是过度信任 — subagent 返回的结果必须由主代理做交叉验证,不能直接采信。
最大收益不是速度,是 context 保护。主窗口干净,复杂任务才能走完最后一公里。
附 Simon 原文链接: