Noeri:AI Agent 智能路由与模型自动优化
Vibecoders, 𝗜’𝗺 𝗯𝗲𝗴𝗴𝗶𝗻𝗴 𝘆𝗼𝘂, stop using the same model for every task.
从具体的开发痛点切入,清晰阐述了“分层调用”的优化逻辑,并自然引出 Noeri 作为自动化解决方案,结构完整且具有实操参考价值。
Vibecoders, 𝗜’𝗺 𝗯𝗲𝗴𝗴𝗶𝗻𝗴 𝘆𝗼𝘂, stop using the same model for every task.
Vibecoders,我求求你们了,别再对每个任务都使用同一个模型。
You don’t need a frontier model to fix a typo. Most of what your agent does all day is easy, and you’re paying 𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿 𝗽𝗿𝗶𝗰𝗲𝘀 for all of it.
修复一个打字错误根本不需要前沿模型。你的代理一天中大部分时间都在处理简单任务,而你却为所有这些操作支付了𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿 𝗽𝗿𝗶𝗰𝗲𝘀(前沿价格)。
→ Use a fast, cheap model for the easy stuff → Escalate when it gets stuck → Save the strongest model for the parts that need real reasoning → Measure how often you actually escalate. It’s probably way less than you think
→ 用快速、便宜的模型处理简单任务 → 遇到卡壳时再升级调用 → 把最强的模型留给真正需要推理的部分 → 统计你实际升级调用的频率。它可能比你想象的要少得多
The catch: nobody has time to keep tuning this 𝗯𝘆 𝗵𝗮𝗻𝗱.
问题在于:没人有时间去手动不断微调这套流程。
That’s why I’m watching Noeri. Routing is just one of the things it improves on its own, along with its instructions, tools and even the models. A person approves every change.
这就是为什么我在关注 Noeri。路由只是它自主改进的功能之一,此外还包括其指令、工具甚至模型本身。每次变更都需要人工批准。
https://x.com/RhysSullivan/status/2115827384108855648/video/1
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力