端到端训练 (end to end)
集里怎么说它
- 《Parag Agrawal:当 AI 成为网络的第二个用户》(17:46起):本集以 Twitter 推荐系统和 Waymo 为例：系统起初是拼凑的模型加启发式规则，随理解加深越来越端到端训练；今天可以一开始就在规模下更端到端地做。
- 《Waymo 谈物理 AI 的七条实战教训》(26:15起):本集将端到端模型描述为从传感器到控制指令的直接映射，最容易构建且前期进步快，但在安全关键环境达到超人类性能时基本版端到端不够，需要用结构增强来引导规模
- 《当无人机变成基础设施：给机队下指令的人》(15:21起):本集说端到端是长期愿景但物理系统上可观测性和可靠性保证很难,应判断流程中哪些环节该换成世界模型表示,凡是想写 if 分支的地方就该交给智能体
① 提到它的金句
10 条
实际上让我兴奋的是，人们在使用智能体和模型的过程中，拓展自己本来能做的事情的范围，或者对事情承担更多端到端的责任。
指向原始笔记的链接
I actually think the what excites me is people like broadening what they could have done, or taking more end to end ownership of things by using agents and models.
—— 嘉宾 · [17:40]
但现在我们谈的是 Claygent 上数十亿次的运行，以及让 Sculptor 为你做端到端任务，或者说这些运行时间非常长的任务，评估就变成了不可协商的。
指向原始笔记的链接
But now that we’re talking about billions of runs on Claygent and having Sculptor do end-to-end tasks for you or these really long-running tasks, evals became non-negotiable.
—— Vishu · [02:11]
他们端到端地完成任务,卡住时会回来找你,而且他们运作起来就像同事一样。
指向原始笔记的链接
They finish jobs end to end, they come back when they’re stuck, and they function just like colleagues.
—— Roshan · [06:28]
智能体可以帮你深挖这些问题，帮你找到可能驱动客户那些感知的指标，但通常不太擅长真正理解这项工作的端到端上下文。
指向原始笔记的链接
And so, you know, agents can help you dig into those and can help you find um uh find the uh the the metrics that might be driving those perceptions from customers or driving those anecdotes, um, but generally aren’t as good at really understanding the end-to-end context of the work.
—— Marc Brooker · [19:28]
归根结底,你可以拥有一个无限聪明的智能体,但你从这个能端到端做越来越多事情的无限聪明的智能体中真正获得价值的能力,其实是给它正确的任务去做。
指向原始笔记的链接
In the end, you could have an infinitely smart agent, but your ability to actually get value out of this infinitely smart agent that’s able to do more and more stuff end to end is actually giving it the right things to do.
—— Jerry Liu · [05:38]
人人都在产出这些端到端由智能体完成的 PR,不是人类加 AI,而纯粹是 AI。
指向原始笔记的链接
Everyone’s producing these PRs that are end-to-end agentic, not human plus AI, but literally AI.
—— Daksh Gupta · [05:16]
它真正的核心在于：传统的在位者受限于他们的记录系统和他们所拥有的那份记录，他们没有完成端到端的工作。
指向原始笔记的链接
And it really centers around the fact that the legacy incumbent is limited to their system of record and that record that they have, and they’re not completing the end-to-end job.
—— Seema Amble · [03:32]
然而，AI 原生创业公司的机会在于说：我们要拥有整条端到端的链路。
指向原始笔记的链接
However, the opportunity for the AI native startup is to say, we’re going to own that entire end-to-end arc.
—— Seema Amble · [04:16]
所以，在部署这个构建的第一周内，我们就把端到端解决率从 10% 提升到了 80%，而且用的是相当朴素的实现。
指向原始笔记的链接
So we went from 10% to 80% end-to-end resolution rate in just the first week of deploying this build with a pretty naive implementation.
—— Matt Lawler · [08:09]
每一层可能会有 10%、20%、2 倍的差距。你把这些优化机会相乘起来，突然之间你就得到了一个很大的端到端机会，不管你想怎么表述，比如每瓦智能或每瓦效益。
指向原始笔记的链接
And so there might be 10%, 20%, 2x across each of these layers. You start multiplying those optimization opportunities through, and all of a sudden you’re left with a big end-to-end opportunity in terms of whatever you want to say, intelligence per watt or good per watt.
—— Amin Vahdat · [24:54]
② 出现在这些集
3 集
- 《Parag Agrawal:当 AI 成为网络的第二个用户》 — 作为概念(提及)
- 《Waymo 谈物理 AI 的七条实战教训》 — 作为概念
- 《当无人机变成基础设施：给机队下指令的人》 — 作为概念
③ 关联
点进去有真内容 —— 本页主要出口
智能体 · 世界模型 · Parallel · Dmitri Dolgov · Suchet Bargoti · token 最大化 · Waymo · Skydea · 开放网络 · Waymo driver