跨国深谈← 返回←Pete Florence

① 他说过的话

7 条

那个模型只会用单一一个小时的机器人数据来训练。
That model would only ever be trained on one single hour of robot data.
—— Pete Florence · [01:30]

指向原始笔记的链接

虽然说出来很简单，我不知道在当时，也许甚至到现在，还有哪个任何人展示过的通用模型，能在几个不同的任务上达到 99% 以上的成功率。
As simple as it is to state, I don’t know of at the time, and maybe even still, another like general model that anybody had shown could achieve like 99% plus success rates on several different tasks.
—— Pete Florence · [10:36]

指向原始笔记的链接

所有这些能力，包括单样本和少样本学习能力，都不是被明确训练出来的。
All of these capabilities, including the one-shot and few-shot warning capability, were not explicitly trained for.
—— Pete Florence · [14:54]

指向原始笔记的链接

你知道的，我认为「涌现」这个词完全就是被滥用得不成样子了，因为人们觉得，哦，如果我说它是涌现的，人们就会兴奋。
You know, I think that the phrase emergent is completely like it just gets clobbered over the head and overused because people think like, oh, you know, people are going to be excited if I say it’s emergent.
—— Pete Florence · [15:22]

指向原始笔记的链接

它的预训练中没有任何机器人数据。
had zero robot data in its pre-training.
—— Pete Florence · [19:50]

指向原始笔记的链接

你真正想要的是一个模型，是的，它能做五指手，但它什么都能做。
What you really want is you really want a model that, yes, it can do five-finger hands, but it can do anything.
—— Pete Florence · [21:32]

指向原始笔记的链接

重要的不只是把你的模型对齐到某个品牌或模型类型，而是在机器学习的发展历程中我们一遍又一遍学到的是，一切都是可能的，而且一切都非常简单。
What matters isn’t just aligning your model to a certain brand of model type, but rather the thing that we always learn over and over again over the arc of machine learning is that everything is possible and everything is very simple.
—— Pete Florence · [26:09]

指向原始笔记的链接

② 出现在这些集

1 集

• 《机器人捡起香蕉当扫帚：Generalist 的物理世界通用智能》 — 作为主持

③ 他谈到的

点进去有真内容 —— 本页主要出口

通才 · Gem 1.5 · 缩放定律 · 少样本学习 · 跨载体 · 手（末端执行器） · VLA · 世界模型 · 物理泛化 · 精通

④ 也在聊「智能体」的人