集里怎么说它

• 《人形机器人上战场：公关跑得比机器人快》(20:42起):本集把机器人训练数据方法分成三派：仿真、远程遥控和视频；说Foundation偏重视频路线，用第一人称GoPro加第三人称三脚架同时拍，约100小时视频可让单一任务完全自主

① 提到它的金句

18 条

于是你意识到你能看到挥舞双截棍的人形机器人的视频，它们是预先编程的，制作那个视频花费了 1500 万美元。
And so you realize that you can see the video of nunchuck wielding humanoids, which are pre-programmed and that costs $15 million to do that video.
—— Qasar Younis · [10:29]

指向原始笔记的链接

我宁愿每隔几年遇到那样的抵制,但继续进化并保持相关性,也不愿选择另一种方案,那就是我们没有视频,没有私信,没有快拍,没有排名。
i would much rather have backlashes like that every couple years but continue to evolve and continue to stay relevant than the alternative which would have been like we didn’t have video we didn’t have dms we didn’t have stories we didn’t have ranking
—— Adam Mosseri · [58:31]

指向原始笔记的链接

因为你对这件事非常有热情、非常兴奋、想做到最后，它很快就会开始感觉像你在玩电子游戏，试着升到下一级。
Because it’s a thing that you’re so passionate and excited about and you want to get to the end, it quickly starts to feel like you’re playing a video game and like trying to get to the next level.
—— 嘉宾 · [39:28]

指向原始笔记的链接

如果我能生成手指移动的视频,如果我能生成手拿起杯子的视频,为什么我不能让机器人做同样的事情?
If I can generate video of a finger moving, if I could generate video of a hand picking up a glass, why can’t I cause a robot to do the same?
—— Jensen Huang · [35:03]

指向原始笔记的链接

我想了解关于世界的一些事情。而且即使我可以生成任何种类或任何结构的视频，我也没有学到我想知道的关于世界的事情。
I wanted to learn something about the world. And even if I can generate video of any kind or of any structure, I didn’t learn what I wanted about the world.
—— Justin Johnson · [11:12]

指向原始笔记的链接

我们可以拿 20 年前的照片和视频，做一些 20 年前不可能做到的重建。
We can take photos and video from 20 years ago and do reconstructions that were not possible 20 years ago.
—— Brian McClendon · [08:25]

指向原始笔记的链接

电子游戏里的物理效果几乎可以了,但还不足以用来训练机器人。
The physics in video games are almost there, but they’re not good enough to train a robot on it.
—— Rafael Levi · [04:44]

指向原始笔记的链接

比如说,NVIDIA 在训练 Cosmos 机器人时,他们丢弃了大约 96% 的视频。
So for example, NVIDIA, when they train in the Cosmos robot, they’re throwing out about 96% of the video.
—— Rafael Levi · [08:00]

指向原始笔记的链接

我们真的相信,如果我们大规模训练一个视频模型,最后我们会得到一个很棒的模型。
we really believe that if we trained a video model at the large scale, we would get a great model at the end.
—— Anastasis · [08:53]

指向原始笔记的链接

文本生成视频不会是答案。人们想要比那多得多的控制力。
text-to-video was not going to be the answer. People wanted a lot more control than that.
—— Anastasis · [19:05]

指向原始笔记的链接

每一个投入生产的视频生成模型,在底层都用了一条复杂的提示词补全流水线。
every single video generation model in production uses a complex prompt completion pipeline under the hood.
—— Anastasis · [22:39]

指向原始笔记的链接

类似地,Gen2 几乎无法创建连贯的视频,但如果你把它扩展,从某种意义上说没有理由它不会成功。
Similar Gen2 can barely create coherent video, but if you scale it up, there is no reason why it shouldn’t work in a way.
—— Anastasis · [26:20]

指向原始笔记的链接

对我来说,实时视频生成是不可避免的,它的用户体验要好得多。
To me, real-time video generation is just inevitable that it has much better user experience.
—— Anastasis · [41:26]

指向原始笔记的链接

我们意外地创造了一个机器人领域最先进的模型——仅仅通过扩展视频模型。
we accidentally created a state of the art model for robotics. By just scaling video models.
—— Anastasis · [56:05]

指向原始笔记的链接

视频模型与世界模型之间的巨大差距,就在于反事实生成这个理念。
the big gap between video models and world models is that idea of the counterfactual generation.
—— Anastasis · [70:15]

指向原始笔记的链接

Scaling loss 起作用不只是科学概念。视频只是比 LLM 和编程晚了两三年，但很显然它最终在视频领域也会起作用，我就决定赌一把。
It’s not just a concept from science that scaling loss work. Video just takes maybe two, three years longer than LLMs and coding, but it was clear that finally scaling loss should work in video as well, and I just decided just to take a bet.
—— Alex Mashrabov · [10:50]

指向原始笔记的链接

今天思考视频模型的方式，就是把它当作现代渲染引擎。把它想成类似 Unreal Engine 或 Unity，只是输入类型不同。
The way to think about video models today, it’s just modern rendering engine. Think about this as like Unreal Engine or Unity, but just different types of inputs.
—— Alex Mashrabov · [25:16]

指向原始笔记的链接

社交媒体营销中实际发生的情况是，当公司开始每周产出数百个广告创意时，他们想要也许更便宜、更可控的模型，因为做出爆款社交媒体视频并不一定需要博士级别的智能。
What’s actually happening in social media marketing as companies start to print hundreds of ad creatives a week, they want to have maybe cheapest, more steerable models, because like PhD level intelligence is not necessarily needed to make viral social media video.
—— Alex Mashrabov · [28:54]

指向原始笔记的链接

② 出现在这些集

1 集

• 《人形机器人上战场：公关跑得比机器人快》 — 作为概念

③ 关联

点进去有真内容 —— 本页主要出口

Mike LeBlanc · Foundation · 人形机器人 · 执行器 · 视觉-语言-行动模型 · 远程遥控 · 世界行动模型 · Boston Dynamics · Optimus · Physical Intelligence