Fei-Fei Li's Startup Raises $230 Million in Funding

Bloomberg Technology
13 Sept 202408:46

Summary

TLDR在这段访谈中,Kahlil Lee博士讨论了他在学术界和私营部门筹集资金的经历,以及他如何将空间智能的概念从学术研究转化为实际应用。他解释了空间智能的含义,即人类和动物在三维世界中理解、推理和互动的能力,并强调了将这种智能引入计算机的挑战。Lee博士还讨论了空间智能在多个领域的潜在应用,包括艺术创作、设计、建筑、机器人技术和制造业,并强调了公共部门和私营部门在推动此类技术发展中的重要性。

Takeaways

  • 💡 李飞飞博士强调了空间智能的重要性,这是人类和动物都具备的一种古老能力,涉及理解、推理、生成和与3D世界互动。
  • 🌟 李飞飞博士提到,尽管她在学术界和私营部门都有经验,但将空间智能变为现实是一个极具挑战性的问题。
  • 🚀 她认为空间智能是人工智能领域的下一个前沿,需要将2D图像识别的进展扩展到3D世界。
  • 🤖 空间智能的应用非常广泛,包括创意产业、设计、开发、建筑、机器人技术、制造业以及增强现实和虚拟现实。
  • 💼 李飞飞博士解释了为什么空间计算需要空间智能,以及这对许多其他用例的重要性。
  • 💸 她讨论了为什么需要私人部门的资金来推动空间智能的发展,包括生态系统中的基础研究和工业驱动的专注。
  • 🏆 李飞飞博士提到了ImageNet项目,这是她在计算机视觉领域的早期工作,她认为空间智能是她职业生涯的延续。
  • 🌐 她强调了公共部门和私营部门在推动技术发展中的重要性,以及她对公共部门资金的持续需求。
  • 🤝 李飞飞博士提到了一些AI领域的知名人士,如Jeff Dean和Geoffrey Hinton,他们对空间智能项目的投资和支持。
  • 🎓 她呼吁增加对学术界和基础科学研究的投资,以支持学生、教职员工和研究人员进行基础科学研究。

Q & A

  • 在学术界筹集资金和在私营领域筹集资金有什么不同?

    -在学术界筹集资金通常是为了支持基础研究和好奇心驱动的研究,而在私营领域筹集资金则需要展示技术的实际应用和商业潜力。

  • 什么是空间智能,它与人类智能有什么关联?

    -空间智能是指理解、推理、生成和在三维世界中互动的能力,这是人类和动物经过数百万年进化获得的能力。

  • 空间智能在计算机视觉领域中扮演什么角色?

    -空间智能是计算机视觉领域中的下一个前沿,它涉及到将图像和视觉信息转化为三维理解,这对于真实世界的理解和交互至关重要。

  • 为什么将空间智能应用到三维世界中是一个挑战?

    -因为现实世界是三维的,而人类的空间智能是基于理解和操作三维的天然能力,将这种能力转化为计算机系统是一个技术挑战。

  • 空间智能在现实世界中的应用有哪些?

    -空间智能的应用非常广泛,包括机器人技术、制造业、增强现实(AR)、虚拟现实(VR)以及任何需要与现实世界交互的场景。

  • 为什么空间计算需要空间智能?

    -空间计算需要处理三维空间中的数据和交互,空间智能提供了理解和操作这些数据的能力。

  • 为什么私营部门对空间智能的研究很重要?

    -私营部门可以提供资金、资源和市场驱动的研究方向,有助于将学术研究成果转化为实际应用。

  • 为什么需要从风险投资那里获得资金?

    -风险投资提供了必要的资金来支持创新技术的研发,并且能够吸引那些相信并愿意解决难题的人才。

  • ImageNet对空间智能的发展有何影响?

    -ImageNet是计算机视觉领域的一个里程碑,它推动了对象识别和图像处理技术的发展,为空间智能的研究奠定了基础。

  • 为什么说空间智能是人工智能的下一个‘北极星’问题?

    -因为空间智能是理解现实世界的关键,它可能会像ImageNet一样,推动人工智能领域的重大进步。

  • 为什么认为公共部门也需要资金来支持研究和开发?

    -公共部门的研究是创新的基础,它为私营部门提供了原始的技术和理论,两者相辅相成,共同推动技术进步。

Outlines

00:00

💼 从学术界到私企融资

在第一段中,Kahlil 被问及从学术界进入私企并筹集资金的经历。他提到,尽管之前在私企工作过,但筹集资金并非易事。他强调了空间智能的重要性,并表示对于能够聚集一群杰出人才来解决这一难题感到兴奋。Kahlil 解释了空间智能的概念,即人类和动物理解、推理、生成和在三维世界中互动的能力。他提到,尽管人工智能在语言处理方面取得了巨大进步,但将这些进展应用到三维空间中是一个更大的挑战。Kahlil 还讨论了空间智能在机器人、制造业和增强现实等领域的潜在应用,并强调了私企在推动这一领域发展中的作用。

05:00

🌐 空间智能:AI的新北星

在第二段中,Kahlil 讨论了空间智能与他之前在 ImageNet 项目中的工作之间的联系。他认为空间智能是计算机视觉领域的下一个重要挑战,并且是他在 AI 领域的下一个'北星'问题。他提到,尽管 ImageNet 项目在二维空间中取得了成功,但空间智能将引领 AI 进入三维空间的新篇章。Kahlil 强调了公共和私有部门在推动 AI 研究中的重要性,并呼吁增加对学术界的投资,以便学生和研究人员能够继续进行基础科学研究。他还提到了一些知名的 AI 专家和风险投资公司对 World Labs 的投资,这表明他们对解决这一难题的信心。

Mindmap

Keywords

💡空间智能

空间智能指的是理解和推理三维空间的能力。在视频中,提到人类和动物都具有这种能力,例如理解空间关系、导航和操纵物体。视频中强调,将这种能力赋予计算机是当前研究的重点,因为它对于理解现实世界的三维性质至关重要。

💡三维

三维是指具有长度、宽度和高度的物理空间。视频中提到,现实世界是三维的,人类的许多活动,如建造城市或与环境互动,都依赖于对三维空间的理解。计算机视觉和空间智能的目标之一是使计算机能够理解和操作三维空间。

💡人工智能

人工智能是指使计算机系统模拟人类智能的技术。视频中讨论了人工智能在过去十年中的进步,特别是在语言理解和图像识别方面。同时,也提到了空间智能是人工智能领域的下一个前沿。

💡计算

在视频中,计算指的是使用计算机处理数据和解决问题的能力。特别提到了GPU(图形处理单元)在进行复杂计算,如深度学习模型训练时的重要性。

💡投资

投资在视频中指的是为研究和开发提供资金。讨论了从学术界到私营部门的资金筹集过程,以及如何通过吸引投资者来支持解决复杂问题的工作。

💡研究与开发

研究与开发(R&D)是指创造新知识或改进现有知识以开发新产品或流程的活动。视频中强调了公共和私营部门在R&D中的投资对于推动技术进步的重要性。

💡公共部门

公共部门通常指的是政府机构。在视频中,提到了公共部门在支持基础研究和教育方面的作用,以及它在推动技术发展中的重要性。

💡私营部门

私营部门指的是由个人或公司拥有和运营的企业。视频中讨论了私营部门在推动技术创新和商业化方面的作用,以及它如何与公共部门合作以支持研究。

💡机器学习

机器学习是人工智能的一个分支,它使计算机能够从数据中学习并改进其性能。视频中提到了机器学习在图像识别和空间智能中的应用。

💡虚拟现实(VR)和增强现实(AR)

虚拟现实和增强现实是利用计算机技术模拟环境的技术,为用户提供沉浸式体验。在视频中,提到了空间智能对于发展VR和AR技术的重要性,因为这些技术需要对三维空间有深刻的理解。

💡机器人学

机器人学是设计、构建和操作机器人的科学。视频中提到机器人学作为空间智能应用的一个领域,因为机器人需要理解并与其三维环境互动。

Highlights

进入私营领域筹集资金的挑战

空间智能的概念及其重要性

空间智能与人类及动物的进化

计算机视觉和空间智能的进展

3D世界对空间智能的重要性

空间智能在现实世界中的应用

空间计算与空间智能的关系

为什么空间计算需要风险投资

学术界与产业界的合作生态

ImageNet对空间智能研究的影响

空间智能作为AI的新北星问题

如何向AI领域的大人物筹集资金

公共部门与私营部门在研发中的资金需求

对公共部门投资的必要性

私营部门对资源的访问

对学术界基础研究的支持

Transcripts

play00:00

What was it like raising funds as an academic now entering this private world

play00:04

to be raising money? You've been in private well before, but

play00:06

to raise funds. Was it easy from this star studded list?

play00:11

Good morning, Kahlil. It's really good to be here.

play00:14

Well, nothing is easy, but what's really, really hard is make spatial

play00:18

intelligence happen. I'm just so excited that we've brought

play00:22

together a incredible group of pixel talents to work on this really, really

play00:29

hard problem that we now call spatial intelligence.

play00:35

Explain to us what you mean when you say spatial intelligence and what is it that

play00:40

you're building exactly? Yeah.

play00:42

Rachel Well, look, humans have spatial intelligence.

play00:44

It's actually a very ancient ability. We have evolved over millions and

play00:49

millions of years. It's the ability to understand, to

play00:53

reason, to generate and to even interact in a 3D world, whether you're looking at

play01:01

a beautiful flower or trying to touch a butterfly or building a city.

play01:07

All of this is part of the capability of spatial intelligence.

play01:12

We see that with humans and we see that with animals.

play01:15

How do you expect that? We'll see that with computers?

play01:19

Well, that's the problem we're working on.

play01:21

We're already starting to make tremendous progress.

play01:24

The past decade of A.I. has been pretty exhilarating, and people

play01:29

hear a lot about language recently. But truly, in the world of pixels of

play01:34

vision and spatial intelligence, we've been making progress, such as

play01:38

understanding what's in the picture. To be able to tell a story of what's in

play01:43

the picture, to even prompting a sentence and get up to the image out of

play01:49

it. But what's really the next frontier,

play01:51

which is such a hard problem to crack, is to bring all this into 3-D because

play01:56

the real world is 3D and human's spatial intelligence is built upon this very

play02:03

native capability of understanding and working with 3D.

play02:08

So let's put into real world context that

play02:12

working with 3D the applications, is it robotics?

play02:16

Is it manufacturing? Is it just us interacting with the real

play02:20

world when we put our AI function glasses on?

play02:24

You're Laurel and Carolyn. Those is such a foundational technology

play02:28

and such foundational ability for for computers that it has implications in a

play02:36

wide range of use cases. To start with, right creators.

play02:40

Creators includes not only artists and VFX creators, but also designers,

play02:48

developers, builders. And this technology has a profound

play02:53

implication for them. But in the long arc of this, of course,

play02:58

robotics manufacturing, a AR, VR. This is you know, there's a reason that

play03:05

Apple calls their vision pro spatial computing.

play03:08

Well, in my opinion, spatial computing needs spatial intelligence.

play03:12

So does many other use cases. Why does spatial computing need VC

play03:18

money? Private sector action.

play03:21

What couldn't you achieve in academia? So this is a whole ecosystem.

play03:28

We've been seeing this in air. We've been seeing this for years now.

play03:33

And this dates back to any technology that our society, our country is

play03:38

building. Right?

play03:39

This ecosystem needs upstream, fundamental, curiosity driven research,

play03:45

which I have spent many years of my life in.

play03:50

But it also needs a very focused drive in industry.

play03:55

We've got great big tech companies working on related problems, but we also

play04:01

what's beautiful about our ecosystem are startups.

play04:05

There has a tall dream, has has the ability to call on the believers who

play04:11

believe in cracking such a hard problem. And we come together and focus all of

play04:17

our energy in solving this truly hard problem that needs to be scaled, needs

play04:23

to be protectionist, and needs to be delivered in the hands of users and

play04:28

customers. Dr.

play04:32

Lee, one of the things that you're most well-known for is image net and big

play04:37

database of millions of images that really help push forward the field of

play04:41

object recognition and images. I'm curious how your work on that played

play04:45

into your decision to start World Labs. How are how do you see those two things

play04:49

as related? Yeah, thanks for asking that question,

play04:51

Rachel. I think they're related in into two ways

play04:55

at least. One is that image, and that is one of

play05:00

the earlier work in the field of computer vision, which is in the pixel

play05:04

space. Granted, in that time, you know, more

play05:09

than ten years ago what image that and the derivative algorithms were able to

play05:14

do are still in the 2D space, right. Recognizing objects in photos and

play05:20

eventually telling stories of pictures. But now this is a an intellectual

play05:25

continuation of the early work in computer vision.

play05:30

Now we're into the next really difficult chapter, which is spatial intelligence.

play05:35

So intellectually, I feel it's my life's work in continuation.

play05:40

And zooming out to one layer image that was my

play05:45

more than 15 years ago. It was my intellectual bet on a big

play05:49

North Star problem, a North Star problem that had really changed the course of

play05:53

A.I.. I do believe spatial intelligence is the

play05:57

next North star for me and for my team, that it will change the course of A.I..

play06:02

One of the things I'm wondering about with the funding for this company is

play06:08

there are a number of big names in A.I. in particular that invested in this.

play06:12

We've got Jeff Dean and Jeffrey Hinton, Andre Carpathia, some of them you've

play06:17

worked with previously. I know several of you were at Google at

play06:20

the same time. How did you pitch them on it?

play06:24

Well, that's the beauty of our field. The first of all, these people have been

play06:29

friends and colleagues for years, or former students.

play06:32

I think they share my belief. I think they see this as such a big

play06:38

problem and they'd be leaving my team when they hear my co-founders, Ben

play06:45

Mildenhall, Christoph Laster, Justin Johnson, and really the entire founding

play06:51

team, they recognize that while this is a tough problem, it needs

play06:56

people who really can have the ability and believe to to crack this.

play07:01

And I think that's why they they supported us.

play07:06

You've got money from big name VCs as well, and he and Horowitz, to name just

play07:12

one. I'm interested more broadly about the

play07:16

rallying call you've had for money to go towards academia.

play07:20

You went to President Biden himself saying that there needs to be more

play07:23

funding in the public sector as well as the private sector when it comes to R&D,

play07:28

ultimately, so universities can access GPU and compute.

play07:32

Are you still feeling that necessity that public sector needs money or have

play07:37

you just sort of given up and gone to the private sector hit

play07:40

a Carlyle? I actually believe even more so now that

play07:45

I'm traversing both the private and the public sector, that seeing the access to

play07:51

compute, access to support in the private sector, I believe none of us

play07:57

will be here in the private sector without the public sector.

play08:00

You know, image that convolutional neural network back propagation

play08:05

transformer models, meaning of this seminal work in AI came from public

play08:13

sector first. So so I think this ecosystem is so

play08:17

critical and missing or or the imbalance of any component is harmful for the

play08:23

ecosystem. And now I have personally experienced to

play08:27

see the access we have to resources. It makes me believe even more that our

play08:32

country needs to invest in our public sector, academia, in this kind of

play08:37

moonshot mentality to support, you know, students and faculty and researchers in

play08:44

basic science research.

Rate This

5.0 / 5 (0 votes)

相关标签
空间智能人工智能机器学习技术创新学术研究私企合作资金筹集计算资源视觉计算3D世界