---
area: "work-career"
category: ai-ml
companies_orgs:
- Meta
date: '2025-12-16'
draft: true
guest: ''
insight: ''
layout: post.njk
project: []
series: ''
source: https://www.youtube.com/watch?v=5qfwNBs3IVE
speaker: EO
status: evergreen
summary: Mike Schroepfer，Meta前CTO，分享了他关于科技创新、气候变化和创业的见解。他讨论了如何识别和投资于具有突破性潜力的技术，以及如何通过创业和创新解决气候变化问题。
tags:
- investing
- tech
title: 高保真双语档案：Mike Schroepfer - 下一代科技投资与创新
---
### 高保真双语档案：Mike Schroepfer - 下一代科技投资与创新

我叫 Mike Schroepfer。
我花了 25 年时间建立、启动和扩展科技公司。在小公司工作过，也在互联网泡沫期间参与过两家小公司。之后，在互联网泡沫破裂后，我创办了自己的公司，并将其卖给了一家更大的公司 - 微软系统公司。然后加入了 Mozilla，负责 Firefox 浏览器的开发，那时版本号还很小。2008 年，我加入了一个叫做 Facebook 的小社交网络，当时它比 MySpace 还小。之后，在接下来的 14-15 年里，我领导了工程团队，建设了数据中心，进入了消费级硬件领域，开发了虚拟现实头盔，管理了多起数十亿美元的收购，建设了 AI 研究实验室，开发了硬件、软件、企业和消费级产品，将团队规模从 100 人扩展到大约 35,000 人。

如今，我通过创办一家名为 Gigascale Capital 的风险投资公司，帮助伟大的企业家们。我们的目标是寻找那些有远大理想、利用技术进步创造更好产品、解决气候和环境问题、同时又能成为优质商业机会的创始人。

一个简单的例子是电动汽车。它们不产生污染，运营和维护成本更低，速度更快，噪音更小。传统上，电动汽车价格更高。但随着电池变得越来越便宜，它们在成本上越来越具有竞争力。我们正在寻找那些具有成本竞争力、人们喜爱、同时对人类健康和地球有益的产品。

这是在 2008 年 8 月或 9 月。我有一个大家都喜欢的产品 - facebook.com。每天都有更多人注册，新的功能不断被添加进来。当时，最紧迫的问题是扩展。坦白地说，这个问题非常棘手 - 我们需要不断确保网站运行顺畅，而支持这种产品的软件和硬件基础架构当时并不存在。

所以，我在那里度过的前几年，大部分时间都在扩展和重建网站的软件架构，然后建设支持这一切的硬件基础设施。当我刚到那里时，我们正在数据中心释放空间并安装服务器。但由于 2008 年的金融危机和房地产危机，人们停止建设大型数据中心，所以我们无法获得更多空间。因此，我们不得不自己建设，这带来了很多挑战，我们必须学习很多新东西。

没有一件事情是一开始就完美的。我们犯了一些错误，不得不修复它们，并且有足够的谦逊意识到，我们需要学习很多新东西。所以，首先要做的就是招聘那些有数据中心工作经验、网络设计经验和相关领域实际背景的人。

像公司里的许多问题一样，如果你是一位创始人或潜在创始人，正在关注这篇文章，那么大多数初创公司做的事情都是出于完全的必要性。就像，我们得不到空间，我们没有其他选择，我们必须解决这个问题。我从中总结出的一条经验是，没有办法逃避难题。你必须直接面对它。

所以，就像，如果我们需要解决这个问题，那么我们就必须想办法解决。关键风险是什么？我们最不了解或技术风险最高的事情是什么？让我们先解决那些问题。我认为，有一种人类的本能，喜欢解决容易的问题。所以，就像，我有这个任务，我想做，但我的房间有点乱。我要去打扫我的整个办公室。我不想做困难的事情。

就像，这是一个非常人性的特质。所以，重要的是要克服它。就像，实际上，不，我的办公室很乱，但如果我不能完成这个提案并为公司筹集资金，那么其他事情都不重要了。我的办公室是否干净并不重要。

早期的 Facebook，事情总是像打补丁一样。我们的问题并没有完全解决，但也不是完全消耗我们的所有精力，以至于我们没有一点精力展望未来。2013 年，Facebook 开始了 AI 研究。

当时，我们面临的一个问题是，我们是否应该建立一个更广泛的研究实验室？当时有像微软研究院、IBM 研究院或其他研究院，它们可以做软件理论、各种技术领域的研究。最令人信服的是，AI 是一个非常重要的事情，它将产生巨大的影响，你不会想花时间研究其他事情。你会想集中所有精力投入 AI。

这是马克和我曾经讨论过的问题，我认为他是那个推动建立 AI 实验室的人。部分原因是他对生态系统的可见性。那时，图像识别的 ImageNet 挑战是一个著名的学术挑战，没有人太关注它，直到第一个神经网络进入挑战并变得比其他所有方法都好得多。

这是科技领域中很少见的时刻，某些东西出现了，就像，“哦我的天啊，这东西太棒了，它比其他任何东西都好 10%，这是一个非常大的差距，相比之下其他差距都很小，它使用了神经网络在数据上进行训练。”所以，这是一种不同的方法。

然后你会看着它说，好吧，它的能力是否已经到了极限，还是才刚刚开始？然后你会说，嗯，驱动它的因素是什么？你会说，驱动它的因素是神经网络的大小、数据集的大小，以及你可以给它的计算量，无论是在训练还是推理过程中。

即使在当时，我们也觉得，哇，我们可以轻松地将所有这些东西扩展数千倍。即使我们不发明任何新东西，你也能从中获得很多成果。所以，这是我所说的具有很大发展空间的技术。它还没有被充分优化。但作为一项技术，它正处于成长和发展的初期。

当我们投资于它们时，我认为这就是我们拥有大量机会的地方，而不是那些已经被优化了 150 年的技术。没有什么是一目了然的，事后看来很简单。我曾经见过人们在某些事情出现时质疑和怀疑它。

三年后，它变得很明显，“哦，我早就知道这会很棒。”而我会说，不，你不知道。我认为这没关系。这是一种人类的天性，在你能触摸和感受到它之前会怀疑某些事情。你必须迈出那一步。

不是每个人都采取那一步。新的技术出现时，开始阶段，总会有一些人说，“好吧，我不相信它会扩大或发挥作用。”有时候他们是对的，有时候事情不奏效，无法扩大。

但通常情况下，每当你把某件事情做到有用的地步，总是会有其他有用的事情它还没做到。所以，我们知道，人工智能最早商业化成功的部分，是它分析图像的能力，说：“好吧，我们可以开始标记这些图像中的东西并理解它们。”

翻译开始工作得相当好，所以你可以做文本到文本的翻译。但这个想法，像一个聊天机器人，我可以与之交谈，它有任何智能的迹象，它比我们以前拥有的任何东西都更好，但仍然不是很好。

演示效果很糟糕，你几乎无法分辨出照片中有多少人？照片中有猫吗？如果你想到“我可以用它做什么”，它并不明显。这并不像已经足够好到可以做任何事情。

所以，你必须看看它并说，“哦，不，这会更好。”这对很多人来说很难相信。然后，直到你能触摸和感受到它，真的像ChatGPT的时刻，大多数消费者第一次与AI聊天机器人有了亲身体验，你才会说，“哦，等等，它实际上可以为我做一些有用的事情，现在我相信了。”

我做过一次类似的经历，关于自动驾驶汽车。很多人一开始都很害怕，就像，“哦，它会让我非常紧张。”我在旧金山坐过几次Waymo的自动驾驶出租车，那是我的最爱之一。

你坐进车里，大约5分钟后，你就会感到无聊。你会说，“哦，它就像一个更好的司机，比大多数心不在焉、疲劳的人类都要好。”然后你会拿出手机，开始玩。

这就是新技术的挑战。我还没有遇到过一种新技术，直到我能以明显有用的方式展示给你，你才能亲自体验。它非常容易让人怀疑。一旦你达到这一点，你就拥有了所有的价值。

所以，很多挑战在于如何识别那些有机会达到这一点的技术，但还没达到。为什么？因为那就是产生影响的地方。

所以，如果我正在研究一项新技术，试图决定它是否可能成为一种突破性、变革性的技术，我认为有三个核心因素我正在寻找。第一个也是最重要的一个问题是，某种意义上的光速测试。

就我们所知，在实验物理学中，你不可能比光速更快。如果我正在制造宇宙飞船，你知道，如果我以光速的99.9%飞行，那么我没有太多加速的空间，对吧？如果你以光速的0.001飞行，我还有很大的加速空间，在我达到任何理论极限之前。

所以，对于大多数技术来说，我的第一个问题是，当前版本距离理论最大值有多远？还有多少提升空间？是几千倍吗？是1%吗？这是问题一。

第二个问题是有利的外部因素吗？是否有一些事情正在发生，这些事情让这项技术变得更好，而你并没有在做这些事情？通常，这意味着有一些组件的输入可以年复一年地改善，而不需要你的工作。

在人工智能领域，例如，获得更多的计算能力正在发生，而不需要我们的努力，因为英伟达、台积电、ASML、整个芯片生态系统都在开发更快、更强大的芯片，一年又一年。

所以，对于同样的钱，我每 18 个月就能获得大约两倍的计算能力。所以，我不需要做任何工作。我不需要建立一个芯片团队并进行芯片设计。每一年，我都会获得更多的计算能力，而不需要我做任何事情。

所以，问题二是我是否拥有某种顺风顺水的条件，让我的产品在即使我睡觉的时候也能变得更好。第三个问题，也是最难的问题，是你要解决什么问题，对你的客户来说有多重要？你的客户可以是消费者，也可以是公司，但有很多惊人的技术已经取得了巨大的进步，但并没有真正解决人们的问题。

最大的例子就是 3D 电视。多年来，每个人都说，“3D 是从 2D 升级的下一步。它更身临其境。”结果发现，人们不想戴特殊眼镜，你知道，坐在沙发上这样做。

所以，3D 电视是一种技术飞跃，但并没有解决消费者的痛点。所以，最终它失败了。我们用先进技术做的是尝试模拟它们，说：“好吧，我还没有制造出这个东西。什么是我可以使用的最好的代理来向人们展示，如果我制造了这个东西，你会喜欢它吗？”

这就是客户探索，它真的很重要，因为最终有人必须购买那项技术，必须有一个商业模式来支付所有费用，必须有人购买它来资助所有的研发。如果没有一个赚钱的循环在那里，它最终会消失。

对我来说，问题很明显。我是说，这是一个我长期以来一直很热衷的事情。我买了第一辆日产聆风，那是第一辆我能买到的电动汽车。早期的产品。很糟糕。续航里程很差。

真正的问题是，是否有办法解决这个问题？它感觉像是一个巨大的、令人不知所措的问题。这需要一些傲慢或天真，认为我可以真正对其产生有意义的影响。

我认为，在某个时候，我决定了这并不重要，是否我可以做到。我必须试试。当我思考人类需要解决的问题时，大量的额外清洁能源是一切的上游。

你知道，如果你从更长远的角度看工业革命，人类真正做的事情就是利用能源，无论是动物、燃料还是可再生能源。

所以，不是手动耕作或手动挖掘，你知道，我们有机器为我们做这些事情。这大大提高了生产力、人类健康和幸福。你知道，我们取得人工智能进步的唯一方式就是大幅增加能源使用。

我们取得人工智能进步的唯一途径就是大幅增加能源使用。我们让人们拥有空调、清洁水和舒适的生活的唯一途径就是能源。那就是能源。

我们知道如何淡化水并使其清洁。我们知道如何让人们在炎热的天气下保持凉爽。我们知道如何制造各种各样的东西。我们知道如何制造超级智能的辅助工具。

我们不知道如何将其扩展到 80 亿人，而不需要数太瓦的额外清洁能源。有很多方法可以解决这个问题，也有许多企业家正在尝试解决这些问题，这些问题可以对世界产生影响。

我花了很长时间才弄清楚具体该怎么做。我想到的结论是，解决可持续性问题就是重新设计我们经济中的数万亿美元。

你知道，能源本身就是一个多万亿美元的行业。如果你想做到这一点，你不能仅靠政府资金或慈善捐款。你需要企业投资。

而实现这一点的最佳方式通常是通过初创公司。你有这批优秀的企业家在那里追逐从聚变到下一代微反应堆到近海 AI 数据中心、厨房脱水器等各种想法。

而我 25 年来在建立公司方面的个人经历与此非常吻合。当我与企业家们一起度过一段时间时，我意识到我可以做很多事情来帮助他们避免在组建团队时常犯的错误。

我如何招聘高管？我如何管理产品开发？所有这些事情。所以，这与我的技能和我想在世界上看到的变革完美吻合。

看到这些产品在市场上取得成功真是令人激动。所以，在清洁能源领域有很多令人兴奋的工作，这些工作在事实上是无限的，意味着我们可以将其提高 10 倍、5 倍，你知道的，远远超过我们今天在全球范围内使用的电力。

最令人兴奋和最雄心勃勃的是核聚变。基本上就是我们太阳的能源来源。我们知道它在宇宙中是有效的。我们实际上已经在地球上实现过它。

我们知道如何实现核聚变。我们只是还没有想出如何将其转化为可靠的能源来源。如果我能做到这一点，我可以建造一个几乎不需要任何投入就能产生大量能量的发电厂。

它是完全安全的，我们可以建造大量的这种类型的发电厂，可以为整个德克萨斯州的奥斯汀供电，一个发电厂每年只需要一辆皮卡车和一年的燃料，这简直是疯狂的。

从效率的角度来看，这几乎是发电的终极目标。所以，这是能源领域，我可以谈论更多关于这些的话题，但让我们再举一个例子，有关人们在家里可能会经历的事情，那就是扔掉垃圾。

当你把食物扔进垃圾桶时，它就会被送到垃圾填埋场。食物腐烂了，释放甲烷。这是近期的主要温室气体排放源和废物来源。

有一家叫 Mill 的公司。它是一个小垃圾桶。看起来像一个小弹出式垃圾桶。但它是神奇的。你把食物扔进去，它会干燥、粉碎并变成类似咖啡渣的东西。

你可以在一个普通家庭中做到这一点，持续一个月，到月底，你会有一个鞋盒大小的这些东西。最重要的是，它没有气味。

你不必为此而清空它一个月。你看，人们真正喜欢扔垃圾。而我会说，你可以减少扔垃圾的次数，它的气味也会减少，你知道吗？

这并不大，它的气味更小，那么每个人都很高兴。这被证明是减少与食物垃圾相关的温室气体排放的主要方法。

这家公司正在创造巨大的收入和利润。这是一个更好、更快、更便宜的例子。它让人们的生活更好，它解决了气候和环境问题。

结果证明这是一项伟大的商业投资。人们经营过大型项目。人们创办过公司，很少有人见过我所经历过的规模。

我建造了数百万平方英尺的数据中心空间。我们已经交付了数百万件消费类硬件产品。你知道，我们已经将团队扩大到数万人，管理了许多数十亿美元的收购。

所以，我很幸运地与一群绝对不可思议的人一起工作，参与了令人惊叹的技术项目。在这里讨论的所有事情中，我们带来的就是如何找到合适的问题、具有发展潜力的技术和有利的外部因素以及客户需求。

然后我们还没有讨论的是人。你知道，我的工作中有很大一部分是找出谁是合适的领导者，无论是技术还是组织领导，来推动某件事情向前发展。

在创业领域，团队最终是你押注的对象。你知道，这些人将建立那家公司。我们正在寻找那些能将公司发展到尽可能远的创始人。

这意味着在 10 名员工、种子轮或种子期阶段的公司与拥有数百名员工、拥有客户、处于 C 轮系列的公司是截然不同的。这种变化率对人类来说是不寻常的。你通常不会遇到变化如此剧烈的环境。

所以，有一小批人可以应对这些变化，我很幸运地与很多人合作过，我也有过这样的经历，所以在挑选创始人时，我会寻找那些能做到这一点的人。

我们也希望创始人具备一些特定的品质，比如无条件的坚韧、果断，以及学习新领域和快速适应的能力。

这并不容易总结成几个要点。但不幸的是，我们不能只通过几个简单的标准来评估创始人。我们只需要去认识创始人，然后做出自己的评估。这是我工作中最重要的部分。

我们通过多次与创始人会面来评估他们。我们通过联系他们的推荐人、与他们合作过的人来评估他们。但我会给你两个清洁技术领域的创始人例子，我认为他们非常出色。

Mill 的创始人马特·罗杰斯（Matt Rogers）。这是他的第二家公司。他的第一家公司是 Nest，他在那里做得非常出色，但他并没有就此止步。他建立了一支伟大的团队。

我们寻找创始人，我知道，他们的工作就是建立一家公司，解决以前从未有人解决过的问题。他们不知道如何解决这些问题，但他们会解决。

当我们遇到 Mill 时，他们说：“哦，我们将从美国政府那里获得批准，以利用食品垃圾并将其转化为动物饲料。”

我们将在X时间前完成。我们并不完全确定如何做到这一点，但我们会做到。大约2或3个月后，他们说：“是的，我们做到了。我们已经完成了。我们现在正在这样做。”

这就像一系列的，我们要解决这个问题，然后我们要解决下一个问题。作为一家公司，每个人都非常出色，但作为一个团队，他们可以解决大问题。

这就是创业的魔力。

<details>
<summary>Original English</summary>

Hi,
I'm
Mike
Sheper.
Um,
I
spent
25


years
building,
starting
and
scaling


technology
companies.
Worked
at
a
small


startup.
Worked
at
two
small
startups


actually
in
the
dotcom
boom.
Then


started
my
own
company
after
the
dot


crash.
Sold
that
off
to
a
bigger
company


on
micros
systemystems.
Then
joined


Mozilla
which
made
the
Firefox
web


browser
helped
ship
15
20
30
back
when


version
numbers
were
small.
They
weren't


in
the
'7s.
And
then
in
2008
I
joined
a


little
social
network
called
Facebook


which
was
you
know
smaller
than
MySpace


at
the
time.
And
then
over
the
next


14-15
years
led
engineering,
built
data


centers,
uh
took
us
into
the
consumer


hardware
business
with
virtual
reality


headsets,
managed
multiple


billion-dollar
acquisitions,
built
an
AI


research
lab,
sort
of
build
hardware,


software,
enterprise,
and
consumer,
and


scaled
the
team
from,
you
know,
100
to


35,000ish.
I'm
now
helping
great


entrepreneurs
through
starting
a
venture


capital
firm
called
Gigascale
Capital


where
we
are
hunting
for
founders
with


big
ideas
using
technological
advances


to
build
products
that
makes
people's


lives
better.
and
it
solves
a
climate


environmental
problem
and
it
turns
out


to
actually
be
a
great
business
along


the
way.
Easy
example
of
this
that
is


you
know
electric
vehicles
spew
no


pollution.
They're
cheaper
to
operate.


They're
cheaper
to
maintain.
They're


faster.
They're
quieter.
They
have
been


traditionally
more
expensive.
But
as


batteries
get
cheaper
and
cheaper,


they're
getting
more
and
more
cost
competitive.
And
so
we're
looking
for


products
that
are
cost
competitive
and


people
love
and
then
also
are
better
for


human
health
and
for
the
planet.
That's


a
version
of
what
we
do.


It
was
August
or
September
of
2008.
We


had
a
product
that
people
loved,
the


website
facebook.com.
More
people
were


signing
up
every
day.
Those
are
new


features
being
added
to
the
site
all
the


time.
At
the
time,
the
most
urgent


problem
was
scale.
Literally
the
problem


was
you
know
every
week
or
every
month


on
keeping
the
site
running
and
the


software
backbone
and
hardware
backbone


for
building
this
sort
of
product
didn't


really
exist.
So
a
lot
of
the
first
many


years
I
spent
there
was
scaling
and


rebuilding
the
software
architecture
of


the
site
and
then
building
the
hardware


infrastructure
to
do
this.
You
know
when


I
first
got
there
we
were
releasing


space
in
those
data
centers
and
putting


servers
in
them
and
and
building
it
up.


But
because
of
the
financial
crisis
of


the
real
estate
crisis
in
2008,
people


had
stopped
building
large
data
centers


and
so
we
couldn't
get
more
space.
So
we


had
to
build
our
own
and
there
was
a
lot


of
challenges
and
you
know
we
had
to


learn
a
lot.
None
of
these
things
were


perfect
the
first
time.
We
made
some


mistakes
and
we
had
to
fix
them
and
we


had
enough
humility
to
know
that
we
had


to
learn
a
bunch
of
new
things
and
so


you
know
the
first
goal
was
hiring


people
who
had
done
data
center
work
who


had
done
network
design
people
with
the


actual
background
in
this
space.
So
the


goal
was
go
hire
a
team
to
understand


this
new
area
and
then
help
us
get
good.


Like
many
problems
in
a
company,
you


know,
if
you're
a
founder
or
prospective


founder
watching
this,
most
of
the


things
a
startup
does,
it
does
out
of


just
complete
necessity.
It's
like,


well,
we
can't
get
space.
We
don't


really
have
a
choice
here.
We
have
to


figure
this
out.
One
of
the
lessons
that


I
took
away
was
there's
no
getting
away


from
the
hard
problems.
You
just
got
to


get
to
it.
So
it's
like
okay
if
we
need


to
solve
this
problem
let's
like
figure


out
what
are
the
critical
risks
what
are


the
things
we
understand
the
least
or


the
highest
technical
risks
and
let's


work
on
those
first.
I
think
there's


this
like
human
instinct
to
solve


tractable
problems.
So
it's
kind
of
like


I've
got
this
task
I
want
to
do
but
my


room's
kind
of
messy.
I'm
going to
go


clean
my
whole
office.
I
don't
want
to


go
work
on
the
hard
thing.
And
like


that's
a
very
human
sort
of
trait.
So


it's
like
important
to
work
against
it.


Say
like
actually
no
it
doesn't
matter.


My
office
is
messy.
like
if
I
don't
get


this
pitch
done
right
and
get
raise


money
for
this
company,
then
nothing


else
matters.
Doesn't
matter
if
my


office
is
clean
or
not.
So,
I
think
that


trying
to
get
people
focused
on
the


right
hard
problems
and
running
at
them


is
is
really
critical.


Early
years
of
Facebook,
it
was
sort
of


just
pants
on
fire
all
the
time.
Wasn't


totally
solved,
but
it
wasn't
so
much


consuming
all
of
our
energy
that
we
had


a
little
bit
of
energy
to
look
forward.


2013,
Facebook
started
the
AI
research.


One
of
the
questions
was,
you
know,


should
you
establish
a
more


broad-reaching
research
lab?
There
were


things
like
Microsoft
research
or
IBM


research
or
others
who
were
kind
of
like


could
do
software
theory,
could
do
lots


of
different
areas
of
technology.
The


most
preient
part
of
all
of
this
was
to


say
AI
is
such
a
big
thing
and
so


impactful
that
you
wouldn't
want
to


spend
time
on
these
other
things.
You'd


want
to
take
all
the
energy
you
had
and


focus
it
on
AI.
This
is
a
debate
Mark


and
I
had
and
I
you
know
I
think
he
was


the
one
actually
who
kind
of
pushed
to


say
like
let's
make
sure
we
just
do
AI.


Part
of
it
was
having
visibility
to
what


was
going
on
in
the
ecosystem.
There
was


you
know
this
the
imagenet
challenge
was


this
famous
academic
challenge
of
object


identification
that
no
one
was
paying


that
much
attention
to
except
when
the


first
neural
net
entered
the
challenge


and
sort
of
was
so
much
better
than


everything
else.
It
is
one
of
these
rare


moments,
you
know,
there
are
these


moments
in
tech
where
something
shows
up


and
it's
like,
"Oh
my
gosh,
that
thing


is
like
10%
better
than
anything,
a
very


large
amount
compared
to
any
other
gap,


and
it
used
neural
net's
training
on


data."
So,
it's
a
different
approach.


You
then
look
at
that
thing
and
say


like,
okay,
is
that
at
the
end
of
its


runway
in
terms
of
capability
or
is
it


at
the
beginning?
And
you
say,
well,


what's
powering
it?
you
say
what's


powering
it
is
like
the
size
of
the


neural
net,
the
size
of
the
data
set,


and
the
amount
of
computation
you
can


give
it
both
in
training
and
in


inference.
And
even
at
the
time
it
was


like
wow,
we
we
can
scale
all
of
those


things
by
thousands
of
times
easily.
And


so
even
if
we
invent
nothing
new,
you


could
get
a
lot
more
out
of
this.
And
so


this
is
what
I'd
like
to
say
is
like


it's
a
technology
with
a
lot
of
runway.


It's
not
fully
optimized.
But
but
the


point
being
is
as
a
technology,
it
was


at
its
infancy
in
terms
of
its
ability


to
grow
and
scale.
And
those
are
the


things
that
I
think
are
really
exciting.


And
I
think
when
we've,
you
know,
when


Meta's
invested
in
them
and
when
I've


invested
in
them,
that's
where
you
have


a
lot
of
opportunity
versus
things
that


have
been
optimized
for
150
years.
And


there's
just
not
a
lot
of
room
for
for


improvement.


Everything
always
feels
obvious
and


afterthought.
And
I
had
definitely
seen


people
who
at
the
time
something
was


coming
out
questioned
and
doubted
it.


And
then
3
years
later,
it
was
obvious,


oh,
I
knew
all
along
this
was
going to


be
great.
And
I
was
like,
no,
no,
you


didn't.
Um,
and
and
I
think
it's
fine.


It's
like
human
nature
to
doubt


something
until
you
can
touch
and
feel


it.
And
you
had
to
take
that
leap.
And


not
everyone,
you
know,
beginnings
of
a


new
technology
takes
that
leap
and
they


say,
"Well,
I
I
just
don't
believe
it's


going
to
scale
or
work."
And
sometimes


they're
right.
Like
sometimes
things


don't
work
and
they
don't
scale,
you


know,
and
then
every
time
you
get


something
to
a
useful
point,
there's


always
some
other
useful
point
it
hasn't


done
yet.
So
we
you
know
the
very
first


part
of
AI
that
started
working


commercially
really
well
was
this
sort


of
ability
to
analyze
images
and
to
say


like
okay
we
can
start
labeling
things


in
these
images
and
understand
them
um


translation
started
to
work
reasonably


well
so
you
could
do
text
you
know
from


one
language
to
another
but
this
idea
of


a
sort
of
chatbot
that
I
could
talk
to


that
had
any
semblance
of
intelligence


it
was
better
than
anything
we
had


before
but
still
not
very
good
at
the


time
the
demos
were
terrible
you
know


you
could
barely
say
how
many
people


were
in
this
photo?
Is
there
is
there
a


cat
in
this
photo?
And
if
you
think
of


this
from
a
like
what
can
I
do
with
it


standpoint,
it
wasn't
obvious
like
this


it
wasn't
good
enough
to
do
anything


with.
And
so
you'd
have
to
look
at
that


and
say
like
oh
no
no
that'll
get
better


which
is
hard
for
a
lot
of
people
to


believe.
And
then
it's
not
until
you
can


touch
and
feel
and
it's
really
the
kind


of
the
chat
GPT
moment
where
most


consumers
actually
had
their
first


experience
with
an
AI
chatbot
and
you're


like
oh
wait
it
actually
can
do
some


useful
things
for
me
now
I
believe.
You


know,
you
have
a
similar
experience
with


self-driving
cars.
Most
people
like


conceptually
get
scared
of
them
and


like,
"Oh,
it'
make
me
super
nervous."


I've
taken
a
lot
of
people
on
Whimo


rides
in
San
Francisco.
It's
one
of
my


favorite
things
to
do.
You
get
in
the


back
of
the
car,
like
5
minutes
in,


you're
bored.
You're
like,
"Oh,
it's


like
a
better
driver
than
than
most


distracted,
tired
humans."
Um,
and


you're
like
now
on
your
phone
and
you're


like,
"This
is
boring."
And
so
this
is


the
challenge
of
new
technology
is
I


have
yet
to
encounter
a
new
technology


that
until
I
could
show
it
to
you
in
a


way
that
was
obviously
useful
that
you


could
personally
experience.
It
is
so


easy
to
doubt
it.
Once
you
get
to
that


point,
you've
like
captured
all
the


value.
So
a
lot
of
the
challenge
is
like


how
do
you
identify
technologies
that


have
an
opportunity
to
get
to
that
point


but
aren't
there
yet?
Um
because
that's


where
the
sort
of
place
to
have
impact


is.
So
if
I'm
looking
at
a
new


technology
trying
to
decide
whether
this


is
something
that
might
be
a
breakout,


you
know,
transformative
technology,
I


think
there's
three
core
things
I'm


looking
for.
The
first
and
most


important
question
is
sort
of


understanding
the
light
speed
test.
As


far
as
we
know
in
experimental
physics,


you
can't
go
faster
than
the
speed
of


light.
If
I
was
building
spacecraft,
you


know,
and
I
was
at
99.9%
the
speed
of


light,
there's
not
a
lot
of
room
for
me


to
get
faster,
right?
And
you
say,
I'm


going
to
make
it
twice
as
fast.
Like


that's
really,
really,
really,
really


hard.
You
know,
if
I'm
at
0.001
Oh,
one


the
speed
of
light.
I
got
a
lot
of
room


to
go
before
I've
hit
any
theoretical


limit.
And
so
for
most
technologies,
my


first
question
is
how
far
away
from


theoretical
maximum
is
the
current


version
of
the
thing?
Like
how
much


headroom
do
you
have
to
scale
an


improvement?
Is
it
thousands
of
times?


Is
it
1%.
Um
that's
question
number
one.


Then
question
number
two
is
are
there


tailwinds?
Are
there
things
that
are


happening
that
make
this
technology


better