Speaker Application
SPEAKER APPLICATION
Sponsor
SPONSOR
Past Events
PAST EVENTS
2026 SF
2025 OAK
2024 AUS
2023 AUS
2022 AUS
2020 SF
All talks
A
L
L
T
A
L
K
S
VC Panel - The Present Future of Data-Oriented Startups | DataEngConf NY '16
Speakers
Matt Hartman
, Partner, Betaworks
David Beyer
, Partner, Amplify Partners
Evan Nisselson
, General Partner, LDV Capital
View transcript
00:00
So, thanks for attending the VC panel.
00:29
The idea behind the panel is, I'm a three-time founder, I'm a software engineer from the
00:36
first bubble, a three-time startup founder, and over the years I've learned a lot from
00:42
living in San Francisco and New York, and talking to venture capitalists, as painful
00:48
as it is, it's been really, really helpful to me.
00:54
So understanding how VCs think as a founder, how they view startups, what's important to
01:00
them, all of this stuff, it's sort of this, like, it's not fully a sub-theme of DataEngConf,
01:06
but I always personally want to make DataEngConf very friendly, not just for engineers and
01:11
scientists, but for engineers and scientists that think they might want to start companies.
01:16
So this is our first attempt to pull this content in, and I hope that you guys find
01:21
this useful and helpful.
01:23
We have some great panelists today.
01:26
We have Tim Devane from NextView Ventures, we have Evan from LDV Capital, we have David
01:34
from Amplify Partners in San Francisco, and we have Matt Hartman from Betaworks.
01:40
And these are all guys that I have known or talked to, or sort of respected through the
01:45
industry, and I think they all have some unique insights on data-oriented businesses, perhaps
01:50
some B2B, perhaps some B2C.
01:53
So we're just going to kind of have at it and see how it goes.
01:59
There's no office hours for this session because we'll go into the party at 5 o'clock downstairs,
02:04
so we'll save some time at the end if you guys have questions as well.
02:09
So let's start, like, in the very early stages of you guys sourcing deals.
02:15
How do you find data-oriented companies?
02:20
And we don't have to go in order, so sometimes I'll call on you and sometimes I won't.
02:27
But who wants to kick it off?
02:32
I'll talk about how we find them.
02:34
So we're set up as a studio and a venture fund.
02:38
So we have a team of data scientists, a couple of data scientists who work with a number
02:44
of our companies in the portfolio and in the studio, and so it's often through writing
02:48
that they do or someone will reach out to them with a new product that's sort of very
02:52
focused on data, and that is a main way that we get investments.
02:58
And then the other group is other VCs who, you know, focus in the area and look for a
03:04
partner who understands that vertical.
03:08
I think for us it's unlocking profound or previously unavailable or accessed data sets
03:15
is tremendously valuable no matter what type of company or startup, B2B, consumer, what
03:23
even the pitch is.
03:24
So as far as sourcing, I think it often comes out of that first conversation, whether the
03:31
company is data-focused from day one as part of their actual goal, mission, in what they're
03:36
building.
03:37
We usually determine that in our first meeting, so I wouldn't say that we go into it looking
03:43
expressly for data as a theme, more so that almost, as I say, 75, 80% of the companies
03:50
that we have invested in have unlocked a very, very valuable data set in something that they
03:57
do as far as their product or their service.
03:59
So it usually comes out of the first conversation, whether it's a focus of their product or if
04:03
it's something that they build into in year two or three and is secondary to what they're
04:09
building.
04:11
Yeah, for us, we're pretty tightly focused on, we do early stage, so seed and some A.
04:19
And we're tightly focused on the idea of the practitioner turned founder.
04:22
So we invested in companies like Datadog and Chef and Fastly primarily because, and it
04:28
informs sort of the theme for the rest of our fund in the sense of finding people who
04:33
may have not started companies before, in fact 80% of our founders or so, our first
04:37
time founders, who had to solve a problem, a big problem in some company that they worked
04:42
at because there was nothing else to buy that could potentially help them solve that problem.
04:48
In the process of building that, they sort of lift their head up and say, actually this
04:51
could be pretty useful for a lot of other people in the enterprise.
04:56
So that's, in focusing on that sort of cohort of people, we end up spending a ton of time
05:03
with just strong technologists at companies like LinkedIn and Pinterest, Google, you know,
05:10
you name it, who are working on interesting problems, sort of at the cutting edge, but
05:15
don't necessarily know how to build a business or think about it.
05:18
That's kind of our job there.
05:21
So LDV Capital, we have a very, very focused thesis.
05:26
So we only invest in people building visual technology businesses.
05:31
And so that's that technology that captures, analyzes, filters, monetizes, or displays
05:37
visual data.
05:38
So to the point of sourcing deals, most all of our companies have a tremendous differentiation
05:45
with the data set that they're analyzing.
05:48
Clarify here in New York, you might be aware of, is one of our companies, we invested the
05:52
first external money, the second in the A, and recently in the B.
05:55
And so finding deep, technically savvy founders or co-founders anywhere in Northern America
06:03
or Northern Europe is our focus, because my background is 18 years building visual tech
06:08
businesses.
06:09
I'm not technical, but I work great with very technical folks and help them with all the
06:14
other stuff that they might not be experts in.
06:17
As another example, a company called Mapillary in Sweden, which is a crowdsourced street
06:21
view, stitching together images from tens of thousands of people around the world to
06:25
make a more real-time Google Street View.
06:28
And those are two examples of very data-focused, of which we get from a summit and from a community
06:34
of these computer vision experts, data experts.
06:37
So always looking for those types of folks.
06:41
So visual image is not something we always think of as interesting data sets.
06:52
What are other aspects where machine learning is sort of transforming, I mean, I think the
06:57
visual focus is interesting.
07:00
Does anyone have other examples of sort of subject matter, interesting niches where machine
07:07
learning is sort of changing our perception of what a new market could be?
07:14
Most of my work at Amplify, we do cloud infrastructure and InfoSec, but I do mostly, actually almost
07:18
all machine learning-based stuff.
07:20
And the three sort of key or really cool areas, I guess you could say, unstructured data,
07:26
which includes visual image data, speech, and text, and the vast majority of the really
07:33
interesting research and deep learning in the past three, four years, actually, before
07:38
that, a lot of the key breakthroughs were around image data, right, Clarify being sort
07:43
of a startup spinning out to address some of those kinds of things.
07:47
But there's some really cool stuff around working with speech, working with text.
07:52
And if you combine all three, you actually start to solve really interesting business
07:56
problems for things like in insurance.
07:59
agriculture, you know, manufacturing, that kind of stuff, yeah.
08:04
So David, I know you invest in developer tools, engineering platforms.
08:10
In my experience as an engineering founder,
08:13
I found that not all VCs are as friendly to the developer tools market.
08:20
Does anyone have any perspective on that?
08:23
David could probably go on for a while about this.
08:26
Does anyone else, do the rest of you guys invest in developer tools at all or platforms?
08:30
Yeah.
08:31
Like you guys do?
08:32
Yeah, two in New York, a company called Code Climate,
08:35
just did a series of announcements,
08:38
which is a real-time code review that runs on the site pre-launch
08:44
and then runs throughout regardless of the size of the site or the product.
08:52
So what we loved about that was that there was a vision for solving a relatively simple-to-understand problem.
09:00
It's basically spellcheck for that type of code.
09:07
But the vision beyond it was there's a desire to put healthy and good source code out into the world,
09:14
and if you can build both a community as well as a unique and increasingly proprietary data set around that,
09:23
that is a tremendous thing to do across a variety of languages.
09:27
So that is one.
09:29
That was a seed investment that we led two years ago.
09:33
And the other is a different take on this in which I think some of the larger services available today
09:42
that have reduced the cost of data analysis and sort of real-time storage,
09:47
things like Kinesis and Lambda Hadoop that are now available to entrepreneurs,
09:52
have allowed for people to take a look at existing and available units of information
09:59
that you might not have previously thought of as unique and valuable data and take a different spin on it.
10:05
So the one I'm referring to in particular is a company that's aiming to revamp the analysis of user and application logs.
10:14
So instead of it being kind of a static view that's tossed out every seven days,
10:18
it's only really accessed at the point of something going wrong.
10:22
Could you actually reimagine how they are both served into the engineer as well as access-tailed and deconstructive, essentially?
10:33
So one, it's a better interface straight up.
10:35
So that's the simple sort of first problem solve.
10:38
And then that working for people and being better than the solutions that are out there today,
10:42
you can then extract what is kind of the lifeblood of a lot of products and services,
10:47
the log itself, and provide that to the entire organization.
10:51
So those are two that I think, you know, two different directions within developer tools,
10:57
but they generate extraordinarily interesting data or access previously sort of overlooked sets.
11:07
Both of those are in New York.
11:09
I think logging data is interesting.
11:13
I've seen some new startups in the field of anomaly detection.
11:20
And a lot of the big companies, Uber and Google and others, Netflix,
11:23
have been building these systems internally for a while
11:26
to use data science to sort of float up notifications of issues with their massive infrastructure.
11:34
It seems to me that we're starting to see a few startups in this category.
11:38
Does anyone want to speak?
11:40
And perhaps the one you mentioned, Tim, is sort of in that.
11:42
Yeah.
11:43
It was a very, very early investment.
11:45
It's called Timber.io if anyone wants to check it out.
11:49
But they're not the only ones.
11:52
I think there's a desire to find a solution better than Papertrail or Elasticsearch
11:58
because they're kind of like a clunky or expensive, you know, really built for a DevOps team,
12:03
not necessarily for a single engineer side project or startup sort of team size.
12:11
David, did you want to comment?
12:12
Just on the question of developer tools, I think it's important to sort of reflect on,
12:18
I think when people say developer tools, it almost kind of has this diminutive sense.
12:24
Oh, it's just, you know, little GitHub projects for developers,
12:28
you know, some library that someone can use in their code.
12:31
Obviously, it's not what literally people mean, but I think it understates the potential.
12:36
And that is because, you know, most of our time is spent with people who buy technology at large companies.
12:43
The buyers that we spend time with are CIOs typically or in the IT function somewhere
12:47
or VPs of Eng at large businesses.
12:50
A lot of them see their businesses transforming from the paper business or the oil business
12:57
or whatnot into technology businesses.
13:00
And so developer tools is really how do we get leverage.
13:05
If we're at J.P. Morgan, we have 40,000 people working in IT.
13:08
It's the largest, aside from the government, the largest IT employers in the world.
13:15
That's not a small sum of people who need things to do their job better.
13:20
And I think from our perspective, of course,
13:23
there are really bad ideas for developer-focused tools that should not become companies, right?
13:30
That goes without saying.
13:32
But what no one has really figured out yet is the channel.
13:35
How do you build a channel to the developer?
13:37
And some people like Atlassian and GitHub have made strides towards that goal.
13:42
But if you look at the sort of previous crop of massive technology businesses,
13:46
HP, IBM, Oracle, those people, they own the channel.
13:50
If they buy a company or if they spool up a new product, they just tell their salespeople,
13:55
you sold this account, this thing, just tack that thing on,
13:59
and that becomes a $100-, $200-, $300-million business in no time.
14:03
So the big opportunity for people and for companies in this space is
14:08
can you find a way to get deeply embedded in the workflow of the developer in some form?
14:15
There's some interesting people doing work around applying machine learning to software development.
14:21
If you become sort of an integral part of that lifecycle, of that workflow, of that person's life,
14:28
there's a lot that you can then do and a lot that you can sell them on top of that.
14:33
So it's a huge, untapped, unsolved distribution problem.
14:37
So just developer tools, I think any VC that doesn't recognize the potential there,
14:43
unless it's not their focus, I think is missing a potentially interesting category.
14:47
So I have a question about that.
14:48
So if you're talking about developer tools that are sort of fundamentally around data science,
14:54
sort of have some connection with data science or are sort of technically difficult a little bit,
14:58
when you're vetting these companies, do you invest in companies that are building technology
15:03
or do you invest in sales teams with technology that is able to be sold into an organization?
15:10
Well, because we're early stage, we're typically investing in the technical teams
15:14
who don't know the first thing about building a sales or go-to-market organization.
15:18
So if your thesis is that the way to win is by having a totally different go-to-market,
15:24
then how do you tell between a really great tech team that has no idea how to go to market and a great tech team?
15:30
Good question. It's not easy.
15:32
I think the question that comes up is, one, with sort of the table stakes for us,
15:37
is this a good engineering team?
15:38
And that's usually, you know, that's evidenced by their background,
15:42
having built really impressive, interesting things or led teams before when they built those things.
15:47
The second question is, is there some way to sort of thread the needle with distribution,
15:54
whether it's around community, whether, in some cases,
15:57
it's an existing open source project that already exists.
15:59
has a lot of legs that you can throw a commercial organization
16:04
on top of that gets you loved by developers.
16:08
And a lot of the time, we'll tell folks
16:11
who are working on this, we don't
16:13
know how you're eventually going to succeed,
16:16
but we're not pressuring you to think about sales right now.
16:19
It almost looks a lot more like the way
16:21
people spoke about Facebook and the social media companies
16:24
in 2006, 2007, get eyeballs.
16:27
It sounds naive, but in reality, if you can get,
16:33
there's I think 1.8 million software developers
16:37
in the United States, if I'm not off by an order of magnitude,
16:40
I don't think.
16:40
If you can get into that workflow somehow,
16:44
and it's not going to be one clever thing,
16:46
it's going to be death by 1,000 cuts,
16:51
there's a lot that you can then do with that.
16:53
Granted, we're a series C and sometimes series A investors,
16:57
so we don't know the answer.
16:59
But we know if you build a product that people love,
17:02
it will sort of sort itself out, obviously not without effort.
17:06
But we think there's a large opportunity there.
17:08
We don't know exactly what it looks like,
17:10
but there's room to create and own
17:13
a couple of channels around this that I think don't exist yet.
17:17
We could be wrong, but that's our convention.
17:20
Just as a note to our audience, this little exchange
17:24
represents the quickest slide you
17:26
can go from a technical conversation to a sales
17:29
and marketing conversation.
17:30
And this will constantly happen when you're talking to VCs,
17:33
as a technical person.
17:34
You'll be, oh, I have this amazing technology,
17:36
and this is going to change the world.
17:37
And then they'll say, oh, how are you going to sell it?
17:40
How are you going to market it?
17:41
How are people going to know who you are?
17:42
What kind of traction do you have?
17:43
Constantly.
17:44
So be ready for this as engineers,
17:47
because if you think the sales and marketing
17:48
and go-to-market stuff is boring and not for you,
17:52
or you need to find a co-founder.
17:54
So this is just a little microcosm
17:55
of how quickly we go from a technical panel
17:58
to the real business issues, which are absolutely important.
18:01
Or finding an investor that would
18:03
love to help that data scientist do that stuff
18:06
until they need to hire someone.
18:07
Right.
18:08
Absolutely.
18:08
Or it's a project.
18:09
I mean, not everything has to be done.
18:11
If you have built something that's profound,
18:14
you don't necessarily want to go sell it into the Fortune 500.
18:17
Because sales and marketing can be really boring sometimes
18:21
and sucks a lot of the time.
18:22
So if it's not even natively part of your DNA,
18:24
or something that you have an ambition for with that thing
18:27
that you've built, that's OK, too.
18:28
But it is something, I'd say, to your question, Matt,
18:32
technology first, authentic experience first, always,
18:35
particularly at our stage.
18:38
But there's a reason to look for and discuss the desire
18:42
to carry the banner for the company.
18:46
That's at events like this.
18:47
It's also at going in to talk to companies
18:51
and interfacing with people that might not
18:54
be the ones you're going to hang out with on a Saturday night.
18:56
Actually, I'll guarantee that they aren't.
18:58
But they can move your business forward.
19:00
And it's only, there's two things.
19:02
I think it's defining when it goes from a technology
19:05
to you decide you want to make this a business.
19:07
That can be a very different thing.
19:09
There's also, there's only so many examples
19:12
of the organic, we're going to open source it,
19:14
give it away for free.
19:15
And everyone falls in love with it, bottoms up.
19:18
That's a beautiful vision.
19:19
And you know, not everybody builds slack.
19:24
And it's not just saying that they're the best in the world.
19:26
But there are a number of things that
19:28
came together to make that suddenly hit for a lot of people.
19:32
You can't just bank on that.
19:34
So Evan, you like to invest in unique, proprietary tech.
19:41
I mean, every VC does.
19:43
But when you find a technical founder
19:45
that you think looks interesting,
19:48
how do you advise them on all of the other team components
19:51
of what it might take to actually turn that
19:53
into a proper founding team or a startup
19:55
that actually has some legs?
19:58
I think it's a good question.
19:59
So once you find that one or two, three people
20:03
that are very technical with the deep domain expertise,
20:06
it really changes depending on those people.
20:08
Because some of them, let's say Matthew Clarify,
20:12
he does have that desire and hustle
20:14
and pieces of the entrepreneurial DNA
20:17
that goes out and closes deals.
20:19
I didn't know that when I first met him.
20:21
I learned that over time and he has that.
20:23
Other people might have other aspects
20:24
that are more about recruiting, but not the sales part.
20:30
So it really changes depending on each person.
20:32
So it really is, at the end of the day,
20:33
people build businesses.
20:37
People make money.
20:38
People recruit other people.
20:40
And so you have to adjust that collaboration
20:43
depending on that person.
20:45
So there's a young Eric from Mapillary,
20:49
sold his last company to Apple.
20:52
And he has some of that business savvy,
20:54
but he's not a sales guy.
20:56
He's not the hustler.
20:57
So then I help him with that.
20:58
And or most importantly, try to help find
21:01
the right people for the team.
21:03
Because I can only help for a little while
21:05
or investors should not be in the weeds for years.
21:08
But I am happy to get in for six months
21:10
and actually help with some of the negotiations
21:13
and figure out together with the founder,
21:16
what's the right fit to then recruit into.
21:19
Because if you don't interact, you don't know.
21:21
I don't like, because of my 18 years as an entrepreneur,
21:24
there's a lot of people that talk but don't do.
21:27
And I can't stand those folks.
21:29
And I don't invest in those folks.
21:32
But I try to be the person that does.
21:33
And if the floor needs mopping, I'll do that.
21:36
So trying to find, it's almost a sociological interaction
21:40
with the founder to figure out what they think they need.
21:43
Keep each other honest.
21:44
I don't know if that answers the question,
21:46
but I think it changes depending on every person
21:48
you invest in and at every stage of the game.
21:50
So another example is either one of those two companies
21:53
now have raised an A or B.
21:55
People they need is very different.
21:56
And I actually-
21:57
Some rich markups too.
21:59
What's that?
21:59
Some rich markups there too.
22:00
Well, hopefully they continue to be valuable.
22:04
And it's good for all the shareholders.
22:07
So there's an interesting thing there.
22:08
So for example, how do I help specifically
22:11
with Jan-Erik, we were first money in
22:14
and every day for the last three years,
22:16
every two weeks we have a standing call.
22:18
We still do since he's done the Series A.
22:20
And we talk about recruiting, firing, product market,
22:24
whatever is appropriate, as almost a sounding board.
22:28
And now we finally just, not as a good or bad thing,
22:31
change it from two weeks to every three weeks.
22:33
So the collaboration doesn't end early.
22:37
And I like to do that if the entrepreneur wants.
22:39
Others we don't talk, we talk once a year,
22:42
but they have my cell phone
22:43
and they can call me any single time.
22:44
I tell them to call me on the cell phone
22:47
if they're in a meeting
22:47
and they don't know the answer to a question,
22:49
go to the bathroom and call me on the cell phone
22:51
and we can talk through the answers.
22:53
And I'm always available.
22:54
Because that's what I wanted when I was an entrepreneur.
22:57
Call your VC from the bathroom, folks.
22:59
I've had those calls and they freaked out.
23:02
And it was, you know, it's not that I have an answer,
23:04
but two smart people discussing it
23:07
and having the confidence that maybe their view was right,
23:11
but they were worried that it was wrong.
23:13
And it's as simple as saying, no, no, you're right.
23:15
You know, call them out on that and tell them they're wrong.
23:19
And if they don't agree, walk out.
23:21
It's the beginning of every great joke,
23:22
a VC and entrepreneur walk into a bathroom.
23:26
One's virtual and one's taking a shit, yes.
23:28
Could be a data ant com meme.
23:31
All right, so let's take it back
23:32
a little bit more technical then.
23:34
So, David, you mentioned image data, speech, text, et cetera.
23:44
So, Matt, you're probably not going to convince me today
23:50
that everybody needs a personal assistant bot.
23:55
I'll try.
23:57
I know you'd love to try.
23:58
That could be at the party.
23:59
That's later.
24:01
But from a technical perspective,
24:03
explain to us how conversational interfaces are
24:07
married to machine learning, data, obviously NLP.
24:12
Like, why is this stuff all happening at the same time?
24:15
Is it a coincidence?
24:15
Or is it because there was stuff that we couldn't do before
24:18
that now we can do, which is driving
24:19
conversational interfaces?
24:21
I mean, I think there's a couple of separate things
24:23
that are happening at the same time.
24:25
One is there's these new places that you can talk to software
24:33
and you do it via text, right?
24:35
So like, it was just a pain in the neck
24:37
to get people to download your app.
24:39
And so now you can, whether it's ship them
24:41
a UI and a Facebook message that has a bunch of buttons,
24:43
or ask Alexa a question.
24:45
Alexa's probably a good example, because it's
24:47
like, there's no buttons.
24:48
You have no choice as a developer
24:49
to send someone buttons.
24:51
So you either have kind of individual commands,
24:53
or you communicate with the person
24:57
through some kind of natural language, right?
24:59
And I think what we found is we did this thing
25:02
over the summer called Bot Camp, where we had eight companies
25:05
building different bots.
25:06
And they basically fell into two categories.
25:08
One was, they said, you know what?
25:10
I can solve this problem by shipping a button,
25:13
or by just having a keyword and someone just asks it.
25:16
And that's actually easier than saying, hey, Facebook
25:19
Messenger, here's where I am, my location.
25:22
Like, will you please send me an Uber?
25:23
Is actually much, much harder than tapping an Uber, right?
25:27
But they either went that direction,
25:29
or they went the opposite direction,
25:30
was they built their own kind of some kind of natural language
25:34
processing.
25:35
And usually, that mattered.
25:38
It only really makes sense when you
25:39
have your own unique data sets.
25:41
Like, as a startup, what we found, at least,
25:45
is that most of the algorithms are out there.
25:47
The data is the thing that's the sort of key resource
25:50
that you don't have.
25:51
And so if you're going to compete algorithmically
25:54
against a data set that Google has a better data set than you,
25:57
then that's going to be a really hard sell to an organization.
26:00
Yeah, the question I have to you there
26:01
is, I've been fantasizing about amazing personal assistants.
26:06
There have been a few companies.
26:09
There's a long way to go there.
26:11
And the technology exists to make those work
26:14
from a startup perspective.
26:16
Google will crush you if they want, right?
26:18
They've acquired a few startups around that time,
26:20
Phil, and some others.
26:22
But they are sitting on the mother load of the data set.
26:25
And the question is, you talk to folks in Google,
26:27
and they say, well, they're not really pushing on it.
26:30
Like, there hasn't been a real update to Calendar,
26:32
for example, in like months.
26:34
How do you invest in companies looking
26:36
to tackle things that have to do with email, Calendar,
26:40
and other sort of personal logistical information
26:43
with a threat that Google could, with one feature, wipe them out?
26:47
Yeah, well, I think there's two things in your question.
26:49
One is, can you, so let's take emails.
26:52
I think email's like a fascinating data set,
26:54
because you do sort of own it.
26:55
Like, you can bring that, your email set of data,
26:59
to any company, right, with one connection.
27:05
I think the issue is actually on the other side, which
27:07
is the consumer behavior side.
27:09
So if you have a personal assistant bot that
27:11
will do anything for you, right?
27:12
Like, people are asking Alexa to do all kinds of stuff.
27:15
And it's like, I'm sorry, I don't know how to do that.
27:16
I think that Google Home is shipping now.
27:18
And that's, people at least can ask that questions
27:21
that they would ask Google, which is sort of a bigger
27:23
solution set.
27:24
But I think part of the reason that both Google and companies
27:28
like Magic, or whatever, that say
27:29
they can be a personal assistant to anyone,
27:32
haven't worked, is less about the technology.
27:34
It will be about the technology once they
27:35
can get people to actually use it frequently.
27:38
But it's less about like figuring out
27:41
what someone said in sentence structure,
27:42
and more about just reminding people
27:44
that they can use this thing.
27:46
If you just have an open text box, when do you use it?
27:48
You don't even use Google for absolutely everything
27:50
when you think about it.
27:51
And my one example, my example I think of
27:52
is, I don't know if you all have used Uber Rush, which
27:56
is where they'll send something to you.
27:58
I use Uber once a week, twice a week, really, really frequently.
28:03
So the once every two month time that I
28:05
need to get something rushed delivered to me,
28:07
I might think of them.
28:08
But I don't know that the startup would work the other way
28:10
around, where it's like, when I think about this,
28:12
I'm just, once every two months, they're
28:14
going to have to acquire me at exactly the right time
28:16
and remind me that they exist.
28:18
So I guess the conclusion that we've come to, at least,
28:24
is, if you can find repeat use cases
28:28
with very, very specific verticals with data
28:31
sets that these big companies either don't have access to
28:34
or can't, for some reason, strategically leverage,
28:37
I think that the don't have access to is a bit easier,
28:40
then you can, then there are opportunities.
28:43
But then that's where I was asking about the sales thing,
28:44
because then you have, what's the moat issue?
28:46
One way to solve that is by getting your own personal data,
28:49
letting people bring in data or companies bring in data
28:52
that other people, because if Google doesn't have access to it,
28:57
we don't have access to it.
28:58
But then you need something else, right?
29:01
That was a really whiny answer.
29:04
No, I think it's interesting about this concept of access.
29:08
And so this access is, I have rights to,
29:12
or I can steal data, OK?
29:16
So Facebook succeeded on stealing, or scraping,
29:22
or using, or borrowing, or I have access.
29:28
They started by stealing all the profile images from Harvard.
29:31
It wasn't legal.
29:32
So a lot of researchers, a lot of very brilliant data
29:36
scientists, well, I can't have access to that.
29:38
So is it because you don't think it's
29:42
legit, or you don't want to scrape it,
29:44
or there's a serial entrepreneur in me who will say,
29:50
do you ask forgiveness, or do you ask permission?
29:54
And so you can imagine which one I'm going to suggest.
29:58
I would argue that in either case
30:00
doesn't matter, because if you have,
30:03
regardless of whether the access is great or not,
30:05
if you have access to it, your startup,
30:07
then I have access to it myself.
30:08
I agree.
30:09
And I'm trying to say, don't let not having access slow you
30:15
down.
30:16
Get the data, build your brilliance on top of it,
30:20
and then figure out afterwards.
30:22
Get a great lawyer.
30:23
Yeah, and figure it out.
30:24
So I'm not saying do anything illegal,
30:26
because I'm on the record.
30:27
I'm just saying prove what you, and either partner and get
30:33
great data, or get the great data and prove your solution,
30:37
and then move forward.
30:38
When you get picked up by the FBI,
30:39
Evan will take your call in the bathroom.
30:43
I think it's one of the ways that if the sort of laggard
30:46
industries that are adopting data-oriented, big data
30:52
strategies now, health care, fintech,
30:56
one of the ways that entrepreneurs have broken in
30:59
is to build a short-term product that
31:00
encourages the individual to provide their data,
31:03
because there's something solved that's simple that
31:07
is a benefit to them.
31:08
And the longer-term vision is then once you have it,
31:12
and maybe there's a data network effect that comes out of that,
31:15
you can build something that is significant
31:17
to the entire industry, that because of HIPAA, or FINRA,
31:19
or whatever, straight up you couldn't just do it at first.
31:22
So if you can do something for the individual that
31:24
says, here's the banking information,
31:28
here's my credit score, here's the results of my test,
31:31
that can be a short-term band-aid or way around
31:35
that, once you start having a lot of people contribute
31:37
to it, the industry has to acknowledge you,
31:40
because you now have something that they don't even
31:42
necessarily have access to, at least on the software level.
31:45
Yeah, I agree.
31:46
That's probably a more diplomatic way
31:47
of saying what I'm saying.
31:50
Well, this idea that you can augment your data
31:53
set with usage, this is sort of a core component of any UGC
31:59
social.
31:59
model that we've seen on the consumer side.
32:03
What's an example of this where you can actually
32:06
add value to your data through B2B usage?
32:09
Anyone want to take that, David?
32:11
Well, I mean, even in the case of Clarify,
32:14
we invest in a company that does machine learning
32:16
for radiology.
32:17
We have one that does sort of bio-derived materials work.
32:22
A lot of these things basically work
32:24
on getting the users or the customers to train.
32:28
And if you're selling to B2B frequently, it depends.
32:32
The scale isn't always enough for it to be substantial.
32:36
But I actually think even if the data, as you were mentioning,
32:40
is out there and theoretically accessible
32:42
to any number of entrants, there is a real advantage,
32:46
a sort of virtuous loop between technology, team, and data
32:49
that exists, where the folks at Clarify
32:53
will be able to, by virtue of the core team,
32:55
recruit the subsequent marginal fantastic engineer, which
33:01
will then make them more efficient in collecting data.
33:05
We invest in this company called Diffbot,
33:06
which is trying to sort of structure the web.
33:08
And they're building, effectively,
33:10
the Google Knowledge Graph.
33:12
And they're a really talented team.
33:15
And by virtue of the accomplishments they've had,
33:18
they've been able to pull additional amazing engineers,
33:21
which puts them at an advantage.
33:22
So even though the algorithms exist out there,
33:26
anyone could theoretically use them,
33:27
there is a difference between folks
33:29
who can actually make them work in production-ready ways,
33:33
efficiently, at scale.
33:35
Those are actual proprietary advantages, in a way.
33:38
Because there are not that many people
33:40
who are willing to work at a startup who
33:41
can do that sort of work.
33:42
So I would sort of qualify the idea
33:45
that algorithms are cheap and easily available to anyone
33:53
with a computer.
33:54
It's not quite true.
33:56
Well, it appears that that's the case.
33:58
Because Google open-sourcing TensorFlow, the machine
34:01
learning platform, and all these other data science tools
34:06
that we're seeing open-sourced, it
34:08
would appear that the largest companies believe
34:12
that classifiers are going to become commoditized.
34:16
And they're relying on their massive data sets
34:20
to train their neural nets, which
34:23
they know that no one else can have access to.
34:24
And so in reality, they're actually
34:26
willing to open-source the platforms,
34:28
and in some cases, even the algorithms,
34:30
on top of that data, because there's
34:32
been a shift in the sense of where the IP exists
34:34
in that value chain.
34:35
Do you agree or disagree with that?
34:37
I think Google is TensorFlow, Kubernetes.
34:43
Those are loss leaders for them, truly.
34:45
I mean, they just want people on GCP,
34:48
and they're willing to do whatever it takes to get them
34:50
there.
34:51
Google has the advantage that it's
34:53
kind of late in competing with AWS,
34:54
but it does have an advantage that it has a much better data
34:57
team and data infrastructure than Amazon has ever had.
35:01
And if you're thinking of building machine learning
35:04
pipelines, you kind of have this inherent advantage
35:08
with Google.
35:08
So that's location data, too, eventually, right?
35:11
That's something they can take advantage of.
35:13
But they're not going to look to make those businesses, per se,
35:16
as far as I can tell, at least not relative to the value of,
35:19
come to Google.
35:20
Spend money on our infrastructure, right?
35:22
I think they're pimping a lot of that stuff out, marketing it.
35:24
And there were a lot of announcements
35:26
from Google, Facebook, et cetera,
35:27
and all of the voice-to-text and voice-to-app action products,
35:33
because none of them are good enough yet.
35:35
And I think that we've made significant advancements
35:39
in the past five years in human language processing,
35:42
whether that's voice-to-text or just text overall.
35:46
But that data set in itself is just insane.
35:52
We talk to each other with such strange colloquialisms
35:55
that it's very, very hard to train for the edge cases.
35:59
And I also think that there's still
36:02
a long way to go to be able to account
36:04
for a product that is efficient and accurate
36:08
and is better than a human boss.
36:11
So everyone sees the vision, and they push that out
36:13
because they need to train faster.
36:17
So I think they would actually take the hit in someone saying,
36:20
oh, now I'm trying to use Siri, and it actually still
36:23
doesn't work that well.
36:24
Because by doing that and by you trying to use it,
36:26
you're helping them advance their own machine learning,
36:32
because you're using it, and you're giving them
36:34
maybe a frustrated use case that they need
36:35
to understand and account for.
36:37
So they would take the hit in marketing long term
36:39
to be able to have a product that actually works really,
36:42
really well medium term.
36:44
I think your question earlier also,
36:46
when you were asking are there platforms
36:49
where user-generated data that can be sold B2B,
36:52
I mean, Mapillary is exactly that.
36:54
So crowdsourced Street View, where tens of thousands
36:57
of people are capturing images, those images
36:59
are being stitched together and sold to navigation
37:01
companies and municipalities and future autonomous driving.
37:06
So if you think about vision, autonomous driving can't
37:08
exist without vision technologies and data.
37:14
And it's really, if you take it to the next level,
37:16
an autonomous car is really a robot.
37:19
And in order for a robot, whether or not
37:21
it's a driving one or a walking one, to work,
37:23
it has to be able to see.
37:25
And in order to see, it has to analyze data
37:28
and have brilliant data scientists and software
37:30
engineers trying to figure out what that 1% signal is
37:35
in the noise of all the data that's coming in.
37:39
And so Mapillary doing that, they're
37:42
gathering the most unique data set
37:45
from cameras and cars and bicycles
37:47
and be able to sell either an image, a dozen images,
37:51
or thousands of data points via an API
37:54
to a business that wants those, whether or not
37:56
you're Nokia, a government, or an autonomous driving company.
38:02
And so I don't think we knew early on that that would
38:05
be such a big thing short term.
38:06
But making sure that you have a defensible, unique data
38:09
set that can grow over time is very exciting.
38:14
And just start small.
38:18
So before we take a few audience questions, just one last thing.
38:22
So obviously, this method or the reality
38:27
that you acquire a novel data set is key.
38:30
What's the most unique way, outside of Zuck scraping
38:34
all the pictures at Facebook, can we
38:38
come up with another example of a novel way
38:40
that you've seen an entrepreneur sort of get
38:42
started with a novel data set that you can talk about publicly
38:45
and actually endorse?
38:51
I've got another one.
38:52
Yeah, please.
38:54
Local company here, MediaChain.
38:57
So content identity tracking leveraging the blockchain.
39:01
So an image is a shared to Pinterest,
39:03
and then shared to Reddit and others.
39:06
They lose all their attribution and their metadata.
39:08
But what if you could scrape images on certain,
39:14
via multiple APIs that are open, scraping that data?
39:19
And a couple of partnerships with big agencies and maybe
39:21
museums that would love that.
39:23
So they've done free scraping to validate, now other scraping,
39:27
and creating a unique identifier to benefit
39:31
the creators and the publishers and the portals.
39:34
And that's a very unique example, very early,
39:38
a two-person, three-person team here
39:40
in Bushwick with Andreessen Horowitz and Union Square
39:43
Ventures and us invested.
39:46
I would encourage people to look random places,
39:51
a couple of companies, one called Kairos Aerospace.
39:55
They're looking for methane leaks.
39:57
It's about a $6 billion problem.
39:59
Regulatory and loss revenue perspective,
40:02
actually from a global perspective,
40:05
they can have a potentially measurable impact
40:07
on climate change, but they are assembling
40:10
some of the best maps of the United States,
40:12
North America, of methane, fixable methane leaks.
40:17
And that's by putting planes in the air
40:19
and they built their own sort of hyperspectral sensor rig.
40:22
And in the process, they're flying over large swaths
40:25
of the United States.
40:26
They're picking up all sorts of other interesting data
40:28
that may be useful for energy companies.
40:30
We also have a company called Recursion Pharmaceuticals,
40:32
which is using massively parallel screens
40:36
for monogenomic loss of function diseases,
40:40
basically genetic diseases that affect
40:45
and kill millions of people worldwide.
40:48
Just throw all sorts of compounds at them,
40:50
use machine learning to sort of featurize
40:53
the phenotype of cells and see which of these compounds
40:56
or combinations of compounds rescue them
40:58
from a visual perspective, and potentially use that
41:00
and work with pharma to use that information
41:02
to inflect the course of a trial at any given point.
41:06
But in the process, they're actually building up a dataset
41:08
just by essentially building it themselves, right?
41:13
They're not requiring users, they're just doing
41:14
vast amounts of parallel computation.
41:17
So you can look inside the body,
41:19
you can look at things from the air,
41:20
you can go into space, it's coming
41:23
in all sorts of weird places.
41:24
There's a lot of data out there in B2B,
41:26
old school B2B businesses that's not even digitized.
41:29
There's a ton of data, but you don't even have to start big.
41:31
So I mean, that's a fantastic example,
41:32
but remember that some of the beautiful things
41:34
you said earlier, just a prototype.
41:36
So when Jan-Erik started Mapillary,
41:38
and I said, I didn't believe him
41:39
in the beginning that it would work.
41:40
And so the next day, he biked around town
41:43
with his camera phone and had a unique dataset
41:46
of a very, very small town in Sweden,
41:50
put his hacked algorithms together and stitched together,
41:53
and he says, it's there.
41:54
And I said, great, now when you get three more people.
41:56
So the next day, he got three more people to do it,
41:58
and now it's a more unique dataset.
42:00
So just remember, starting in the beginning,
42:02
just like when Google started.
42:03
You know, you have to have a big vision.
42:05
You want to make the world searchable.
42:07
But they only had one product when they started,
42:10
and they empowered other search engines.
42:12
They didn't have the hundreds of products they have now.
42:15
Amazon, when they started, huge vision.
42:17
Let's build a search engine as big as the Amazon.
42:22
But does anybody remember the first product
42:23
that they started with?
42:25
I mean, it's a fraction of what they are now.
42:28
What kind of books?
42:30
It was, no, it was research books.
42:33
Just because it was the only deal they could get.
42:36
That was the first thing.
42:37
They had a huge vision, but they weren't selling soap
42:40
and mops and crap like that in the early days.
42:42
And so when we think about what you guys want to do
42:45
or what dataset you need, or just start,
42:47
like Tim was saying, I've got a problem.
42:50
I want to solve that.
42:51
Let's prove it, and then we'll get more.
42:54
At least on the consumer side,
42:55
I would answer that in reverse.
42:58
In a world in which individuals want their data back,
43:04
and the holy grail of unique datasets
43:08
and network effects and data network effects
43:10
actually potentially becomes less valuable
43:12
because the consumer, the individual online
43:16
wants to have access to their data,
43:20
decide when they're gonna share it,
43:21
decide that they don't want to share it at all
43:23
and have more security.
43:24
Generally, from Amazon, Facebook, Google, et cetera,
43:28
they grew because people were willing
43:30
to surrender their information,
43:32
their data in exchange for information.
43:35
And that was the value.
43:36
And if that starts to change, at least to certain degrees,
43:39
or if certain people say,
43:40
I want to know exactly who's got access, when,
43:43
and I want to know what I'm getting back for it,
43:46
or I want to hold it off at certain points,
43:48
the value of a mode around data
43:51
actually becomes quite a lot less.
43:55
So what is the mechanism for giving that back to individuals
44:00
and allowing them to participate more
44:03
as an active and transparent party in the exchange of data?
44:09
My inclination is that that has to do with the blockchain,
44:12
but I would challenge everybody else
44:14
to think about that some,
44:16
because I think we're quickly approaching
44:17
a world like that.
44:18
And that's actually becoming, I think,
44:19
particularly relevant in Europe.
44:23
We invest in a company that tries to basically
44:25
help you figure out data sovereignty issues.
44:27
You're a business and you have data
44:29
that may belong to a German,
44:31
and then it somehow ends up in, I don't know, Taiwan.
44:35
The penalties that foreign governments are imposing
44:38
for data misuse, even if it's accidental,
44:43
can be phenomenal.
44:44
And so you're seeing a regulatory push
44:46
against companies like Facebook, like Google,
44:48
to ensure that data is highly controlled.
44:51
Whether that's good for the long-term
44:53
fate of civilization, I don't know.
44:55
But it's certainly gonna propel people, I think,
44:59
to build businesses that sort of keep track of that.
45:03
And then you can potentially sell your data
45:04
if you know where it is, right?
45:07
Great.
45:08
Let's take a couple questions
45:09
before we go for beers downstairs.
45:12
Oh, thanks.
45:14
Hey, let's do it.
45:15
I'm just gonna grab you a mic.
45:17
Right here, Adam.
45:21
Panel, if you repeat the questions
45:23
for the video, that'd be great.
45:26
Sorry, I'll start.
45:27
So we saw a lot of trends in startups
45:30
from social to mobile, then it was machine learning.
45:34
Everyone was just machine learning
45:35
plus something startup.
45:36
Then it became, what data do you own?
45:38
And I feel like, with special autonomous cars,
45:42
with special autonomous cars,
45:43
soon the government will force us
45:45
to actually release where the car is,
45:46
and you have to kind of release
45:47
some of your information,
45:48
because you have to do combined data across systems.
45:51
What do you think is next?
45:53
Do you think data's gonna always be the owner,
45:56
or will that be the next bubble
45:57
where no longer is data,
45:58
or what is next for technology?
46:03
What's next for data technology?
46:06
It doesn't have to be data.
46:07
Data's currently the hottest thing, but what's next?
46:09
Depends if Trump wins or not.
46:11
All bets are off in that case.
46:13
I had to throw that out there.
46:15
You had to go there, didn't you?
46:17
That's one of the bigger issues facing our industry
46:20
and every other industry right now.
46:22
So who knows?
46:24
I think one of the things we're seeing
46:25
is that the function layer,
46:27
I kind of think of Amazon Lambda as a metaphor
46:30
for what's happening in a bunch of different places,
46:32
which is that you either have your processing layer
46:34
or your capturing data,
46:36
or you're making your algorithm smarter,
46:38
and it sort of lives everywhere.
46:40
We don't have a lot of examples of companies
46:42
that look like this yet
46:43
where you don't actually care what the API looks like.
46:46
So the only, I used this example earlier,
46:47
but the only analogy I can come up with
46:49
is Uber is cars driving around.
46:53
That's the depth of their service,
46:54
and they don't care about how you connect with them.
46:56
It's just like, how do I get this atom
46:58
from point A to point B?
47:00
Whether that's on your Apple Watch,
47:01
or it's on your phone, or it's Alexa,
47:04
it doesn't matter, right?
47:05
Their service lives somewhere else.
47:07
I can imagine worlds where you're taking
47:09
specific algorithms, like in a future world
47:12
where we all have our own data, we all control it.
47:14
We're pulling algorithms to run specific computations
47:17
on our data.
47:18
I think it's like, that's why it kind of reminds me
47:21
of Lambda, because you just sort of pull
47:22
this one specific function.
47:24
It's like object-oriented programming
47:27
has functions inside of the objects, right?
47:29
It just kind of fixes it out the other way,
47:30
where now we go back to having individual functions
47:32
that accompany us, and you apply it to your objects.
47:35
So I think it's similar, like the elimination
47:37
of a lot of the middlemen, the aggregators,
47:40
because that's what a lot of the big winners,
47:42
like the biggest in this cycle have done,
47:44
is they got people to give them their data,
47:46
they aggregated it, and they win,
47:47
because now they're so big that you can't compete
47:49
with Google, can't compete with Amazon,
47:51
can't compete with Facebook, that social commerce search.
47:54
And it continues to get better as more people use it.
47:56
But a couple more leaks, and a couple more,
47:59
Like, you know, absolutely horrific losses of that data or misuse of it.
48:07
And people say, I don't want to be blindly tracked anymore.
48:09
I don't want to be advertised things that I don't want.
48:11
And I want to know exactly when you're going to do that.
48:13
And I want to tell you when you can.
48:16
I think that's a pretty profound shift.
48:18
And that would be elimination of any lack of clarity, but also a lot of middle firms.
48:24
Uber is not a ride-sharing company.
48:27
Don Tapscott wrote the book Blockchain Revolution.
48:29
He gave a talk yesterday at QuarkBench.
48:31
He says that Uber is not a ride-sharing company.
48:33
It's an aggregator also because they sit in the middle.
48:35
And they become worth whatever bajillion dollars they're worth because they sit there.
48:40
And they actually control the service.
48:41
If it was more peer-to-peer, it would actually be sharing.
48:45
And everyone would be in control of sort of their person and the person that they give access to online.
48:55
I think that puts way too much – gives way too much credit to humanity, honestly.
49:00
I understand.
49:01
I share your vision on that.
49:02
But I just – I think –
49:03
Is this the extension of next week's issue?
49:06
Well, I don't know.
49:07
But I just don't think – I'm skeptical that people care that much about –
49:13
I mean we invest in InfoSec, right, as a core category of one of the things we do.
49:18
And we used to think, well, the target breach followed by this breach followed by that breach.
49:23
We used to think at some point somehow the world would sort of come to an end.
49:27
But people have become inert to it.
49:29
And so if you tell them their data is breached, they may theoretically care.
49:33
But I'm skeptical that aside from bureaucratic sort of overhang and aggressive regulation,
49:41
actual people fundamentally feel the impacts or care about their data being diffused and not under their ownership.
49:51
That's a philosophical –
49:53
I agree.
49:54
And I think it's what comes next that more drastic things would have to happen.
49:58
But I feel like there is a path where it could.
50:03
And then it's either people serve themselves up to a world like the movie WALL-E,
50:09
if you've ever seen that, in the end, where human is kind of just like –
50:12
it's a play for efficiency which becomes laziness and kind of a different kind of existence.
50:16
Or you decide to take action and a different type of technology comes in the way.
50:21
So I understand what you're saying.
50:23
So obviously you guys all know my thesis, my focus these days, but I'll say what I – the answer to your question.
50:28
The next thing that's going to be larger than Internet of Things is called Internet of Eyes.
50:33
And everyone here will have multiple cameras in their fabric, on their eyeglasses, in these lights.
50:40
And I believe it's actually going to help improve our lives and society.
50:44
There's also, on the flip side, everybody says, what about privacy?
50:47
And privacy is dead.
50:49
So figuring out that data set from all of these different things, from robotics or thermal or anything,
50:58
and there's phenomenal businesses and phenomenal data sets that we haven't even started to gather.
51:04
Well, the Internet of Eyes botnet brought down, like, everything two weeks ago, right?
51:09
I know, and it validated that sector, so it's only going to get more interesting.
51:13
Apparently there's more Internet-connected security cameras up there than any of us knew
51:16
because it was enough to DOS half of the East Coast.
51:19
That's true. That's Black Mirror Season 2, Episode 3, something like that.
51:22
It's actually Person of Interest.
51:24
Just watch Black Mirror if you want to know what's coming.
51:26
It's one of my favorite series. It's horrible acting, but it's Person of Interest.
51:29
They know everything, and to the point earlier, it's images, voice, and text,
51:35
and understanding that signal for different use cases.
51:38
I think a lot of it, a camera is going to be able to know your heart rate right now
51:42
compared to your heart rate. That's huge.
51:45
That's terrifying.
51:47
Have it terrifying as whatever you want, but I think it's a huge opportunity.
51:51
And everything from the first industrial age, it was terrifying for many people.
51:55
They were not usually investors.
51:57
So it's interesting that you're terrified with what the reality of the future is going to be.
52:01
Let's do one more, and then we can chase these guys downstairs for questions.
52:07
What about corporations and governments?
52:09
They're not going to like that answer that privacy is dead.
52:12
They don't want to get hacked, that cybersecurity is more important.
52:16
It seems like those are actually at odds with each other.
52:20
I think you're right. They are at odds with each other.
52:23
Privacy is very important to me, but the reality of all what we're doing,
52:29
like you said earlier, we're doing things for an end goal.
52:34
I don't think most people are thinking that, okay, I'm willing to give my data away to do this.
52:38
They're just doing it and after finding out about it.
52:41
And so that's what I mean by privacy or what we know privacy is not going to exist any longer.
52:47
I am not supportive against killing it, and I think there will be many iterations of protecting it going forward.
52:53
I actually think somebody is going to come up with a tool that says,
52:56
Evan, we just noticed five-year images popped up on the Internet from the security cameras at Grand Central Station.
53:01
Do you want me to kill them or take them down?
53:04
I mean, that's the flip side of going against privacy.
53:09
My sense with your comment on privacy being dead is that it's de facto.
53:14
It's sort of people have made – we all have made this decision.
53:18
I was thinking to myself, like when we first started talking, I had forgotten there was a camera.
53:23
And it was like, okay, but do I care, actually?
53:26
I got gas. Maybe that's good. Maybe it's bad. Like, I don't know.
53:31
I go and I do an Internet search, and then the best result comes up because they've been following me around on the Internet all day.
53:36
I was like, well, is that bad? Is it good? Like, I just don't care.
53:39
I think that's what people figured out.
53:41
I have my little cousin's 13, and she was showing me her Snapchat.
53:45
And she said, I take pictures of my Snapchat as my primary phone.
53:50
I was like, you don't use the regular phone?
53:52
She says, yeah, I don't want the photos on my camera to take up space.
53:55
So that's like it's a feature to her because she values – I call it her privacy, right?
54:02
Her personal photos, right?
54:04
The privacy of those photos is so little that it's not a function of like, oh, I now have to make this exchange.
54:12
It turns out that it doesn't matter anymore to a lot of people.
54:16
And I'll just follow up to your point.
54:19
Actually, I think it is the job of investors to be equally excited and terrified about the things that we're investing in
54:27
because when it comes to machine learning, we're actually investing in things that could potentially result in massive labor dislocations
54:34
that will destabilize things further and give political space for potentially an authoritarian to come into role.
54:42
I can't predict all of those things, right?
54:45
But it's important to weigh the pros and cons.
54:50
I mean, for example, OpenAI, sort of spun out of Y Combinator and funded through some ex-Google folks,
54:57
founded sort of explicitly because you're seeing this blurring of line between academia and research,
55:03
but it's mostly you're sending a gravity shifting to corporations, to Google, to Facebook, to Baidu and other places.
55:11
And ILIA had wanted, let's build an organization that doesn't have, say, the corporate profit motive as its primary directive
55:20
given that some of this research is going to be consequential.
55:25
And it's important to sort of start planning, I think, the ethics around this,
55:30
as people are doing now with CRISPR and genetic editing,
55:33
thinking very sort of stringently about what are the consequences of all these actions?
55:38
And how do we build sort of best practices and norms so that we at least try to manage the future on the downside?
55:47
I agree you have to balance, but I think the main point from mine is, and it relates to your question,
55:52
I think over time there's kind of two camps.
55:55
There's the same thing. Do we ask permission or do we ask forgiveness?
55:59
of these issues get pushed because of technological advancements balanced with human nature,
56:07
okay? And I'm more on the part of, I'm not sure the answer. I'd rather protect it to
56:12
improve society. But when the technology goes forward and creates new solutions to problems,
56:18
it'll create more problems. And then there'll be more solutions. And so I think you're right,
56:23
you have to balance the, you know, what ifs and the scary to figure it out. And there's
56:27
probably solutions on both those. But I think that I'm more of the ask forgiveness and validate
56:36
stuff. And let's prove there's a market out there and people want it. And there's a reason
56:41
people want it. So then the laws, most of all the laws, are legacy laws that might have to
56:47
evolve when they're pushed to evolve. When you guys co-invest in your first company together,
56:53
let me know. What's that? I want to see you guys co-invest in your first company together.
56:56
Okay. I can't wait.
56:59
Thank you for a lively, fascinating panel. Each of you, this was awesome.
57:03
Thanks for the questions. Chase these guys downstairs.