0:01

You're listening to Gradient Dissent, a

0:04

show about making machine learning work

0:06

in the real world, and I'm

0:08

your host Lucas B. Wald. Christopher

0:10

Alberg is an old friend and

0:13

the CEO and founder of

0:15

Recorded Future, an AI cybersecurity

0:17

startup that was founded in

0:19

2009 and sold recently to

0:21

MasterCard for $2.65 billion. Previously

0:23

he was the founder of

0:26

Spotfire information visualization

0:28

startup. And we talk about

0:30

AI applied to cyber threats,

0:32

war in Ukraine, and compare

0:35

notes on being two-time founders.

0:37

I hope you enjoy this one.

0:39

All right, so Christopher, welcome

0:41

to Grady the said. Thank you, Lucas.

0:43

Thank you for having me. So you

0:45

have like my favorite kind of guest,

0:48

which is someone who's taking AI and

0:50

applying it to an important domain that

0:52

I don't understand super well. So I

0:54

think for our audience, you're going to

0:56

have to tell us about your company

0:58

recorded future and the problems that you

1:01

solve. So maybe we could start with

1:03

that. Yeah, I know for sure, you know, you

1:05

could sort of, I wouldn't use this

1:07

description. widely, but in this context, maybe

1:10

you could say they were doing AI

1:12

for intelligence or intelligence for AI, if

1:14

you want. We're an intelligence company, threat

1:16

intelligence, we try to get our hands

1:19

on everything bad that's going on in

1:21

the world, but maybe more importantly in

1:23

the sort of the internet world that

1:26

I like to say that there's sort

1:28

of this opportunity that came along to

1:30

think about the internet as a... an

1:33

incredible intelligence sensor that happened at the

1:35

same time as the world have sort

1:37

of slowly migrated onto the internet and

1:40

and as that migration is happening it's

1:42

sort of going to the point where

1:44

actually the world is actually becoming a

1:46

reflection off the internet crazily enough and

1:49

so in that we've done hard work

1:51

over the last 10-15 years to try

1:53

to organize this data collected you can

1:56

imagine all kinds of crazy data we've

1:58

had to use a lot of you know,

2:00

call it machine learning, big data, analytics,

2:03

now AI, pick your favorite words. to

2:05

sort of make sense that a lot

2:07

of data, a lot of that stuff

2:10

got started in natural language and processing,

2:12

these sort of things, but then all

2:14

kinds of other data. And in the

2:17

end game, it's all about producing great

2:19

insights. These could be insights about geopolitics,

2:21

they could be insights about cyber threats,

2:24

that's sort of where our main stuff

2:26

is, it could be insights that support

2:28

a warfighter, it could be insights of

2:31

all kinds, but all in this sort

2:33

of... bad news domain and the threat

2:35

Intel domain and we built a good

2:38

business there. Sometimes I summarize it by

2:40

saying it's sort of the Bloomberg for

2:42

intelligence. And so I'm sure that, like

2:45

a lot of our guests, maybe the

2:47

most interesting things that you've done, you

2:49

can't say. But is there any particular

2:52

insight that you pulled out that you're

2:54

proud of that you can talk about?

2:56

Yeah, you know, it was through the

2:58

years, there's been wild things, you know.

3:01

going back to in 2016 when the

3:03

election stuff sort of was the first

3:05

time around sort of thing we found

3:08

some Russian dude who was selling access

3:10

to the electoral access commission or electrical

3:12

yeah whatever it's called you know incredible

3:15

stuff that was probably the first sort

3:17

of that moment where you're just like

3:19

wow yeah what's going on how did

3:22

you talk about how did you figure

3:24

that out that's amazing like what what

3:26

how does that work in that case

3:29

it's this sort of wonderful dark web

3:31

world where that back then was in

3:33

these forums that still exists. A lot

3:36

of that has now moved on to

3:38

telegram, if you want. That's sort of

3:40

where a lot of the bad news

3:43

of the world now happens. But some

3:45

guy there had broken into this thing

3:47

called Electoral Assistance Commission, EAC, and he

3:50

had packed it in good ways using

3:52

a sequel inject, extracted a bunch of

3:54

information, but maybe more importantly. He went

3:57

back out and said, I'm selling access

3:59

to this. Who wants to get the

4:01

access? And we found it before somebody

4:04

else. and actually bought the axis crazy

4:06

enough. Really? Yeah. Wow. And took custody

4:08

of it and then went back to

4:10

the government and said, you may want

4:13

to make this thing go away. And

4:15

yeah, it was a little bit of

4:17

a wild thing. I remember when this

4:20

guy, a colleague, fantastic colleague Andre calls

4:22

me in the middle of the night

4:24

and just says, you know, something not

4:27

so great is going on. What should

4:29

we do? And a long time ago,

4:31

but good story. Okay, you're like instantly

4:34

pulling me off my script, which I've

4:36

been trying to stay a little more

4:38

on script But I can't help myself

4:41

like what what is the dark web

4:43

like is like is this like forums

4:45

that people log into is this like

4:48

on tour like what actually is that

4:50

it is you know like the and

4:52

it's a portion just to be clear

4:55

about and it's it's changed recently we

4:57

can come back to this and talk

4:59

about telegram, but no this stuff is

5:02

still there there there are these forums

5:04

and it turns out if you're in

5:06

the business of buying and selling information,

5:09

you need a place to buy and

5:11

sell it. You can't just sit on

5:13

it, then you're not going to make

5:16

much money. And these guys have, the

5:18

only interest really is in making money,

5:20

different from spies and government spies. But

5:22

these guys, they want to make money,

5:25

so they need places to buy and

5:27

sell. And some of them still exist

5:29

20 years later, 25 years later, crazy

5:32

enough. So they are, most of them

5:34

are on tour, some of them are

5:36

not, and they're sort of essentially marketplaces.

5:39

and we keep a close eye on

5:41

them if we put it that way.

5:43

So today, right now, somebody as of

5:46

an hour ago or a couple hours

5:48

ago was selling access to a particular,

5:50

I'll just call it big software company

5:53

servers and it's crazy creating havoc right

5:55

now, right this moment. And we got

5:57

our hands on all this data. We're

6:00

trying to upload it into our system,

6:02

generating alerts. the clients and this stuff

6:04

is never ending. So yeah. So like

6:07

so literally right now someone's selling access

6:09

to a company's servers. Many of them

6:11

yeah no the the it's software that

6:14

is installed because something is going on

6:16

in real time I'll leave it sort

6:18

of a little bit fuzzy but it's

6:21

a big prominent software company and they're

6:23

selling access to this yeah. And who

6:25

would typically buy it like do you

6:28

pretend to be like someone that wants

6:30

to buy it to to learn, but

6:32

like, who even would want that access?

6:34

It can be, you know, so typically

6:37

these guys will sell access. That's one

6:39

thing. And, you know, if you think

6:41

about the guys who want to buy

6:44

access, there could obviously be somebody who

6:46

wants to get after something very special.

6:48

But what happens in this world is

6:51

that it's very specialized. You have people

6:53

who will... Then, for example, sell ransomware

6:55

software, or rent access to ransomware software.

6:58

You have people who help you get

7:00

the rent, execute the ransomware. You have

7:02

people who then... handle the actual Bitcoin

7:05

flows all the way to moneymules, a

7:07

highly specialized world of criminals there because

7:09

they are criminals. A large portion of

7:12

them are in Russia, not all of

7:14

them. And in this case, it's probably

7:16

somebody who has realized that he can

7:19

sell access to all these different places

7:21

that opens up all kinds of opportunities

7:23

for ends somewhere. I'm sort of guessing

7:26

in this case since it's happening real

7:28

time, but it's a pretty interesting world.

7:30

It's an interesting world, I guess, but

7:33

I'm totally not aware of this. want

7:35

to buy this data. Is that other

7:37

criminals or is that like other criminals?

7:40

Other criminals. Probably the ransomware actors. So

7:42

there is a whole slew of these

7:44

ransomware actors. So there's guys who write

7:46

and operate ransomware software. They then franchise

7:49

out the use of this ransomware software

7:51

to other actors and they'll use it

7:53

execute and in return for holding on

7:56

to say 80% of whatever money they

7:58

get on the ransomware. action, the operator

8:00

gets another, get the other 20% and

8:03

it becomes one of the best business

8:05

models ever operated on planet Earth. And

8:07

presumably if someone's selling something illegal, they

8:10

want to make sure they're selling it

8:12

to a criminal and not you. And

8:14

presumably they're worried about selling it to

8:17

you, so how do you convince them

8:19

that you're not going to turn them

8:21

in with something like that? So, you

8:24

know, that's our job. Okay,

8:27

love it. All right. Is there any

8:30

AI involved in that or is that,

8:32

that seems like a totally different. There's

8:34

AI in, because it turns out that

8:37

these, it's, I wouldn't call it big

8:39

data, big forums, big, but they're tricky.

8:41

You had, they might be locked down

8:43

in all kinds of different ways. So

8:46

you need a lot of trade craft

8:48

to get in and it's a combination

8:50

of human social engineering and smart scraping

8:52

if you want. there could be six

8:55

layers of access to some of these

8:57

places. So that's pretty complicated sort of

8:59

thing. So it's a combination. I wouldn't

9:01

necessarily call it AI. But then at

9:04

the same time, these scrapers need to

9:06

operate in human-like behaviors. So I like

9:08

to joke that they sort of might

9:10

actually pass a very limited touring test,

9:13

because they need to sort of operate

9:15

as if I'm a cyber criminal. I

9:17

don't know. Is that a touring test?

9:20

I guess it's a one kind of

9:22

touring test. by whose definition you go

9:24

with by it. But yeah, I guess

9:26

it is. Yeah. Yeah, I guess there's

9:29

a human looking at it. It's for

9:31

sure. It's during tests. Yeah. No, no.

9:33

So artificial criminal behavior. So I guess

9:35

that's, you know, that's that's a touring

9:38

test of just a special kind. Okay,

9:40

well, can I get a back on

9:42

script that I'm sure we're going to

9:44

come come back to this? Yeah, one

9:47

of the things that we have in

9:49

common is that, you know, we're both

9:51

repeat entrepreneurs and you actually, you know,

9:53

you founded Spot Fire back in, I

9:56

think, 1996 and sold it 11 years

9:58

later. And then you found a recorded

10:00

future a little bit after that. And

10:03

I was wondering if you could go

10:05

back in time to when you started

10:07

recording future and what you were thinking

10:09

at that moment and what the insight

10:12

was that led you to start recorded

10:14

future. Yeah, that's a great question. So

10:16

Spotfire was data visualization. We started that,

10:18

sort of that, I'm aging myself here,

10:21

but in the early 90s, working for

10:23

my PhD on how to visualize large

10:25

data sets and kind of a. predecessor

10:27

to Palantir and a whole bunch of

10:30

other sort of things, not to make

10:32

any claims, but that we were the

10:34

first by the means because we're not,

10:36

but we did a nice job with

10:39

certain kinds of data and so on.

10:41

So then we sold that to a

10:43

company in Palo Alto, Co, Tipco, worked

10:46

out great, and we were very excited

10:48

about that. But then sort of the

10:50

idea struck me that never sort of

10:52

all about visualizing what was in an

10:55

Oracle database or an Excel spreadsheet or

10:57

what have you. That sort of data.

10:59

And so we were, it sort of

11:01

struck me, I was on the treadmill

11:04

running and literally while we had signed

11:06

the deal to sell Spotify, but we

11:08

hadn't closed yet. So it was like

11:10

in this weird in-between timing. And it

11:13

struck me that, oh, wonder if we

11:15

instead of thinking about analyzing what's in

11:17

an Excel spreadsheet, what if I could

11:19

hook up an analytical engine straight to

11:22

the internet? So, you know, use that

11:24

sort of approach to things. So, and

11:26

now the internet is at the surface

11:29

level, at least, is mostly human-produced text.

11:31

So now you had to deal with,

11:33

look at for entities and events and

11:35

these sort of things out of text

11:38

and try to make sense out of

11:40

that and organize it in a way

11:42

that you could do analysis. And yeah,

11:44

that was sort of the inspiration. That's

11:47

how we got into it. Interesting,

11:49

and the neighbor recorded future is super

11:52

evocative. Did that come to you from

11:54

the start? No, not maybe immediately. It

11:56

was one of my co-founders here, Eric.

11:58

very clever guy. And the idea from

12:01

the beginning sort of was this idea

12:03

that we would find these future time

12:05

points in text, you know, like Zee

12:08

Jing Ping is traveling to Moscow on

12:10

Friday. And if you could actually keep

12:12

tabs of everything that's known in the

12:15

world from all the sources, we should

12:17

get a good picture of where Zee

12:19

Jing Ping is. to use an extreme

12:21

example. And from the beginning, Eric suggested

12:24

we should call it, I guess, recorded

12:26

time, I guess in one of the

12:28

Shakespeare's at Macbeth, until the end of

12:31

recorded time, was sort of that thing.

12:33

Then of course, the domain name, recorded

12:35

time, was taken, it became recorded future,

12:37

which was probably a better name to

12:40

begin with. But so, yeah. And it

12:42

fits well with the idea that, you

12:44

know, most secrets are known. It's just

12:47

got to get to where they are

12:49

where they are. Let's get to all

12:51

of them. It's interesting, you know, it

12:54

seems like a little bit of an

12:56

entrepreneurial Anti-pattern here of like it's not

12:58

it doesn't sound like you're really starting

13:00

with like a specific customer pay import

13:03

and working backwards It sounds you kind

13:05

of started with your your sort of

13:07

interest like was it obvious that intelligence

13:10

was going to be like a big

13:12

customer this because it does seem like

13:14

as broadly as you would find it

13:17

like almost any organization would benefit from

13:19

this kind of analysis Totally Totally. And

13:21

that's probably my weakness, but maybe it's

13:23

also a decent strength. But you know,

13:26

so Spot Fire started as we could

13:28

visualize any data set. In that case,

13:30

we stumbled on to visualizing and analyzing

13:33

pharmaceutical discovery data, very super high-end super

13:35

valuable application, and it worked out great.

13:37

Then in that, we actually ended up

13:39

doing a lot of counterterrorism work in

13:42

the sort of CT space in intelligence.

13:44

I've always loved that world and we've

13:46

done a lot of good work in

13:49

that. So when we started to record

13:51

a future, to your point, we knew

13:53

that there was an application in this

13:56

sort of strategic foresight or whatever. Turns

13:58

out that we ended up in the

14:00

cyber world with it, but no, it's

14:02

certainly more of... inventing a big ass

14:05

hammer and trying to figure out where

14:07

you could apply it rather than the

14:09

other way around but you know isn't

14:12

that how a lot of good stuff

14:14

comes up I don't know what What

14:16

you would say, I think a lot

14:19

of good ideas come that way. Well,

14:21

look, I mean, you've been incredibly successful.

14:23

So I think anything, you know, people's

14:25

strengths and weaknesses often always connected. I'm

14:28

just curious about your process. Like, you

14:30

know, when you started this company, did

14:32

you have like a big list of

14:35

possible use cases and you kind of

14:37

ran down the list talking to people?

14:39

Or how did you, how long did

14:42

it take you to kind of get

14:44

to these like specific use cases and

14:46

how did you approach that? We also

14:48

thought about commercial intelligence and sales intelligence

14:51

and lots of other sort of things,

14:53

you know, for sure. And there's people

14:55

who've built similar type companies in lots

14:58

of different spaces. We had a long

15:00

list. I think I still have those,

15:02

whatever you want to call them, presentations,

15:04

you know, you can imagine the first

15:07

venture presentations. Yeah, yeah. We quickly went

15:09

to Incatel. and Google Ventures barely existed

15:11

at the time, but ended up taking

15:14

money for Minketel and Google Ventures back

15:16

in 2008-9 or something like that, and

15:18

sort of honed in on this Intel

15:21

thing pretty early, but could have gone

15:23

many other places too, for sure. So

15:25

one question that I always get asked,

15:27

that I feel like I never have

15:30

a good answer to, but I think

15:32

I'm going to turn around to you

15:34

and ask it. to you is you've

15:37

now done these two companies and you've

15:39

done both for quite a long period

15:41

of time. What did you do differently

15:44

in your second company? Like what did

15:46

you take away as a second time

15:48

entrepreneur? Oh. That's the sort of Peter

15:50

Thiel question that sounds like you know

15:53

like that. I only bring that one

15:55

up because literally every podcast I do

15:57

I always get us a question and

16:00

I never know what to say so

16:02

I think maybe all of a better

16:04

response to me. The sort of the

16:06

if I started with the bad side

16:09

you know so with the first company,

16:11

it took us a little time and

16:13

we stumbled on, we started very generally

16:16

and then we focused on pharmaceutical discovery

16:18

and killed it in that domain and

16:20

then worked from there. That domain ended

16:23

up being fairly limited and probably limited

16:25

the outcome of the company at some

16:27

level. We sold that for $195 million.

16:29

Nothing to sneeze at, but you know.

16:32

Congratulations, man. Oh, no, but these days,

16:34

people are, yeah, whatever. People are like,

16:36

it's... To put it in terms like

16:39

now, that would be like a $5

16:41

billion exit in 2025, just for the

16:43

younger listeners. Man, man. Yeah, so then

16:46

here we, I think we, like the

16:48

failure point, we thought we could do

16:50

multiple application errors. We for multiple years

16:52

ran an intelligence track and a quand

16:55

trading track to create quand trading signals.

16:57

And not, and we had a. serious

16:59

sea investor came aboard and he you

17:02

know you shouldn't listen too much to

17:04

your board but but this one guy

17:06

he just said kill that and I'm

17:09

like you're actually right right we should

17:11

just kill it and we killed it

17:13

and that was a great sort of

17:15

like freed us and we could just

17:18

run and and it is sort of

17:20

already started withering a little bit but

17:22

so so That was certainly that we

17:25

got, what do you call it, when

17:27

you think too big, or you know,

17:29

like we thought too highly of ourselves.

17:31

In terms of what we did differently,

17:34

that was better. We probably picked a

17:36

bigger problem. That was great. This cyber

17:38

intelligence turned out that we ended up

17:41

sort of riding a massive wave. This

17:43

investment in cyber security that have sort

17:45

of gone through the last 10, 15

17:48

years has been incredible. And even though

17:50

we certainly dissolve a subset of it,

17:52

you know, by writing a big wave,

17:54

you can make a lot of mistakes.

17:57

You know, you've seen that in AI

17:59

also, like the wave is big enough.

18:01

It allows for a lot of mistakes.

18:04

And maybe also... No, that would probably

18:06

be the answer to that. I'm sure

18:08

there's more, but. No, actually great segment

18:11

to my next question, right, which is

18:13

that, you know, you've been in this

18:15

massive wave of, not just, just like

18:17

cyber security, but the application of ML

18:20

and AI throughout your kind of 15-year

18:22

arc and recorded feature. I'm really curious

18:24

how the availability of data and the

18:27

new applications of ML and I changed

18:29

recorded futures business over the time that

18:31

you operate it. It was good. So

18:33

there is sort of two aspects. One

18:36

is the data and then there's like

18:38

what you now can do with data

18:40

here. So we started off with, you

18:43

know, the stuff that we called AI

18:45

and oh, oh, eight or nine, ten

18:47

sort of thing. We wrote our first

18:50

entity extractors and event extractors with like

18:52

lots of if-then- else, else just stacks

18:54

of if-then- else statements, sort of, they.

18:56

We also wrote a lot of good

18:59

test cases to test whether it was

19:01

right or wrong, but this test... text

19:03

lead to this, that sort of thing.

19:06

Though obviously ridiculous from a point of

19:08

view of like what people are doing

19:10

now and we've probably gone through three

19:13

generations even for those entity extractors and

19:15

we use some commercial stuff now we

19:17

sort of own all that internally and

19:19

it's been a great journey with that

19:22

and now that's all based on these

19:24

sort of models as you could expect.

19:26

The availability of data for sure have

19:29

sort of helped through that. We used

19:31

to have entity extractors for each language.

19:33

Now that that is one coherent model

19:35

that sort of spans, I don't know,

19:38

1530 languages and it's amazing how it

19:40

cross learns between languages that are not

19:42

just inside into European languages, but it's

19:45

like cross languages. It's sort of bananas

19:47

that this stuff even works. It's mimicking

19:49

human brains somewhere. Somewhere there is something

19:52

wild going on in that. And so

19:54

that's sort of super interesting. The other

19:56

big piece, probably, sort of could spend

19:58

a lot of time on this, but

20:01

is how that was one thing to

20:03

do feature extraction if you want, or

20:05

like, yeah, feature extraction, data extraction, out

20:08

of content. Now with all this generative

20:10

stuff, you could obviously sleep. So in

20:12

intelligence, there are sort of two aspects

20:15

to it. You think about if you

20:17

run fancy intelligence agency, you've got the

20:19

collectors, the James bonds running around in

20:21

the world, and you've got the analysts

20:24

who sort of putting stuff together. The

20:26

collection part is sort of what I

20:28

first described. The second part of writing

20:31

to, you know, everything from the simple,

20:33

summarize what happened in Somalia last week,

20:35

to what are the second order implications

20:38

of what happened in Somalia last week,

20:40

you know, like so from basic questions

20:42

to more advanced, or summarize the second

20:44

order implications of what happened in Somalia

20:47

and get a report written in Arabic

20:49

that I can share with my partner

20:51

in Egypt about that. Those are like

20:54

pretty juicy, juicy, heavy questions. And by

20:56

the way, do that every week for

20:58

me and deliver me at 8 a.m.

21:00

You know, now you take an fair

21:03

amount of work and stacked up there.

21:05

Now, and we do that. And it's

21:07

just, it's just happens. It's sort of

21:10

mind-boggering, isn't it? So, so if somebody

21:12

would have told me that. five years

21:14

ago, I would not have believed it.

21:17

Maybe, maybe three years ago, but it's

21:19

it's pretty mind-boggling that this works, and

21:21

it works with a mix of text

21:23

and images, and actually we collect a

21:26

lot of other weird data, sort of

21:28

more net flow data, malware data, very

21:30

technical data, that we spend a lot

21:33

of time on how to wrap that

21:35

in human language to the LLMs can.

21:37

handle that sort of data too. And

21:40

even there it works. So, you know,

21:42

it's a, I'm full off the chair

21:44

here and, you know, like, maybe that's

21:46

because I'm dumb, too dumb to understand

21:49

it, but it's pretty mind-boggling. I totally

21:51

agree for what it's worth. Actually, I

21:53

wanted to ask you. When ChatGPT came

21:56

out, I think it was two or

21:58

three years ago, was that like a

22:00

the same like watershed moment in the

22:02

Intel community that it was here at

22:05

Silicon Valley or did it take longer

22:07

for people to realize the implications? I

22:09

think, you know, so first even inside

22:12

recorded future, it took me just like,

22:14

I remember I was sort of debating

22:16

with, you know, my co-founder Safant Reebad

22:19

sort of like, you know, are we...

22:21

coming to AI sort of yet another

22:23

AI winter? Or is it about to

22:25

take off? And I'm like, ah, it

22:28

feels like it's done. That was Christopher

22:30

Commons and stuff. And I was like,

22:32

it's going to take off, buddy. And

22:35

I'm like, he was, of course, very

22:37

right. And I was very wrong. So

22:39

to begin with. But then I think

22:42

there were people in DC who ran

22:44

with this very cool, in a very

22:46

cool ways. I don't want to talk

22:48

about them, put names to things and

22:51

so on. But there are areas in

22:53

there where people have been extremely forward

22:55

leaning and built incredible stuff. And we've

22:58

had the good chance to spend time

23:00

with those people if they listen now,

23:02

they'll know who I'm talking about. And

23:05

where we do comparison of notes on

23:07

what we're doing, what they're doing, and

23:09

you know, sometimes their benefit is if

23:11

they have access to very special data,

23:14

we have access to the internet in

23:16

a way that we think. maybe other

23:18

people don't and we try to collaborate

23:21

on some of that but no I

23:23

would say in general the government is

23:25

never the best that seeing advantage of

23:27

new tech we're not never that's not

23:30

true because sometimes they put crazy satellites

23:32

in the sky and stuff but in

23:34

general not always the best adopter of

23:37

tech but in this case some parts

23:39

of the government was pretty pretty amazing

23:41

so okay another big moment that I

23:44

think happened around the same time as

23:46

GPT was the invasion of Ukraine. And

23:48

I know that recorded features, this is

23:50

an important moment for recorded future. Could

23:53

you kind of think me that back

23:55

to that moment inside of recorded future

23:57

and what you were doing and how

24:00

you responded? Yeah. So I would say

24:02

a couple of different things. We have,

24:04

they were. customer from before in some

24:07

areas and you know we we have

24:09

probably 47 different countries around the world

24:11

that use us in in some sort

24:13

of national capability and all sort of

24:16

in the West plus plus or the

24:18

extended West are sort of a weird

24:20

way of describing the world especially with

24:23

those who live in the East but

24:25

you know it's sort of a got

24:27

to explain it in one way or

24:29

the other. So in that world 47

24:32

countries and Ukraine was one of them.

24:34

But so when the invasion happened, we

24:36

did not approve of that. So we

24:39

said, let's help out. And we provided

24:41

our technology and it's been deployed nicely

24:43

in a number of places over there.

24:46

It has been very good for them,

24:48

I think. You can find a lot

24:50

of good. I'll leave it to you,

24:52

to find the... the good proof points

24:55

that they've been willing to talk about,

24:57

very very very specific such. Well, wait,

24:59

why don't you tell us about the

25:02

proof points? I would love to hear

25:04

about them. It's a lot to brag

25:06

on the scene. You have to be

25:09

careful with this and that. But there

25:11

are some very specific examples where we've

25:13

been able to. So because part of

25:15

this is that when something is like

25:18

sort of think about when. a spy

25:20

tries to again we talk about the

25:22

criminals who can be pretty brutal we're

25:25

not brutal of course they're brutal but

25:27

pretty non-suddle they're like just they want

25:29

to go make money they many cases

25:32

they don't really care about whether they

25:34

burn something whether they create havoc as

25:36

long as they make money and and

25:38

you know they don't necessarily think the

25:41

most the longest term a spy if

25:43

it's a doesn't matter if the Russian

25:45

spy or Chinese spy or frankly our

25:48

own spies if they want to go

25:50

get access to information for are willing

25:52

to take, you know, if you have,

25:54

if the policy is that we should

25:57

know what's in Putin's diary or in

25:59

Zing Ping's travel records or the other

26:01

way around, one is willing to spend

26:04

any number of years. to get to

26:06

that information, and you have to be

26:08

very subtle about it. So if you

26:11

assume that Russian spies or Chinese spies

26:13

or anybody who's trying to come out

26:15

Ukraine are trying to do that, they'll

26:17

be very careful, very subtle, and so

26:20

on. So what it turns out then,

26:22

or for that sake, when somebody wants

26:24

to come and destroy something in the

26:27

world of computers, When they do that,

26:29

they sort of become a gigantic honeypot

26:31

if you think about it. It's like

26:34

just everything shows up there. So when

26:36

we deploy the recorder future across a

26:38

whole bunch of places there, we ended

26:40

up also learning a time. So you

26:43

sort of get in the way of

26:45

the absolutely most modern malware of various

26:47

sorts and all the other stuff that

26:50

comes around that. And so... any number

26:52

of times, we've been able to sort

26:54

of detect incredible things and then be

26:56

able to have that data flow into

26:59

all our other customers in a way

27:01

so that they can be defended against

27:03

the same. So it ends up being

27:06

sort of a, I don't know, probably

27:08

not a popular inality, but virtual iron

27:10

dome type of thing that this turns

27:13

into, and so pretty incredible sort of

27:15

outcome out of that, and yeah, continues

27:17

well to this day. Interesting,

27:19

one question that comes to mind for

27:22

me there is, you know, is it

27:24

ever tricky to decide what governments to

27:26

work with and not work with? I

27:28

mean, running weights and biases, I've been,

27:30

you know, kind of surprised by how

27:33

many different, you know, customers we have

27:35

where, you know, employees might object and

27:37

then, you know, I think, like, it's

27:39

really not like... my expertise to figure

27:41

out like who the good guys and

27:44

the bad guys are but you know

27:46

we're not exactly at the same level

27:48

that that you are I mean it

27:50

kind of is your job maybe to

27:52

figure out who the good guys and

27:55

bad guys are and it's actually like

27:57

probably pretty complicated like what do you

27:59

really start digging in like how do

28:01

you approach that like if a new

28:03

government wanted to work with you do

28:06

you have some kind of like flowchart

28:08

that decides if they're on our team

28:10

or not? Yeah. And how does that

28:12

work? You know, as you can imagine,

28:15

it's not something we necessarily publish, but

28:17

I just read some general principles. But

28:19

you're right, first of all, it is

28:21

our job to be able to make

28:23

those sort of judgments. And you have

28:26

to make choices and no such choices

28:28

will be perfect. And it's not like

28:30

we can claim to have the most,

28:32

we're... software dudes, we're not necessarily, or

28:34

software people, we're not necessarily moral philosophers,

28:37

just to be in not overstate our

28:39

own abilities. But, you know, we chose

28:41

to not have customers in China and

28:43

Russia and Iran and North Korea, those,

28:45

that was sort of an easy choice.

28:48

There are countries that you as law

28:50

and European law nor the same countries

28:52

to some degree but also places like

28:54

Cuba and Sudan and you know like

28:56

there's six seven countries that Syria that

28:59

you actually go straight to prison if

29:01

you sell them stuff so that makes

29:03

it easy that's easy those are the

29:05

easy ones yeah then you have maybe

29:07

other places that are more tricky and

29:10

they are not going to start naming

29:12

but we made our choices where there

29:14

is a number of countries we just

29:16

stay away from and then as well

29:18

as companies there is a number of

29:21

countries sorry companies that for example you

29:23

know engage in illicit hacking behavior that

29:25

we stay away from so there we

29:27

we call it an audit list call

29:30

it whatever you don't publish that but

29:32

and and then there is the companies

29:34

that help out some of these things

29:36

and so on. So there's a, we

29:38

have a process maybe and it is

29:41

a flow chart is not a bad

29:43

way of thinking about it. And then

29:45

we have sort of a committee at

29:47

the very core of this when there's

29:49

choices to be made. And sometimes you

29:52

actually have to make choices and we

29:54

make those choices. So I make no

29:56

claims about that being easy, but it's

29:58

never been that hard either. It's sort

30:00

of. Sometimes it would be hard to

30:03

write down. But we have sort of

30:05

a committee and in the end game,

30:07

that committee is who decides and I

30:09

think we've been able to do that

30:11

in a way that we can go

30:14

to bed nicely and so on. Now

30:16

our stuff is for cyber defense also.

30:18

It's pretty worthless. It's not like you're

30:20

not going to use our software too.

30:22

decide where you're going to drop a

30:25

bomb or to do, you cannot even,

30:27

you cannot physically use us to intrude

30:29

into a phone or do any of

30:31

the malicious, it's cyber defense. So even

30:33

if a bad guy got their hands

30:36

on it, it wouldn't be great, but

30:38

it's not like it's going to like

30:40

create complete havoc in the world. So

30:42

in the end game, it's hosted softers.

30:45

If the wrong guy got their hands

30:47

on it, we can cut it any

30:49

time. So, you know, there's a number

30:51

of controls to it. So it. So

30:53

it's good. So it's good. So it's

30:56

good. I guess one of things that's

30:58

important to me is that, you know,

31:00

our customers know that we're not going

31:02

to cut them off if they get

31:04

sort of politically unpopular, which is why,

31:07

you know, I want to be like

31:09

super clear about our criteria. Do you

31:11

worry about that at all for record

31:13

features? Just so kind of clear what

31:15

you guys are willing to do and

31:18

not. So first of all, our stuff

31:20

is fairly expensive, so we only have

31:22

2,000 customers. You know, whereas somebody else,

31:24

you might have many, many more. So

31:26

that then, so and we know who

31:29

they are. They run every customer of

31:31

these 2,000 goes through sort of a

31:33

KYC process, know your customer, sort of

31:35

thing. So it's, if somebody shows up

31:37

and they're called a company name that

31:40

we have no idea of, they have

31:42

no website, they have no website, nothing.

31:44

It's just not going to spend $100,000.

31:46

If our stuff ranges from that up

31:49

to millions and millions, you know, that

31:51

itself selects a lot of different things.

31:53

It's like, whereas if you have a

31:55

little dev shop with four guys who

31:57

might use some AI tools, it's I

32:00

think you might be more difficult for

32:02

you to make those sort of choices.

32:04

So I don't know, tell me if

32:06

I'm wrong, but I think we sort

32:08

of, there's a whole bunch of variables.

32:11

that have made it so that this

32:13

has been less of an issue, actually.

32:15

That's great. Actually, shifting gears, I think,

32:17

you know, we found a saying of

32:19

yours that I think you repeated earlier

32:22

here that I want to ask you

32:24

about, which is saying, you know, I

32:26

think it's like in the last 25

32:28

years, the internet reflected the world and

32:30

the next 25, the world will reflect

32:33

the internet. Can you expand on what

32:35

you mean by that? the trivial one

32:37

is sort of like everything that's happened

32:39

in the physical world sort of whether

32:41

it's everything from sort of transactions to

32:44

interactions to like how we do business

32:46

how we sort of the simple stuff

32:48

is sort of moved on to the

32:50

internet and and and that the internet

32:52

then through that becomes a reflection of

32:55

what's going on in the world and

32:57

I think we're all very happy to

32:59

see that and there's not a lot

33:01

of drama that comes out of that

33:04

then But then sort of saying that

33:06

as soon as those systems here is

33:08

sort of its internet first is sort

33:10

of what's starting to happen and whether

33:12

that sort of. the trivial social, when

33:15

you know, you see kids where, where,

33:17

what's going on in Tiktok or Facebook

33:19

or whatever is sort of that, maybe

33:21

not kids and Facebook, that's not true,

33:23

but, but, you know, whatever the preference

33:26

of social network of preference is, it's

33:28

sort of, it starts in the internet

33:30

world. When democracy is sort of first.

33:32

internet and then on to the world

33:34

when business is first internet and and

33:37

now we end up in a world

33:39

where what happens on the internet is

33:41

actually what defines the world and that's

33:43

sort of what I think becomes pretty

33:45

interesting because now if you run a

33:48

country it's becomes tricky you know who's

33:50

who's in power is it the country

33:52

or is it Mark Zuckerberg with Facebook

33:54

and I think it's been pretty interesting

33:56

even observing Facebook and just seeing how

33:59

Facebook switched head of government affairs at

34:01

the same time as years. if we're

34:03

president and that was probably not my

34:05

window. So I think we're going to

34:07

see a lot of this and if

34:10

you sort of fast forward where that

34:12

will be 25 years from now, I

34:14

wouldn't, I'm not keen to be a

34:16

politician but large, but I certainly wouldn't

34:19

want to try to be a politician

34:21

in 25 years because it's probably going

34:23

to be pretty tricky. And I guess,

34:25

you know, you talked about elections earlier

34:27

and we found a lot of people.

34:30

on the show actually talking about different

34:32

kinds of election interference and mitigation efforts.

34:34

When you kind of roll forward, the

34:36

trends that you're seeing, do you think

34:38

that, you know, the US society has

34:41

to operate a different way to be

34:43

successful in this more vulnerable world? Big

34:47

question. I'm a huge democracy fan. I

34:49

sort of like whether you take it

34:51

from a positive way that how positive

34:53

isn't it that we can live in

34:55

a world with sort of humanity's been

34:57

around for tens of thousands of years

34:59

and and it's sort of most of

35:02

the time has been the strong man

35:04

that rules and you sort of follow

35:06

and and us only and even when

35:08

there had been democracy it's only or

35:10

some it's only being for the super

35:12

privileged or whatever you have sort of

35:14

thing. pretty damn positive that we can

35:17

have a vote. So that's sort of

35:19

pretty amazing that we get to live

35:21

in that time. Now, if you then

35:23

fast forward and start thinking about what's

35:25

going to happen there. First of all,

35:27

I was going to say that the

35:29

other non-positive ways is to say there's

35:32

a lot of other ways and maybe

35:34

democracy isn't the best. But it's the

35:36

best of the worst. There's sort of

35:38

like none of that stuff is amazing.

35:40

It all has problems and so on.

35:42

But it's... I certainly haven't seen anything

35:45

remotely as good as democracy. Now, so

35:47

if you go forward and you just

35:49

say, what are the ways the democracy

35:51

could develop? Again, there is this sort

35:53

of thing, what, because we think very

35:55

much in democracy in terms of countries,

35:57

and I think that. That's important and

36:00

that's probably going to remain true. But

36:02

again, if the sort of the groups

36:04

of people are not necessarily following national

36:06

borders, it might make it very difficult.

36:08

Now, maybe we are sort of a

36:10

new era right now. I can't really

36:12

make that judgment where. There's a new

36:15

set of nationalism going on in a

36:17

bunch of different places. Better for worse,

36:19

we can only observe it and that

36:21

seems to be the case, but maybe

36:23

that's a reaction and what's going on

36:25

would happen on the internet. And I

36:27

think there's many other of these sort

36:30

of things. I think maybe the negative

36:32

view would be that there's the world

36:34

that's been manipulated in many different ways

36:36

and there's amazing opportunities in front of

36:38

us to manipulate if I put on

36:40

my... I want to be evil brain.

36:43

There's many ways to be evil. But

36:45

maybe on the other hand, people are

36:47

also, let's be positive and say, maybe

36:49

people are getting smart and, you know,

36:51

a lot of manipulation is. pretty primitive

36:53

so hopefully you know like I'm sort

36:55

of always say look people always come

36:58

up with wonderful text solutions to deal

37:00

with this how about if we do

37:02

what they do in Finland and just

37:04

get people great education and and we

37:06

can sort a lot of these problems

37:08

let's make sure that people can read

37:10

really well and other sort of things

37:13

and it can work out so I

37:15

am rambling a little bit but I

37:17

guess I'm a generally an optimist democracy

37:19

is better than anything and and yeah

37:22

That's great to hear. I mean, what

37:24

do you like? Do you think that

37:26

when you think about sort of like

37:28

AIs like offensive techniques and defensive techniques?

37:31

Do you have a sense on like

37:33

what wins in terms of like, I

37:35

don't know, election interference or kind of

37:37

any other conflict that we come into

37:40

here? Like, um, Yeah, like you can

37:42

use AI to look for security vulnerabilities

37:44

and fix them or look for security

37:47

vulnerabilities and exploit them. I mean, if

37:49

you think about like sort of, if

37:51

cyber attacks with large have this happened

37:53

faster and faster, and so if I

37:56

start there just because that's sort of

37:58

closest to home, and clearly, so for

38:00

now the AI stuff you're really seeing

38:02

is really only. these sort of better

38:05

fishing emails and so on. But even

38:07

there, it's like, you know, like, there

38:09

was a reason most cyber attacks happened

38:11

in English speaking countries for a long

38:14

time because right in the fishing emails

38:16

was easiest to do in English, sort

38:18

of thing. But now you can do

38:20

that in all kinds of different things.

38:23

You can do more targeted stuff, blah,

38:25

blah, blah, blah. But people are going

38:27

to start writing software that once it

38:29

gets in, it traverses through systems automatically.

38:32

That for sure people are. slow, slow,

38:34

moving things that tries to not detect,

38:36

or it's the sort of stuff where

38:38

they don't really care and there's a

38:41

prum, run through the system. That's gonna

38:43

happen. That's gonna set up the bar

38:45

for the defenders to try to build

38:47

things that you're not gonna actually be

38:50

able to run counter that with just

38:52

humans. You're gonna have to use AI

38:54

to sort of get or some version

38:56

of. automated analysis, but we can call

38:59

that AI to fight against that for

39:01

sure and that who wins that there?

39:03

Yeah, I like there's a guy Rob

39:05

Joyce who's a terrific man who used

39:08

to run cyber defense for the US

39:10

government at NSA. He before that ran

39:12

the hacking team there one of the

39:14

groups and famously called TAO at the

39:17

time and he has a great saying

39:19

where he said you have to know

39:21

your network better than the bad guy

39:23

coming at you. And if you're a

39:26

big company. It's freaking hard to know

39:28

their network. And a, you know, a

39:30

crawler plus some AI is going to

39:32

outsmart most IT guys even if they

39:35

work there for 25 years. So, so

39:37

there you have that. And I don't

39:39

think it's certain who wins that. We,

39:42

for sure, in disinformation to your point,

39:44

I'd like to think that we should

39:46

be able to write. you know so

39:48

in the recorded future when you get

39:51

text and images and all this stuff

39:53

now we try to classify and say

39:55

instead of information you're looking at here

39:57

is very likely machine generated and it

40:00

being machinery generated with these models that

40:02

sort of stuff to help the researcher

40:04

now when that can be in an

40:06

our operational system so whether I'm on

40:09

tic-toc or my email or whatever to

40:11

say that, but maybe on the other

40:13

hand, maybe 75% of what I'm going

40:15

to deal with in the future is

40:18

going to be machine-gen rated. So maybe

40:20

that's not enough. It's like the world,

40:22

the future is maybe machine-gen-rated. So I

40:24

think that's sort of interesting. And then

40:27

you have the actual warfare stuff, where

40:29

what's going to happen is that I'm

40:31

going to paint a square and staying

40:33

in this 10 by 10 K kilometer.

40:36

There's going to be drones hanging over

40:38

that stuff with guns and grenades that

40:40

what you pick. And anything that moves

40:42

in there is dead. You know, like

40:45

that sort of stuff. And people are

40:47

working on that right now. That's going

40:49

to change the nature of warfare in

40:51

a brutal way. So no, there's be

40:54

all kinds of nasty stuff in front

40:56

of us. So sorry, just I'm doom

40:58

and gloom here, but. In a vein,

41:00

I guess, and maybe it's like even

41:03

a, you know, practical question, like what

41:05

does it mean that the. the dark

41:07

web is moving from kind of forums

41:09

to telegram. I mean, I'm sort of

41:12

vaguely aware of telegram as an alternative

41:14

to signal. I've watched many, many, maybe

41:16

most of my CEO friends gradually moved

41:18

to signal in the last... you know,

41:21

year or two to the point where

41:23

I feel like I'm using it quite

41:25

a bit more than I expected to.

41:28

Is telegram a signal like equivalent or

41:30

why didn't, shouldn't a CEO be using

41:32

the same, you know, network that the

41:34

criminals are using? Would that be the

41:37

most? So first of all, use signal,

41:39

good choice. Make good choices, look, because

41:41

use signal, just so you know. Signal

41:43

is great. Signal publishes their source code.

41:46

you could argue this is actually I

41:48

have still yet really figured it out

41:50

but but there's no real server to

41:52

attack in signal there's lots of really

41:55

good things and all the fancy spies

41:57

of the world in all the good

41:59

countries use signal okay the fact that

42:01

they've got uncomfortable now that mean might

42:04

mean that they have a big cabal

42:06

where they all share you know or

42:08

you look at your mind emails but

42:10

no I've seen plenty of very smart

42:13

people use use signals I'm a huge

42:15

signal fan of The good telegram on

42:17

the other hand, you know, so the

42:19

guy, I'm not going to remember his

42:22

name now, but you know, very cool

42:24

entrepreneur in many ways. He built the

42:26

UK first, sort of the Russian Facebook.

42:28

He was forced to sell that because

42:31

the Russian government did not appreciate that

42:33

he had built this. They forced him

42:35

to sell it to an oligarch in

42:37

Russia. He was sort of really... Imagine

42:40

your arm being up here and you

42:42

sort of have to say yes, and

42:44

he had to say yes, and he

42:46

sold it. He then started telegram and

42:49

moved to Dubai to be able to

42:51

sort of run it, Abu Dhabi, or

42:53

Dubai, I think it's Dubai, and built

42:55

a messaging platform that is unlike signal,

42:58

if I understand correctly, does have sort

43:00

of a centralised place to be, it

43:02

has encryption that is not... at the

43:04

same way of end-to-end encryption as of

43:07

signal and others do in any number.

43:09

Like some of this is way above

43:11

my pay grade, but there's a whole

43:14

set of reasons why telegram is not

43:16

as good. It is interesting from this

43:18

sort of, and it's dangerous in their

43:20

world to sort of say good and

43:23

evil, but the set of people that

43:25

we don't necessarily love that are, or

43:27

many of them, are on telegram. signal

43:29

in a different way, but it ends

43:32

up being a good place to go

43:34

look for, and especially the criminals are

43:36

on there, whether it's sort of cyber

43:38

criminals or people in trafficking and all

43:41

kinds of interesting stuff, are there, and

43:43

it's turned into a very good information

43:45

source. Is there something about telegrams feature

43:47

set that it's better for criminals like

43:50

when if any criminals like listening to

43:52

this podcast you could be convinced to

43:54

switch over from telegram to signal I

43:56

mean surely they wouldn't watch it. I

43:59

think it's no I think it's it's

44:01

social this is sort of like and

44:03

it also turns out that Ukraine there

44:05

it does obviously in in the Russian

44:08

sort of world. is very popular, so

44:10

it's popular in Ukraine as well. So

44:12

it's not, I'm not by any means

44:14

saying that all telegram people, users are

44:17

bad by any means. It just sort

44:19

of happens to be a certain class

44:21

of criminals that ends up being concentrated

44:23

on telegram and, okay, interesting stuff. All

44:26

right, well, like, rolling forward into the...

44:28

There are some people who'd be very

44:30

mad at me for saying this stuff,

44:32

but... Wait, why? For saying what part

44:35

of it would they be mad at

44:37

it? Because once you sort of say

44:39

that one platform is more criminal than

44:41

the other, you know, like I'm... You

44:44

know, this is... Well, we can edit

44:46

it out if you want, but... No,

44:48

no, no. It's all good. So good.

44:50

Yeah, actually another question maybe that comes

44:53

to me is like, yeah, I feel

44:55

like, you know, in my very kind

44:57

of like low level of celebrity, you

44:59

know, like every like, you know, month

45:02

or two, I get someone reaching out

45:04

to LinkedIn, like, say it, I'm going

45:06

to come kill you or something, but

45:09

you must get these kind of threats

45:11

like all the time. Like, do you

45:13

worry about you and your family's like

45:15

safety being so kind of involved in

45:18

this world? Do you like, do you

45:20

walk around with the bodyguard? You must

45:22

be constantly concerned about getting hacked. Like

45:24

how do you think about that? Do

45:27

I look afraid? You don't look afraid.

45:29

Is that for cause or for hubris,

45:31

I guess? No, it would be funny

45:33

here. No, I'm not afraid. That's sort

45:36

of, you know, and you choose when

45:38

you get involved in this that that

45:40

you sort of and no bodyguards, unless

45:42

it's needed. It's sort of one way

45:45

of thinking about it. One should be

45:47

cautious. The sort of Russia deemed as

45:49

a, whatever they call it, undesirable enemy

45:51

of the state. So just before Christmas,

45:54

there was a verdict, if you want,

45:56

from the national prosecutor or the... The

45:58

general prosecutor of the Russian Federation put

46:00

out a statement around why, record a

46:03

future, is an enemy of the state,

46:05

is an undesirable, very specific language, not

46:07

like random stuff. It was very specific.

46:09

So that was less great if you

46:12

want. That sounds bad, man. people that

46:14

are undesirable, according to Russia, like get

46:16

turned out of windows and stuff. I'm

46:18

not super up on this, but... I

46:21

can assure you I'm not traveling to

46:23

Russia any time soon. Sure. That seems

46:25

to be after the... No, look, I,

46:27

you know, the... I think we're very,

46:30

very careful with information security here. We

46:32

run a good physical security program here

46:34

as well. We try to be very

46:36

careful and thoughtful about how we travel

46:39

what we do and be thoughtful about

46:41

what we do and be thoughtful about

46:43

things. then at the same time, once

46:45

you're not overestimated where you are on

46:48

the list of problems that the bad

46:50

guys have, you know, whether it's the

46:52

Chinese government or the Russian government, they

46:55

have a lot of shit to deal

46:57

with and I don't think we're sort

46:59

of at the top of that list.

47:01

So that's sort of, I don't want

47:04

to simplify it, but information security, we've

47:06

got to be on our egg game.

47:08

All right, well let's let's let's wrap

47:10

this up and come back to the

47:13

here and now so a couple months

47:15

ago You announced that you you were

47:17

getting acquired by by Masterguard for 2.65

47:19

billion. It's got to be one of

47:22

the biggest You know acquisitions of the

47:24

of the last year. Can you talk

47:26

about what the thesis was and how

47:28

that came about and what your plans

47:31

are going forward? Yeah, no, so we

47:33

had a long discussion with them, you

47:35

know, multi-year sort of discussions that has

47:37

been going on. We, the sort of

47:40

the simple thesis is that it's sort

47:42

of maybe not as known, but they

47:44

run a great payments business, of course.

47:46

They have a services business that actually

47:49

includes pretty nice assets on the cyber

47:51

and intelligence side. They wanted to expand

47:53

on this and we could be a

47:55

nice fit into that. We also, the

47:58

financial intelligence side is super interesting, that's

48:00

especially as we talked about cyber criminals

48:02

and so on, this sort of financial

48:04

end point to things. If we get

48:07

our hands on a stolen credit card,

48:09

it would be awfully interesting to know

48:11

whether a card has been used or

48:13

not. And the cards that have been

48:16

used or not actually tells you a

48:18

whole lot about the whole supply chain

48:20

of where those cards came from, just

48:22

to have one random sort of example.

48:25

So there is a lot of those

48:27

sort of things. So we're going to

48:29

be. building out our business. We're continuing to

48:32

operate on our own standalone. There's a lot

48:34

of cool synergies that we can do together.

48:36

And it was also one of those

48:38

where it was just felt like a

48:40

really good home for the company as

48:43

great people and sort of long-term value

48:45

in like mindset of things, which was

48:47

important to me, because kind of to

48:49

your point for both you and I

48:51

have built companies for a long time,

48:54

you want to make sure they end

48:56

up in places that are not just

48:58

going to be like five by night

49:00

type stuff, but people who are serious

49:02

and want to do things over long

49:05

term. any number of reasons that sort

49:07

of stacked up where this was a

49:09

great place to take it. Any

49:11

advice for me for success post

49:13

acquisition? I think, you know, make

49:15

sure they'll keep doing execution very well.

49:18

It's sort of like you have to

49:20

keep at your A game. We're early

49:22

into it, but, you know, build good

49:25

friendships, good relationships with people because this

49:27

stuff is not easy. When you do

49:29

it, as you know, it's like, you

49:31

know, it's easy to come up with

49:34

a lot of clever thoughts on paper,

49:36

but we're going to make it do

49:38

it, it's like with people. So

49:40

make sure you have good relationships

49:42

with those people. Yeah, I think

49:44

that's it. And don't just think

49:46

it happens sort of automatically. It's

49:48

sort of like it's, it's, unfortunately,

49:51

I guess it continues to be

49:53

hard work. It's sort of unavoidable.

49:55

Awesome.