0:01

Welcome to Zero Knowledge,

0:04

I'm your host Anna

0:06

Rose. In this podcast

0:09

we will be exploring

0:11

the latest in Zero

0:14

Knowledge Research and the

0:16

decentralized web, as well

0:19

as new paradigms that

0:21

promise to change the

0:24

way we interact and

0:27

transact online. Nico and

0:29

I chat with Andrew and Vivic from

0:31

Cursive. We catch up with them about

0:33

what led them to start working

0:35

on a series of activations and

0:38

experiments using programmable cryptography.

0:40

We map out the timeline of these

0:42

activations. Often they were part of in-person

0:44

events, including events like ZK Summit 11

0:46

and ZK Summit 12 last year. They

0:49

walk us through some of the challenges

0:51

and also insights they had. as they

0:53

speed ran the deployment of

0:55

experimental applications using experimental libraries

0:58

with real users. cursive is one

1:00

of the teams I feel are on the

1:02

edge of the application development front in ZK

1:04

and in the adjacent fields like MPC and

1:07

FHE and they've been working on some

1:09

concepts that I think could be quite powerful

1:11

in the future. Concepts like passively sharing digital

1:13

signals from private user devices or what they

1:15

call digital pheromones as well as the idea

1:18

of narrow casting. Nico and I have actually

1:20

mentioned these concepts before on the show, so

1:22

it's great to get a chance to catch

1:25

up with the team who invented them.

1:27

Now before we kick off this episode, I

1:29

want to share a few things from our

1:31

ecosystem. First, if you haven't seen it yet,

1:33

ZK Summit 13 is happening in Toronto on

1:36

May 12th. Applications to attend just opened. Space

1:38

is more limited than usual this time around,

1:40

as our venue caps out at around 400,

1:42

so we are encouraging people who really want

1:44

to join to get their spots secured early.

1:47

It is invite only and we do prioritize

1:49

previous attendees and long-term members of the ZK

1:51

community. So if you're newer and you really

1:53

want to join, just aim to apply early

1:55

for a better chance of getting a spot.

1:58

The application to speak can be found. the

2:00

same form. The deadline to apply to

2:02

speak is March 15th. Next up on

2:04

the ZK Hack front, if you don't

2:06

know ZK Hack as a hub for

2:08

learning about ZK, if you're just getting

2:11

into the space, this would be a

2:13

really good place to check out. One

2:15

of our projects there, the ZK Whiteboard

2:17

sessions, just wrapped up. We released our

2:19

final two modules in the past month.

2:21

These were hosted by Dan Bonet. They

2:23

were focused on Fry and Proximity Proximity.

2:25

We also completed our first study group

2:27

that went through these videos kind of

2:30

in detail, but we are looking to

2:32

launch a new one in the coming

2:34

months focused just on these Frye modules.

2:36

So if that's interesting for you, be

2:38

sure to head over to the ZKHAC

2:40

Discord. I'll add the link to that

2:42

and show notes. That's where we organized

2:44

the study groups. I also just want

2:46

to mention that ZKHAC Soul has been

2:49

unfortunately postponed. We weren't able to secure

2:51

the venue that we needed, and so

2:53

we've decided not to do the event

2:55

this year. In the meantime, though, we

2:57

are hosting ZKHAC meetups in Sophia around

2:59

the ZK Proof's event and in Denver,

3:01

right around the East Denver event. We

3:03

are also developing our next hackathon. It

3:05

will probably be in Berlin in June,

3:08

around the Protocolburg and DAPCon events. We'll

3:10

be sharing more over on the ZKHAC

3:12

channels in the lead up to that.

3:14

So I think that wraps up all

3:16

the upcoming events. Hope to see you

3:18

at some of these. Now, here's our

3:20

episode with Andrew and Vivic from cursive.

3:25

Today, Nico and I are here with

3:27

Vivic and Andrew from Cursive. Welcome both

3:29

of you to the show. Excited to

3:32

be here. A longtime ZK podcast fan

3:34

boy. Oh, I'm so glad. Nico, thanks

3:36

for joining. Hello. Hello. This episode is

3:39

a long time coming. We have actually

3:41

been trying to schedule it for the

3:43

last two months or so. Niko, you

3:45

and I, we did an episode back

3:48

in October where we talked a little

3:50

bit about some of the cursive activations

3:52

and experiments. And then in our end

3:55

of year episode, we also mentioned what

3:57

you guys had been up to. So

3:59

yeah, it's so good to have you

4:02

on so we can act. dig into

4:04

all of these events that you've been

4:06

present at and what you're thinking in

4:08

terms of using programable cryptography in the

4:11

real world. So let's kick off with

4:13

a little backstory on the two of

4:15

you. Where did you get your start?

4:18

What got you excited about this? And

4:20

maybe at what point did your path

4:22

start to merge? Vivic, let's start with

4:25

you. Yeah, so back in 2016 was

4:27

the first time I started doing cryptography

4:29

work. I was part of this like

4:32

a research program at MIT where I

4:34

was paired with some grad students to

4:36

work on computer science projects and they

4:38

just like randomly slotted me into cryptography.

4:41

And yeah, I like started out doing

4:43

some stuff with consensus alternatives like proof

4:45

of space, so very crypto economic, but

4:48

involved cryptography, mercal trees. And then I

4:50

did some authenticated dictionary work with Olean

4:52

Tamasku, who I think has actually been

4:55

a guest on the show as well.

4:57

Oh yeah. Yep. So we worked together

4:59

for two, three years using KZG commitments

5:01

for like authenticated dictionaries and then kind

5:04

of just dropped off cryptography for a

5:06

while because I kind of thought, I

5:08

don't want to be an academic and

5:11

everything in industry seems like it's weird

5:13

crypto scams or something. Like it was

5:15

like during like an FT craze. And

5:18

I like briefly interned at like a

5:20

crypto trading firm and I was like,

5:22

this is degeneracy, I don't want to

5:25

do this. Crypto is scary and weird.

5:27

That is true. Absolutely. And that has

5:29

like the exact same process when I...

5:31

Yeah, studied cryptography and then it was

5:34

like, I can't get into all this

5:36

crypto scam nonsense. Yeah. Like I have

5:38

to work somewhere else. No, exactly. And

5:41

then I did like machine learning research

5:43

for a few years at MIT and

5:45

then near the end of graduation. One

5:48

of my best friends at Yish who's

5:50

also been a guest on the show,

5:52

he's like good friends with Brian Gu,

5:54

the founder of Xerox Park. I think

5:57

they go back to like high school.

5:59

Okay. And he told me they were

6:01

spinning up this new. They're doing this

6:04

weird ZK thing and I was like,

6:06

what is this? Like, you know, it's

6:08

probably another crypto scam. And then I

6:11

started reading the Vitalic blog posts and

6:13

I was like, oh my God, this

6:15

is like KZG commitments. Like, this is

6:18

the same stuff I was doing. It's

6:20

just like applied in a different way.

6:22

And so yeah, I basically started with

6:24

them in like late 2021, early 2022

6:27

doing some initial experiments around anonymous speech.

6:29

Nice. So that was Xerox park. What

6:31

were you, I mean, were you helping

6:34

out? as an employee or were you

6:36

just kind of learning along, were you

6:38

kind of a resident? I actually don't

6:41

know how it was structured back then.

6:43

Early on I think it was structured

6:45

very similar to PSC in that it

6:47

was essentially like giving out a bunch

6:50

of grants. A lot of them were

6:52

like Bryan's ideas for various projects he

6:54

wanted to see, including the stuff that

6:57

I was working on, like anonymous speech

6:59

was something that he was championing for

7:01

a while. Okay. So I was just

7:04

a grantee there for. About a year,

7:06

and then I was briefly a full-time

7:08

employee working on SU pass. Were you

7:11

there during like the Dark Forest era

7:13

or did you join after that? No,

7:15

I was not. I like knew it

7:17

was happening. I'd like known Brian from

7:20

some other programs and I knew they

7:22

were working on it, but I was

7:24

not involved with Dark Forest. Okay. And

7:27

what about you Andrew? How did you

7:29

get your start? I guess I had

7:31

a less involved past with cryptography with

7:34

cryptography. I never studied cryptographyography. in college,

7:36

but one of my friends from doing

7:38

math together kind of reminded me that

7:40

programmable cryptography existed. He's basically like, oh,

7:43

this ZK thing is really cool. You

7:45

should go check it out. And I

7:47

was completely fascinated. This was maybe in

7:50

2023. Okay. In the springtime. Yeah. So

7:52

quite recent, actually. And I got into

7:54

it purely from the math side. I

7:57

was like, this is completely magical. the

7:59

idea that you can prove things with

8:01

revealing any extra information is completely absurd

8:04

and it makes no sense and the

8:06

fact that it's just built on algebraic

8:08

objects and numbers is incredible so that's

8:10

kind of what drew me in at

8:13

the beginning and after a little while

8:15

I was like okay this is really

8:17

cool I want to work on this

8:20

for a little while let me go

8:22

meet some people in the space

8:24

first I joined a PSE project for

8:26

a few months. It's working on

8:29

identity wallets for a bit, based on

8:31

zero knowledge proofs. And then afterwards,

8:33

I went to meet a bunch of

8:35

people who were working in ZK. I met

8:37

Lockshman from Persone. I don't know

8:40

if he's been on the show,

8:42

but Lockshman and I had a

8:44

conversation where we were talking about

8:46

how... ZK enables people to express themselves

8:48

differently, how it can be connected

8:50

to identity in the physical world.

8:53

One cool concept we toyed around

8:55

with was the idea of being

8:57

able to change an outfit or

8:59

wear different pieces of clothing and

9:01

basically become a different person, have

9:03

different identity personas. And from

9:05

that, Lachshman was like... Okay, I

9:08

know someone else who has similar

9:10

thoughts or is thinking in similar

9:12

directions and that's when he introduced

9:14

me to Vivik and we've been

9:16

working together since then. Oh, cool.

9:18

Vivik, you were doing Zu Pass. You sort

9:21

of just mentioned that. This is Zuzalo Zu

9:23

Pass? Yep, same one. So I guess, did

9:25

you guys meet after that? Yeah, so I

9:27

was, yeah, working on the Zu

9:29

Pass team. And actually I think

9:31

we're cursive, like, like, like, properly

9:33

started Zuzalu Zuzulu. There I met

9:36

Rachel who's like currently like one

9:38

of the co-founders and like our

9:40

head of design. She had been

9:42

working at PSC for a few

9:44

years like helping with design for

9:46

a bunch of projects there and

9:48

I also met Althea who was like

9:50

a core member of our team during

9:52

the early stages like also kind

9:55

of a design and like Com's

9:57

background and so like it was near the

9:59

end of Zuzalu where we all kind

10:01

of met and built out kind of

10:03

an initial NFC experiment using like signatures

10:06

and NFC cards which we can get

10:08

more into in a bit and then

10:10

I think that was like May 2023

10:12

and I met Andrew like July 2023

10:14

over video call. I actually when I

10:17

first called Andrew I first called Andrew

10:19

I thought he like thought what we

10:21

were doing was really boring like I

10:23

was like oh I don't think I'm

10:26

ever got to talk to him again

10:28

that's unfortunate. Did you find it boring

10:30

Andrew? No, no, I thought it was

10:32

awesome. Otherwise I probably would not have

10:34

reached out again. That's hilarious. I didn't

10:37

know that. Cool. You also had long

10:39

hair. I never actually saw a long

10:41

hair Andrew in person. I just saw

10:43

it over that morning. I did have

10:46

really long hair and video. There's very

10:48

few people in the ZK space who

10:50

have seen that rare version. Before we

10:52

dive into cursive, because that's kind of

10:55

what I want to cover primarily in

10:57

this. episode cursive in the experiments you've

10:59

run. I don't know if we've ever

11:01

talked about zoo pass on the show

11:03

and it might be worth it to

11:06

talk a little bit about what that

11:08

was because yeah this is like at

11:10

Zuzalu that two-month pop-up city kind of

11:12

the first one that prompted I mean

11:15

since then there's been lots of variations

11:17

and and I think there's news Zuzalu's

11:19

in different places but yeah there was

11:21

this zoo pass so maybe tell us

11:23

a little bit about what that was.

11:26

Yeah, definitely. So I was one of

11:28

the first two devs on the team

11:30

and I was there for sort of

11:32

like roughly the founding of it. I

11:35

would not call myself a founder, but

11:37

like I was basically just like hanging

11:39

out with Brian around the time when

11:41

he was coming up with the concept

11:43

for Zupas and he was like, do

11:46

you want to work on this? And

11:48

I was like, yeah, let's do it.

11:50

And yeah, I mean, I would say

11:52

it's basically an identity wallet that I

11:55

think it's innovation as far as I

11:57

can understand is like. they targeted a

11:59

specific in-person event and also one that

12:01

was very friendly to trying out new

12:03

technologies. Yeah, we're able to sort of

12:06

like digitize a bunch. various like tickets

12:08

that were involved with events and like

12:10

other kind of like data for citizens

12:12

of Zuzalu or other sort of like

12:15

pop-up cities. Yeah and I think in

12:17

in the past year and a half

12:19

I haven't followed their work super closely

12:21

so I'm probably missing some details but

12:24

I think there's been a bunch of

12:26

innovations on like data format and like

12:28

ways of using Zupas more easily as

12:30

a developer so they're just continuing to

12:32

like I think. make it easier to

12:35

use ZK, kind of friendly data, and

12:37

develop with it. Yeah, I think I

12:39

had it. I mean, it was kind

12:41

of used as, I think you actually

12:44

needed to get it for your initial

12:46

ticket to like get in or something.

12:48

I think there's maybe ways around it,

12:50

but it was pretty integrated into the

12:52

experience. I also know that I like,

12:55

I lost it at some point. And

12:57

then for Dev Connect, you needed it

12:59

again. There was a way to do

13:01

that. So I was able to get

13:04

it back, which I, yeah, I don't

13:06

actually know how that worked under the

13:08

hood. But it's actually something that we

13:10

would love to talk about more, but

13:12

with a lot of this stuff that

13:15

you have like true ownership over, it's

13:17

actually like a very fragile system. In

13:19

the case of ZK data, it's less

13:21

of like a signature private key, but

13:24

more of like an encryption private key,

13:26

which is usually derived from a master

13:28

password. And if you forget that, you're

13:30

kind of just screwed. Like you just

13:33

like cannot recover your data. And so

13:35

yeah, it's pretty interesting, like these systems

13:37

can be a lot more fragile because

13:39

there isn't really easy data recovery. So

13:41

I guess, is this a good time

13:44

to ask what is cursive? Yeah. discover

13:46

and deepen their connections. So we want

13:48

to make experiences that align with user

13:50

empowerment. We believe that users should have

13:53

control of their experience. And we think

13:55

programmable cryptography is a really cool toolkit

13:57

for enabling this, giving users ownership of

13:59

their data. giving a lot more choice

14:01

over what types of front ends and

14:04

applications that they might actually be able

14:06

to use, enabling new ways for them

14:08

to express their identity, and ultimately this

14:10

is all in the aim of connecting

14:13

people and bringing them together. Is

14:15

cursive like a lab? Is it a

14:17

company? Do you have a product? I

14:19

feel like this is something you've actually

14:21

probably heard a couple people ask. I

14:23

think I might have asked you this

14:25

a couple times too, but yeah. What

14:27

is cursive exactly. I think we're

14:29

currently transitioning from closer to an

14:31

R&D lab to more of

14:33

just like a product team. I

14:36

think like we have a lot of good

14:38

lessons on the research front and

14:40

like what is kind of possible

14:43

and you know more difficult with

14:45

cryptography. So I think 2025 will

14:47

definitely be a very like product

14:50

driven year for us. We

14:52

were in a very interesting position

14:54

to begin with because I think

14:57

we had a lot of product-oriented

14:59

thinking in our team, but at the same

15:01

time a lot of the technologies

15:03

that we wanted to explore and

15:05

use simply we're not ready or

15:07

we're not packaged into user-facing form

15:10

and so we had to spend a lot

15:12

of time not only iterating on the

15:14

product side but also just fundamentally seeing

15:16

what is possible, what can be done

15:19

with these tools. So in a way

15:21

we were forced to do both, but

15:23

I think we took a lot of

15:25

really good learnings from that. Nice. I

15:27

want to ask, what's your approach

15:29

to designing these products? Right? I've seen

15:32

a lot of teams who fall in love with

15:34

the technology and say, like, this is

15:36

great, I want to use it. And I get

15:38

a sense that you guys do it the

15:40

other way around, right? I mean, I

15:42

think there's a healthy amount of that

15:44

as well. Like, I think Andrew and

15:46

I definitely get nerd sniped and like

15:48

just want to figure out new ways

15:50

of using stuff. We think a lot

15:52

about... constraints in our process and a

15:54

lot of this is spearheaded by like

15:56

our head of design ritual like just

15:58

creativity really I think easier to access

16:00

when you give yourself a box to

16:03

work with them instead of being like

16:05

you know the sky's the limit. I

16:07

think one of those constraints is like

16:09

around like human connection and data ownership

16:12

like what are sort of the existing

16:14

problems with with systems that we have

16:16

right now and in what ways does

16:18

cryptography help in what ways does it

16:21

make the UX bad? Like I think

16:23

we're also very honest as a team

16:25

and very pragmatic. I think we understand

16:27

that like it would be lovely to

16:29

have the purest system with like you

16:32

know like a trustless or something close

16:34

to that. I think often that just

16:36

needs to be balanced with like what

16:38

users care about and like how good

16:41

the experience can be. So yeah I

16:43

think that's been the kind of the

16:45

process in the past I think moving

16:47

forward we're looking to be a lot

16:49

more user-centric just like focusing on a

16:52

specific persona like kind of more traditional

16:54

startup stuff given all the knowledge we've

16:56

acquired on different cryptographic tech and progressed

16:58

that way but that's a little bit

17:01

more new. I think is going to

17:03

be like a 2025 focus. Yeah, I

17:05

think in the early stages of any

17:07

new technology, if you want to be

17:10

both on the cutting edge and also

17:12

for consumers, there is sort of this

17:14

dialectic between product and technology. Right, you

17:16

build some things, you give it to

17:18

users, you see how they feel, but

17:21

at the same time you have to

17:23

turn to the technology and say, oh,

17:25

can we build this? And also, what

17:27

else can we build? And this new

17:30

thing, like, do people want that? So

17:32

I think it's difficult in the early

17:34

stages to be too much in one

17:36

or the other, but slowly, that'll change.

17:39

Let's do a rundown of the experiments

17:41

that you did in the last year

17:43

and a half or so. And then,

17:45

yeah, because I'm really curious to see

17:47

kind of like what you learned maybe

17:50

with each one. Did you do something

17:52

after Zuzalu once cursive was founded before

17:54

Dev Connect or was Dev Connect sort

17:56

of the first time that you guys

17:59

went out as cursive? Actually, I think

18:01

a Dev Connect, we weren't even properly

18:03

called cursive, but it was the same

18:05

kind of team. I think cursive. itself

18:08

was launched at Denver. Okay. But there

18:10

was some experiments in between there. We'll

18:12

just quickly run through 2023 stuff. Yep.

18:14

And then pass off to Andrew for

18:16

2024. So at Zuzalu, basically we did

18:19

a really simple experiment with NFC cards

18:21

where there wasn't even an EasyK involved,

18:23

but you could basically collect like kind

18:25

of like a private Po-up where an

18:28

NFC card had its own private key.

18:30

and it would produce like unique signatures

18:32

upon every tap. And so you could

18:34

basically assign an FC card to a

18:36

specific event at Suzalu, like let's say

18:39

attending a town hall, and people who

18:41

tapped that would essentially get this unique

18:43

signature. It would be stored privately in

18:45

their zoo pass. And then theoretically they

18:48

could use it later, you know, to

18:50

prove they attended, either just with the

18:52

signature itself or using a ZK wrapper.

18:54

We didn't actually get around to doing

18:57

the ZK step, it was mostly just

18:59

like, kind of maybe more on the

19:01

side of like verifiable credentials than like

19:03

zero knowledge proofs, but yeah, I think

19:05

we just wanted to create more data

19:08

for people to make ZKPs with. That

19:10

was kind of the problem we identified

19:12

during different workshops with Rachel and Althea

19:14

of like, yeah, this tech is cool,

19:17

but like no one can use it

19:19

because like there was like ZK email

19:21

and TLS notary vaguely existed, but. it

19:23

was too hard to use those so

19:26

you couldn't make proofs about anything. I

19:28

guess it's also on your side laying

19:30

the groundworks for what's to come next

19:32

right? Like all the infrastructure that you

19:34

guys are gonna reuse? Absolutely yeah like

19:37

once people have a you know a

19:39

little supply of signatures you can start

19:41

to you know do a lot more

19:43

fun stuff with that. Yeah so that

19:46

was that and I think like the

19:48

next step was basically adding these signatures

19:50

and so this came around yeah like

19:52

late 2023 this also when Andrew joined

19:54

the team. which was an amazing level

19:57

up for us. Oh my God. We

19:59

were just suddenly able to do so

20:01

much more because Andrew is a force

20:03

of nature. Yeah, basically we experimented with

20:06

this concept called quests, which is basically

20:08

like a user-friendly wrapper of CK proofs.

20:10

So essentially, like, we create a quest,

20:12

which was, I think the best example

20:15

was like, oh, to attend this event,

20:17

you need to meet like three out

20:19

of five of the organizers. So the

20:21

idea here being like, instead of there

20:23

being like a fixed invite list, it

20:26

was just sort of proof of knowing

20:28

the people involved. And instead of calling

20:30

it a ZK proof, we wanted to

20:32

experiment with different terminology and wrap it

20:35

better, just so you didn't need to

20:37

be a ZK person to like get

20:39

it. And I think that was like

20:41

a really fun experiment. I think we

20:44

did a lot of interesting things like

20:46

event invites and scavenger hunts. And we

20:48

also iterated this at East Denver, just

20:50

a much larger version of this in

20:52

collaboration with IYK, like another NFC team.

20:55

Main takeaway from that though is that

20:57

CK felt like a little bit of

20:59

a limiting toolkit in order to build

21:01

like full user facing applications around in

21:04

that like a lot of what you

21:06

can do is roughly like privacy preserving

21:08

access control. You can basically prove you

21:10

meet some threshold and then get access

21:13

to something. And it turns out like

21:15

this often kind of just looks like

21:17

tickets or like gating some like online

21:19

content. And I think we felt a

21:21

little bit limited by having that be

21:24

our only kind of cryptographic tool which

21:26

led to like a lot of future-facing

21:28

experiments at like CK 11 and onwards.

21:30

Cool. At Dev Connect it was that

21:33

where you were actually doing those like

21:35

you must know three out of five

21:37

of the organizer type experiments? Yep. We

21:39

ran a few of them for different

21:41

like parties and events. I was not

21:44

aware and sadly didn't go to any

21:46

of these. I did not know three

21:48

out of five of these organizers clearly.

21:50

I guess so. Yeah, we had like

21:53

200 cards produced. It was actually quite

21:55

a fun sprint. Like we basically were

21:57

able to add baby job job curves

21:59

into these. cards like so basically their

22:02

Java cards you code them with like

22:04

kind of jank Java like not

22:06

full Java like Parfield Java and

22:08

we were able to get some

22:10

valid signatures going and so we were

22:12

able to build a full ZK experience around

22:15

it and yeah there are only 200 of

22:17

them so I think it was like kind

22:19

of hard to to get it to everybody.

22:21

It's okay. I forgive. It was

22:23

a big event. No worries. But it

22:26

sounds cool. I'm kind of sad.

22:28

I missed that part. Let's fast

22:30

forward then to East Denver, 2024. Because

22:33

I mean, I really got to

22:35

know what you guys were doing

22:37

at CK11, which was April. So

22:39

just like a month or two

22:41

after East Denver. What had you

22:43

actually done at East Denver? Yeah,

22:45

I think. Two different things, I

22:47

would say the main experience

22:49

that was available to every

22:52

attendee was related to completing

22:54

these ZK quests, proving that

22:56

you met certain people, that

22:58

you went to certain events

23:00

or had certain experiences, and

23:02

being able to claim prizes

23:04

as a result of that.

23:06

But at the same time,

23:09

we were also working on

23:11

new experiments specifically related

23:13

to private set

23:15

intersection. And this was using

23:17

a BFV library in collaboration

23:20

with John Ojaya, who's a

23:22

long-time collaborator of ours. And

23:24

I think what we've noticed

23:27

is from all these activations

23:29

and all these experiments, it

23:31

takes a while to take

23:33

a new technology in programmable

23:36

cryptography and actually be

23:38

able to productionize it

23:40

successfully. At Denver, we

23:42

first played around with

23:45

this BFV PSI toolkit

23:47

and Vivic spent multiple

23:49

all-nighters trying to get

23:52

it to work. While

23:54

we were building out

23:56

the main app experience

23:58

as well. And I think

24:01

we were able to make a

24:03

proof of concept, but there were

24:05

still some discontinuities in the experience.

24:07

And fast forward to ZK11. That's

24:09

when we actually were able to

24:11

deploy it successfully, and it basically

24:13

worked perfectly during the event. And

24:15

we've noticed like- Yeah, it really

24:17

did. the lessons we learned from

24:19

trying to get it to work

24:21

and it struggling to work those

24:23

carried over and eventually we did

24:25

actually like do it successfully but

24:27

at the same time at ZK11

24:30

we wanted to also experiment with

24:32

with folding, specifically regards to being

24:34

able to prove you met multiple

24:36

people, each one of those being

24:38

an individual proof, but being able

24:40

to fold that together into some

24:42

sort of Spotify wrapped like experience

24:44

for your whole ZK11 journey. These

24:46

were all the people I met,

24:48

these were all the speakers I

24:50

met, and kind of generate one

24:52

big folding proof related to that.

24:54

And we didn't. learn our lesson

24:56

at all. Deploying technology for the

24:58

first time still is a huge

25:01

trouble. Specifically with regards to folding,

25:03

we ran into issues with key

25:05

sizes, basically having to download very

25:07

large keys and you can imagine

25:09

like trying to get, I don't

25:11

know, what was it, like 300

25:13

attendees, 400? Maybe more. There's 550.

25:15

Okay. Right. Okay. 11. Yeah. 550

25:17

attendees to download like tens of

25:19

megabytes of like keys on conference.

25:21

Everyone on the same Wi-Fi. Yeah.

25:23

I don't know if everyone did

25:25

it in CK-11. I feel like

25:27

it was the first time we

25:30

worked together and we did announce

25:32

it at the beginning of the

25:34

event, but a lot of the

25:36

community weren't used to it. By

25:38

ZK-12, I think the percentage of

25:40

attendees using it was actually up.

25:42

Yeah. But yeah. But I do

25:44

want to touch more on the...

25:46

PSI or private set intersection experience

25:48

because I think this was something

25:50

that we learned that was actually

25:52

quite successful. So specifically what this

25:54

allowed you to do is you

25:56

have a list of all the

25:58

people that you've met at the

26:01

conference. These are represented by their

26:03

signatures, signatures from their public keys

26:05

rather. You have a list of

26:07

interests or things that you're interested

26:09

in and when you meet someone

26:11

you can basically privately figure out

26:13

what you have in common. And

26:15

at ZK11, we found that a

26:17

lot of people were really interested

26:19

in trying this out. It got

26:21

received pretty well at the event.

26:23

And people were like, oh my

26:25

God, like I could actually meet

26:27

someone new, connect with them instantly,

26:29

and all my data stays on

26:32

my device, and it's all private.

26:34

So I think that was a

26:36

really good learning from that event

26:38

that even something pretty small, like

26:40

PSI, it seems like a very

26:42

simple primitive. can be quite expressive

26:44

and contrast that with heavier things

26:46

like full-on FHE, like let's say

26:48

you want to do FHEML or

26:50

something. This is quite complicated to

26:52

run. This takes a lot of

26:54

computation and a lot of memory

26:56

and it's quite impractical for a

26:58

lot of day-to-day consumer experiences. In

27:01

contrast, something like PSI is already

27:03

practical today. It works, it works

27:05

on your mobile browser and it

27:07

can actually create new social dynamics.

27:09

So I think that was a

27:11

very important learning for us. I

27:13

think like one other meta point

27:15

here is like this is our

27:17

first time trying like multi-party stuff

27:19

where you took data from like

27:21

multiple people and did some computation.

27:23

ZK is inherently a very single

27:25

player endeavor. There's like a verifier

27:27

obviously so it's kind of a

27:29

multi-party computation. I think some people

27:32

would get angry at that. The

27:34

person with private data is only,

27:36

there's only one of them. The

27:38

Z811 experience opened a lot of

27:40

new doors in terms of like,

27:42

okay, cool, like, what are the

27:44

new social dynamics? we can explore

27:46

when multiple people have private data

27:48

involved. And it's a much less

27:50

explored design space than CK. So

27:52

that's one meta point. And just

27:54

building off the other thing that

27:56

Andrew said, like, yeah, I think

27:58

like a repeated lesson, or like

28:01

kind of a repeated pattern for

28:03

us is like, we very ambitiously

28:05

try to do some new technology

28:07

at event, like X. We like

28:09

completely fail. It like blows up.

28:11

Like Andrew did mention, like for

28:13

Denver, like we tried putting in

28:15

like the private center section, it

28:17

literally like broke the entire app.

28:19

Mind you, we have like 12,000

28:21

NFC cards that we're giving out,

28:23

like in five hours. Oh my

28:25

God. Yeah, the night before. There's

28:27

two a.m. We were doing some

28:29

last minute tests before we went

28:32

to bed. And then it just

28:34

totally broke everything. And so we

28:36

had to like revert the commit,

28:38

clean up a bunch of stuff,

28:40

test another few hours. for me

28:42

tonight. I love the term programmable

28:44

cryptography because it's it makes you

28:46

forget how hard it is to

28:48

program this stuff. That's yeah honestly

28:50

it's true it makes it seem

28:52

a little too easy but I

28:54

think now we have a good

28:56

awareness of like push to try

28:58

to get it for an event

29:01

but like don't be too sad

29:03

if it doesn't work out it'll

29:05

probably work out for the next

29:07

one and just a word of

29:09

advice that like this stuff always

29:11

takes longer than you think it

29:13

will, especially if you're the first

29:15

person, like actually getting into the

29:17

weeds of this stuff and trying

29:19

to put it on like a

29:21

mobile phone or a consumer device.

29:23

Yeah. It's great to see like

29:25

your cryptographic like palette sort of

29:27

expanding, but at the same time

29:29

you're helping everyone else, right? Because

29:32

now that you guys have tried

29:34

this, someone else can come in

29:36

and try like more PSI stuff.

29:38

Yeah, totally. That's definitely, I think

29:40

one of my favorite parts of

29:42

my favorite parts of the field

29:44

of the field is how open

29:46

source friendly it is, like how

29:48

much of a collaboration it all

29:50

feels like. Well, just kind of

29:52

like one big happy family, like,

29:54

you know, at least some corners.

29:56

Some siblings bigger, but it's a

29:58

big family. It's a big family,

30:01

like everyone's building off each other.

30:03

And I really like that, like it comes

30:05

in contrast to like ML, for example, which

30:07

is like a lot of close source work

30:09

happening. And then you see like Deep's Heek

30:11

come out and everyone's like, this is

30:13

such a gift to mankind. And it's

30:16

like, well, yeah, we've just been doing

30:18

that in cryptography land. Yeah, I think

30:20

in program book cryptography, a lot of

30:22

people understand that siblings might fight,

30:24

but we're all fighting a much bigger

30:26

battle. the grand scheme of things like

30:28

people are either going to adopt this

30:31

stuff or they're not and it's going

30:33

to take a lot of effort to

30:35

make it happen. There's a lot of

30:37

inertia and so I think it really

30:39

is important for teams to collaborate and

30:41

understand that either everyone wins or no

30:44

one wins. There is one thing I

30:46

wanted to bring back about the way

30:48

you did PSI. So PSI has sort

30:50

of a history and literature of being

30:52

a pure like two-party thing like you

30:55

have two parties. finding an intersection.

30:57

And I think in your case you

30:59

guys did something quite interesting where you

31:01

had like a server helping users

31:03

compute their PSI. Could you tell us a bit

31:05

more about how that worked? Yeah, for sure.

31:08

Essentially what we did was probably a

31:10

little bit too overpowered of a technique,

31:12

but it like works pretty nicely for

31:14

our use case, which is a multi-party

31:16

FHE or like threshold FHE. So what

31:18

goes on here is like, in this case

31:20

it's two parties. So let's say like

31:22

me and Andrew doing a private set

31:24

intersection. We essentially, like, together, perform

31:27

an initial NPC where we create,

31:29

like, a shared public key. So,

31:31

between us, we can see the public

31:33

key, but neither of us knows

31:35

the corresponding private key. It's kind

31:37

of like a split between us.

31:39

And so with this public key,

31:41

what we can do is we

31:43

can encrypt our data fully homomorphically

31:45

to the shared public key. So now

31:47

me and Andrew have like encrypted

31:50

versions of our two sets of

31:52

contacts, interests, interests, whatever. And

31:54

so now what we can do

31:56

is we can run a comparison

31:58

in FHE and... Like usually this

32:01

is actually a very costly operation,

32:03

so the nice thing here is

32:05

you can like offload it to

32:07

a server in a way that

32:09

doesn't actually leak any privacy. The

32:11

server is essentially just doing a

32:13

bunch of like FFTs for you

32:15

and then gives you back the

32:17

encrypted result. And then at the

32:19

end, me and Andrew come together

32:22

and we do a decryption MPC,

32:24

where essentially we decrypt the result

32:26

of this computation. without ever actually

32:28

surfacing the kind of shared private

32:30

key. We just just decrypt it

32:32

and thus only get the result

32:34

to the computation. And this technique

32:36

works for stuff past private center

32:38

sections. You can like do any

32:41

computation with this method. But you

32:43

relieved the user devices from the

32:45

heavy workload. Exactly. Yeah. And that's

32:47

what I found super interesting. Yeah,

32:49

like essentially the user is just

32:51

responsible for encrypting and creating the

32:53

shared public key. which tend to

32:55

be very light operations in general.

32:57

And so, yeah, you basically do

32:59

the setup and then a server

33:02

just handles all the heavy lifting.

33:04

And also when we started exploring

33:06

PSI, we didn't have a good

33:08

sense of whether or not PSI

33:10

was going to be the main

33:12

thing that provided a lot of

33:14

application facing value. So I believe

33:16

that a lot of the... Taylormade

33:18

PSI schemes are specifically useful for

33:20

private set intersection, but using multi-party

33:23

FHE gave us the flexibility to

33:25

say, oh, maybe we can play

33:27

around with different types of primitives

33:29

here as well, maybe we can

33:31

expand this to different things if

33:33

they prove to be valuable. Yeah,

33:35

I think another shout out here

33:37

to John Magaya, again, like without

33:39

him, like none of this would

33:42

be possible. Specifically, he has a

33:44

really great toolkit called Phantom Zone.

33:46

It's like an MIT licensed FHG

33:48

VM, a different licensing to other

33:50

sort of FHG tooling, and it

33:52

basically is focused. on this multi-party

33:54

FHE techniques. So I would highly

33:56

recommend in Devs to check that

33:58

out. Really cool library can definitely

34:00

build some fun stuff with it.

34:03

I remember when you first approached

34:05

me for ZK11, I want to

34:07

say, like, I was really excited

34:09

about the ZK, the fact that

34:11

there'd be kind of like a

34:13

tool on site, but I was

34:15

also a little bit wary because

34:17

in my, like, in the past,

34:19

long before I met you. had

34:21

sort of seen a lot of

34:24

these experiments go very, very wrong

34:26

at events. And when you're running

34:28

an event, obviously you're under a

34:30

lot of pressure in general, you

34:32

want to make sure that anything

34:34

that you're bringing will add to

34:36

the experience of the participants. So

34:38

I already knew you, Vivic, and

34:40

I guess that maybe is why

34:43

I took that chance, but I

34:45

remember after the event being so

34:47

happy that we had. But it

34:49

really was unlike, you know, I

34:51

really didn't know what to expect

34:53

going into it. And so obviously

34:55

when we started speaking again about

34:57

CK 12, I was like very,

34:59

very excited to work together again.

35:01

I know that between CK11 and

35:04

CK12, there was another experiment that

35:06

you ran though. So let's talk

35:08

a little bit about that frontiers.

35:10

Yeah, I'll step in real quick

35:12

and not assuage any of your

35:14

fears. Maybe you didn't know this,

35:16

but so. Quick shout out, we

35:18

also worked with Jack and Ian

35:20

from Mock 34. They work on

35:22

a bunch of CK programmable cryptography

35:25

tools as well. They helped us

35:27

integrate folding into the ZK 11

35:29

experience. And at the end of

35:31

the day, like folding worked intend.

35:33

It just was like very poorly

35:35

performing due to key sizes during

35:37

the actual event. We got it

35:39

working. in production 20 minutes before

35:41

the event started. That totally does

35:44

terrifying. You never told me that.

35:46

I probably should not have told

35:48

you that, but now you know.

35:50

Amazing. Things come down to the

35:52

wire. Wow. One last ZK-11 anecdote

35:54

for me is how one hour

35:56

before the end Anna realizes that

35:58

not only is there like all

36:00

this cool ZK and PSI functionality,

36:02

there's a leaderboard. Yeah, that's true.

36:05

So it's only a nice like,

36:07

I cannot lose my own event.

36:09

So she's running around tapping everyone's

36:11

cards. Right. And I think you

36:13

won, right? I did win in

36:15

Z-K-11. I wasn't just that I

36:17

cannot lose my own event. I

36:19

think it was just that I,

36:21

yeah, I realized that. No one

36:23

was ranking, no one was competing,

36:26

and so I started to get

36:28

into it. For ZK12, we did

36:30

fix that by actually offering incentives

36:32

and bringing back the concept of

36:34

prizes, which you had had at

36:36

East Denver and we didn't think

36:38

of for ZK11. But I know,

36:40

yeah, let's talk about that in-between

36:42

experiment with frontiers and this hiring

36:45

experiment, because I thought of that

36:47

as, like, when we started talking

36:49

about ZK12, you were telling me

36:51

about this, it was so... of

36:53

exactly these tools that you're working

36:55

with. So let's talk about that.

36:57

Yeah, there's actually a really nice

36:59

tie in here with CK11, which

37:01

is that. Frontiers is like a

37:03

paradigm-sponsored event. It's like focused on

37:06

open-source rust and Ethereum applications, and

37:08

Georgeos is like the main curator

37:10

of that event, and he actually

37:12

first saw our technology at CK11.

37:14

And that's why he reached out.

37:16

He reached out in our public

37:18

channel. He was like, let's do

37:20

something. And then we were like,

37:22

sure, not really expecting it to

37:24

go anywhere. But yeah, it ended

37:27

up being a really good collaboration.

37:29

And so also appreciate UNICO, like

37:31

digging a bit deeper into multi-party

37:33

FHE, because that comes into a

37:35

play, actually a lot more in

37:37

this activation. So specifically we wanted

37:39

to take that toolkit we were

37:41

using for just private set intersection

37:43

and we wanted to expand it

37:46

to like a more general purpose

37:48

function. And this is right around

37:50

the time where John Magaya opened

37:52

up Phantom Zone as like a

37:54

full, I guess not a full

37:56

VM, it's closer to a circuit

37:58

builder, where you could just write

38:00

more. general purpose multi-party FHE code.

38:02

And so we basically were looking

38:04

into matching as like a core

38:07

verb or core action of like

38:09

how can we match people together

38:11

based on private information in a

38:13

way that like is more efficient

38:15

or gets to use kind of

38:17

like more data about people. And

38:19

in the context of like a

38:21

professional event like frontiers we kind

38:23

of stumbled into hiring as

38:26

an interesting example. I think

38:28

there's two main... interesting

38:30

private information like pieces here. The first

38:32

is like salary. So as a recruiter

38:34

there's some like usually like upper bound

38:36

of salary or like ideally want to

38:38

pay somebody and as an employee there's

38:40

some lower bound that you're like willing

38:43

to work for and you'd be excited

38:45

to work for. And so like discovering

38:47

overlap there without having to give

38:49

up that information we thought could

38:51

be really interesting. So instead of

38:53

there being like a lengthy negotiation

38:55

and potentially even just figuring out

38:57

later in the process that. this

38:59

isn't going to pay the... Yeah, it just

39:02

like is not enough basically. You can

39:04

sort of like front load that without

39:06

actually giving up the information. And the

39:08

second bit of privacy that I think

39:11

is useful is like, as a job

39:13

candidate, if you are already employed somewhere, but

39:15

maybe like, you know, are trying to

39:17

figure out what else is out there,

39:19

or maybe are unhappy with your job

39:21

and like want to, you know, start

39:23

applying other places, you can just post

39:25

on LinkedIn. Like, I hate my

39:27

job, get me out of here.

39:29

And so we were thinking, yeah,

39:31

like it could be interesting to

39:33

basically like privately attest to the

39:35

fact that you're looking for new

39:37

jobs, but you only get notified

39:40

of jobs that kind of match

39:42

your ideal circumstances. So you can

39:44

kind of like, it's almost like

39:46

manifesting in a way, you're like

39:48

kind of like manifesting your ideal

39:50

job, like the salary you want,

39:52

the people you want to work

39:54

with, and you're just putting it out

39:56

to end. using the same exact scheme

39:58

where the sort of like heavy lifting

40:00

of this computation is done on a

40:02

server, the users are just responsible for

40:05

encrypting their data. And yeah, it was,

40:07

it was like successful. I think like,

40:09

I got a few, like, okay, I

40:12

put in a fake application, because, you

40:14

know, I'm happy, bad cursive, but I

40:16

wanted to test out the feature, so...

40:19

Like the job market catfish. I think...

40:21

Exactly. Which actually raises an interesting point

40:23

around verifiability that we can get into,

40:26

but... For now I basically put myself

40:28

as like the like ideal candidate like

40:30

willing to work for like 10,000 a

40:32

year like willing to work like 60

40:35

hours a week you know like just

40:37

the most ridiculous ends of each of

40:39

these spectrum so I could ensure some

40:42

matches. Yeah I got some job offers

40:44

you know or like not offers but

40:46

like some things that matched and so

40:49

yeah that was a really fun experiment

40:51

I think exploring yeah what you could

40:53

do with a bit more than jest

40:56

like. This like you're really replacing the

40:58

recruiter here too. Like usually you have

41:00

a quote unquote trusted third party who

41:03

has all of this information that recruiter

41:05

will know that you're looking but your

41:07

employer doesn't know they'll know often your

41:09

salary bounds but like the potential hiring

41:12

company won't really know exactly what your

41:14

your your range is and this is

41:16

sort of like directly bringing the two

41:19

parties together in that private way where

41:21

they can find each other. That's so

41:23

cool. Yeah, absolutely. Like I think this

41:26

comes up a lot in our explorations

41:28

as like a lot of this stuff

41:30

just tends to be much more direct

41:33

with the consumer and you can cut

41:35

out the middleman. And it's interesting, like

41:37

there's attention here where we talk to

41:39

some of the recruiters at the event

41:42

about like what they thought about it.

41:44

And in some ways it's like, in

41:46

some ways, it's like kind of, yeah,

41:49

not synergistic, but I think one positive

41:51

framing that one of the recruiters found

41:53

is like... Actually, it's nice because someone

41:56

like Georgeos can just put out a

41:58

private job description and he himself can

42:00

match with really good people. instead of

42:03

their needing to be kind of like

42:05

a middleman. And I think you can

42:07

actually create an experience that's like a

42:09

little bit less kind of extractive or

42:12

kind of, I don't know, like clearly

42:14

there's a bunch of recruiter sharks coming

42:16

around trying to hire people. It's less

42:19

of head hunting and more just like

42:21

peer-to-peer discovery of good overlap. And it

42:23

can just be done directly between engineers.

42:26

Yeah, it's interesting. There's like positive and

42:28

like negative framings for recruiters. In this

42:30

experience, was there also like a communication

42:33

feature or people could actually send something

42:35

through this? Or was it just to

42:37

the matching and then they had to

42:39

kind of find each other elsewhere? I

42:42

think we just like, once you match

42:44

you just got their telegram or their

42:46

email, like you could just choose how

42:49

you wanted to be contacted and then

42:51

you could just reach out on those

42:53

platforms. Okay. But is this still like

42:56

you're going up to somebody and like

42:58

tapping a card? Are you connecting? Are

43:00

you standing in front of the person

43:03

when you realize it's a match? Or

43:05

is it like there's a pool of

43:07

people and you're getting matched kind of

43:09

willy-nilly? Yeah, in this experiment, you actually

43:12

had to go up and tap people

43:14

and build your own social graph. And

43:16

I think the idea there was that

43:19

if you met. people in person you

43:21

would be able to talk with them

43:23

have sort of a deeper relationship and

43:26

it would be more of a warm

43:28

referral. I think this relates to the

43:30

question of data import which is a

43:33

big question we spend a lot of

43:35

time thinking about in our work because

43:37

a lot of these matching experiments work

43:40

well when there's really rich data available

43:42

about the people who are getting matched.

43:44

And so in this case, the data

43:46

here is the social graph, as well

43:49

as the form inputs for your hiring

43:51

preferences. And you could imagine that in

43:53

an experiment outside of a conference, you

43:56

could get the social graph through other

43:58

means, be it maybe like ZK email

44:00

from your calendar invites, ZKTLS from maybe

44:03

your LinkedIn connections. could find other ways

44:05

to import this data. And you could

44:07

also perhaps do that with job searches.

44:10

You could import data about where you

44:12

worked in the past, for example, and

44:14

have that be filled in automatically. And

44:16

so I think being able to easily

44:19

import data is a question that's very

44:21

relevant to making this experience richer for

44:23

matching. And it's something that we have

44:26

to spend a lot of time getting

44:28

the UX right because people filling out

44:30

forms and having to manually meet people

44:33

in person to build that social graph

44:35

that's not going to be the long-term

44:37

solution and so still an open question.

44:40

And it kind of addresses the catfishing

44:42

concerns right like if you have verifiable

44:44

sources of data? Yeah definitely. I was

44:46

this whole experiment still built on PSI

44:49

only or did you need other primitives

44:51

in there? Yeah so we actually basically

44:53

just wrote a full Boolean circuit. Specifically

44:56

under the hood, it's using few F-H-E-W,

44:58

which kind of takes like Boolean circuits

45:00

as input. The main kind of new

45:03

functionality we needed was a comparator, like

45:05

a greater than, because we wanted to

45:07

compare the recruiter salary upper bound with

45:10

like the candidate's lower bound. And everything

45:12

else ended up just being like a

45:14

bunch of aunts and oars and stuff.

45:16

Like, you know, is there overlap on

45:19

an experience level? Like, I only want

45:21

to hire people with like a bachelor's

45:23

or higher. and just making sure that

45:26

that like take the end of that

45:28

bit vector and the candidate's bit vector

45:30

is there like overlap basically. Cool. So

45:33

I just remember when we were planning,

45:35

when we were talking about the ZK12

45:37

activation, we had actually thought a bit

45:40

about this hiring experiment as maybe also

45:42

something we could run there. We decided

45:44

against it. I remember us kind of

45:46

going over a few iterations of what

45:49

Like what could we get out of

45:51

this moment with these people? What can

45:53

they get out of that as well?

45:56

Like where would they get a kick

45:58

out of this? I think what we

46:00

ended... up landing on was having sort

46:03

of preferences again because in ZK11

46:05

I just remember it being like

46:07

number of talks and number of

46:09

contacts. at CK 12 we actually

46:11

had something more like what are

46:14

you into what are you like

46:16

what are you not like or

46:18

I think the question was which

46:20

of these are overrated but people

46:22

got confused on the scale was

46:25

like was three overrated or was

46:27

one overrated yes I got a

46:29

little bit confused how to vote

46:31

for it which led to some

46:33

great conversations I thought That's on

46:35

me. But yeah, but maybe you

46:37

share a little bit about how

46:40

this differed. What were you using

46:42

for this? I think ZK-12 for us

46:44

was more of an ideological

46:46

progression. I think during

46:48

the actual app experience, we

46:50

did do more of these

46:53

hot takes finding more rich

46:55

ways for people to actually

46:57

connect with each other based on

46:59

more meaningful data. I think what

47:01

we realize is the questions you

47:03

ask people and what they're responding

47:05

to really matters. You want two

47:07

people to be able to meet

47:09

each other and when they learn

47:11

each other's preferences, actually be like,

47:13

oh, that's cool. This is a

47:15

conversation now, or I'm really excited.

47:17

Why do you think this? So

47:20

I think we learn a lot

47:22

in that direction. But more specifically,

47:24

we... kind of had two new

47:27

concepts that we were playing

47:29

around with, which were just

47:31

framings of things we had already

47:33

done, specifically digital pheromones

47:36

and narrow casting, and narrow casting

47:38

also being relevant to Trinity, which

47:41

was a new MP scheme that

47:43

we were looking into at the

47:45

time. Maybe I can talk a

47:48

little bit about digital pheromones and

47:50

Vivic you can take narrowcasting with

47:52

Trinity. For us we were thinking

47:55

about, okay, like what is actually

47:57

going on here? We have this

48:00

to match multiple people on private

48:02

information, sort of build more meaningful

48:04

connection through these interactions, how can

48:06

we describe what is happening here?

48:08

And I think what we ended

48:10

up on is this idea called

48:12

digital pheromones, specifically pheromones being these

48:14

discrete. chemical signals that you're constantly

48:17

sending out in search of human

48:19

compatibility or a mate or whatever

48:21

it might be and digital pheromones

48:23

are the digital equivalent. So they

48:25

can be applied to many different

48:27

settings. It could be online dating,

48:29

could be private hiring, but crucially

48:32

what ties this all together is

48:34

the idea that they are private

48:36

and safe. So when I have

48:38

like my physical pheromones, like this

48:40

is not something that I'm just

48:42

yelling out or screaming into public

48:44

space, these are much more discreet

48:46

and these are things where, okay,

48:49

when you actually interact with me,

48:51

then they get surfaced in passive

48:53

ways. And I think digital pheromones

48:55

is kind of a similar concept

48:57

where These are all private signals.

48:59

They're all based on potentially sensitive

49:01

data that remains private to you.

49:04

But the only time you actually

49:06

realize something is going on is

49:08

if you are connected with someone

49:10

or if you do meet someone

49:12

that matches your signals. And so

49:14

I think we thought this was

49:16

really cool because now there's this

49:19

way. for people to just be

49:21

walking around going about their daily

49:23

life, potentially walking around a conference

49:25

or event, just doing what they

49:27

are normally planning on doing, but

49:29

behind the scenes, all this matching

49:31

is going on and all this

49:33

compatibility and connection is being found,

49:36

but none of it is being

49:38

done. on a big server that

49:40

sees everything, for example. Yeah, this

49:42

reminds me of contact tracing. Yeah.

49:44

When so the pandemic broke out,

49:46

you know, there was sort of

49:48

two worlds. There was this dystopian

49:51

world where one government would put

49:53

out an app that tracks everyone

49:55

and knows all the interactions of

49:57

every people. And there was sort

49:59

of the utopian version of that

50:01

a lot of economics were pushing

50:03

for. And it's something like this,

50:05

where you would disperse your scent.

50:08

If you were positive, you could

50:10

send out a signal and those

50:12

who picked up your scent would

50:14

be alerted, but privately, only they

50:16

would know. So I'm curious to

50:18

see where this goes, like what

50:20

other sort of use cases we

50:23

can find for this. Yeah, I

50:25

think probably was the first kind

50:27

of wide scale deployment of something

50:29

like this. And then the hiring

50:31

thing we just talked about is

50:33

very much kind of like also

50:35

a proto version of the pheromone

50:38

here literally just being like, like,

50:40

I want this job. a private

50:42

manifestation kind of. It's like you

50:44

kind of put out what you

50:46

want for yourself or like who

50:48

you are and where you want

50:50

to be and you can sort

50:52

of just discover matches and overlap.

50:55

You know, if this is to

50:57

be something that is like fully

50:59

akin to like natural pheromones, like

51:01

chemical pheromones, I do think cryptography

51:03

is the right substrate for it

51:05

in that it is like very

51:07

neutral, it's very peer-to-peer, there's no

51:10

dependence on a centralized server. Like

51:12

I think this is like one

51:14

of the rare things where a

51:16

TEE does not kind of fully

51:18

subsume it because it I think

51:20

does fundamentally change the nature of

51:22

something like this where it's you

51:24

do have to go through some

51:27

service that is like providing this

51:29

matching versus it being like yeah

51:31

literally my phone is letting out

51:33

signals to the people around me

51:35

or like on an online platform

51:37

and they can just parse it.

51:39

So yeah I think this was

51:42

a really fun kind of framing

51:44

and wrapping for a lot of

51:46

our technology that made it easier

51:48

to explain and think about. And

51:50

we actually picked it up at

51:52

D-Web, a decentralized web camp in

51:54

California, just from some conversations. with

51:57

kind of things that just need

51:59

to be happening in the background.

52:01

Oh, another concept that you've been

52:03

very excited about is this idea

52:05

of local first hardware. So I

52:07

think this is a great fit

52:09

for digital pheromones because these are

52:11

kind of things that just need

52:14

to be happening in the background.

52:16

And you could run it on

52:18

your phone, but the problem is

52:20

Apple is very restrictive. about the

52:22

things you can do on a

52:24

phone. That is just their philosophy.

52:26

I think everyone who has developed

52:29

on iPhone knows this and they

52:31

are one of the most fun

52:33

products to use as a user

52:35

and one of the worst companies

52:37

to deal with as a developer,

52:39

as many people know. Yep, exactly.

52:41

So there's limits to what you

52:43

can do with an iPhone with

52:46

existing devices, but I think a

52:48

really cool concept is having open

52:50

hardware. that like somehow is connected

52:52

to your sensitive data but keeps

52:54

it securely on that hardware and

52:56

you can actually walk around maybe

52:58

wearing a pendant or something that's

53:01

just constantly sending out these signals

53:03

to other people and finding compatibility

53:05

that way just in your normal

53:07

day-to-day life. And so I think

53:09

that's a very like interesting area

53:11

of exploration. If anyone is very

53:13

knowledgeable about hardware and wants to

53:16

work on this stuff, please let

53:18

us know. So I think another

53:20

concept that we introduced at CK

53:22

12 was narrowcasting, which essentially is

53:24

like the opposite of broadcasting. You

53:26

can basically pick like specific criteria

53:28

that you want your message to

53:30

be received by and kind of

53:33

the cryptographic version of this is

53:35

that like literally it could only

53:37

be decrypted by people that fit

53:39

this criteria. So like witness encryption

53:41

is kind of like the purest

53:43

version of this, but you can

53:45

build something with similar affordances using

53:48

a mix of like ZK and

53:50

MPC. Another related result to this

53:52

that we published for the first

53:54

time at CK-12 was a new

53:56

scheme called Trinity, which was developed

53:58

in-house, basically mixing like garbled circuits,

54:00

plunk, and then a very recent

54:03

result using KZG witness encryption, really

54:05

beautiful paper, would recommend people to

54:07

check it out. And yeah, like

54:09

one really nice affordance is that

54:11

you can build something like narrowcasting

54:13

really easily because it has... Really

54:15

low rounds of interaction between in

54:17

this case the two parties? This

54:20

is like a general flaw of

54:22

MPC if you want to build

54:24

apps with it There's like usually

54:26

a lot of back and forth

54:28

and it's pretty annoying for an

54:30

app developer to deal with if

54:32

you're trying to make a consumer

54:35

app Just for the reason that

54:37

people put their phones away like

54:39

literally just like once the phone

54:41

goes away just like no longer

54:43

can respond to requests as easily

54:45

like Apple exposes some things was

54:47

this part of the motivation for

54:49

your server-aited MPC model? Yes, absolutely.

54:52

Like that's another place where multi-party

54:54

FHG shines is you can delegate

54:56

a bunch of stuff to the

54:58

server and so you don't need

55:00

to be online for those phases.

55:02

But ultimately it still has like

55:04

about four rounds of back and

55:07

forth interaction because now I need

55:09

to send something to the server,

55:11

server needs to get back to

55:13

you, you need to then do

55:15

this decryption. Whereas Trinity is basically

55:17

just one round of interaction. Like

55:19

I basically just send like an

55:22

encrypted email to somebody. and they

55:24

can decrypt it if they match

55:26

some criteria. And the other nice

55:28

thing that Trinity adds in is

55:30

actually verifiability of input. So this

55:32

relates to Nico, what you're talking

55:34

about with catfishing. Like, this is

55:36

kind of a general problem with

55:39

MPC is like, you know, if

55:41

the inputs aren't like verified or

55:43

signed in some way, then you

55:45

can basically say whatever you want,

55:47

right? You can sort of like,

55:49

discover the other person's private information's

55:51

private information by just like repeatedly

55:54

querying them. And so... Trinity also

55:56

basically adds a really like clean

55:58

tie between like zero knowledge proofs

56:00

which can prove various properties of

56:02

your input data, and like a

56:04

multi-party computation, which in this case

56:06

is done with garbled circuits. Cool.

56:09

Moving on to your next events, Andrew,

56:11

you and I jumped on a call

56:13

after ZK-12 and you were at Edge

56:15

City. Were you running something similar to

56:17

like Zoo Pass in Edge City? What

56:19

was, what were you working on there?

56:21

I think, first off, shout out

56:24

to Janine and to Moore

56:26

and the rest of the

56:28

Edge City team for... letting

56:30

us do something throughout the

56:32

whole month at their event.

56:34

I think they were very

56:36

gracious and kind in supporting

56:38

all of the hitches along

56:40

the way as we were

56:42

developing this. At Edge City

56:44

Lana we built an app

56:46

called Accursive Connections and the

56:48

idea was to better connect

56:50

people within this specific sphere

56:52

of Edge City Lana. And

56:54

so some things were similar

56:56

related to people being able

56:58

to form their own social

57:01

graphs from NFC tapping where

57:03

everyone had a wristband, but

57:05

some things were also different

57:07

in that we had matching

57:09

that was more specific to

57:11

the Edge City community. So

57:13

one big thing at Edge

57:16

City was the Tensions game,

57:18

I think timber ran that,

57:20

and the Tensions game was

57:22

hot takes where you could

57:24

fill out your opinion on these

57:26

hot takes and when you

57:28

actually met someone you would

57:30

figure out whether or not

57:33

you agreed on some of

57:35

them but more importantly which

57:37

ones you disagreed on and

57:39

within the Edge City community

57:41

this was a big topic

57:43

of conversation and another interesting

57:45

experiment we tried out at

57:47

Edge City was we wanted

57:49

to see if we could

57:51

match people within a specific

57:53

event. So we chose the Halloween

57:55

party that they were throwing and

57:58

we wanted people to be able

58:00

to walk into a room, kind

58:02

of upload their their preferences here,

58:04

and within the event to basically

58:06

find who else who is there

58:08

that they might be compatible with

58:10

or might want to spend more

58:13

time with after the party. Like

58:15

a dating app? We framed it

58:17

more as a friendship finding, but

58:19

I think this could easily be.

58:21

But really, it was dating. Dating

58:23

up in certain contexts. Yeah. We're

58:25

just surfacing the connection, however far

58:27

that takes you. Cool. And I

58:29

think there are a few interesting

58:32

takeaways here. Number one is just

58:34

a very practical lesson of... people

58:36

don't want to take out their

58:38

phones at events. Oh, yeah. And

58:40

I think this is actually what

58:42

inspired some of the hardware directions

58:44

as well. The idea that you

58:46

can just have hardware that's just

58:48

like passively there that maybe lights

58:51

up when you find connection, that

58:53

would make much more sense at

58:55

this kind of event as opposed

58:57

to people needing to check out

58:59

their phones. And another. related learning,

59:01

which I guess we'll talk about

59:03

more, is specifically we wanted a

59:05

lot of the computation to be

59:08

non-interactive. So we wanted to be

59:10

able to surface the matches directly

59:12

within the event without people having

59:14

to check their phones. And so

59:16

this naturally surfaced the idea of

59:18

using T's. because the problem with

59:20

multi-party FHE is you do have

59:22

this interactive decryption step. So we

59:24

thought of this example as one

59:27

where you could enter room and

59:29

you understand that the data you

59:31

share within that room or within

59:33

that ephemeral context will go to

59:35

secure hardware because it needs to

59:37

match you non-interactively. cryptographic, purely like

59:39

software, mathematics. We have these hardware

59:41

tools, some trust them, some don't.

59:43

They kind of fulfill a lot

59:46

of the functionality we want to

59:48

use cryptography for. And so my

59:50

question is, as a design team,

59:52

where do you draw the line

59:54

between staying in pure cryptography or

59:56

reaching out for the T's? Yeah,

59:58

it's actually a question that we've

1:00:00

grappled with a lot this past

1:00:02

year, and there's a lot of

1:00:05

different angles to look at it

1:00:07

from. One angle is like funding,

1:00:09

so we... So far I've been

1:00:11

funded by privacy and scaling expirations

1:00:13

at the Ethereum Foundation with grants

1:00:15

and so like they support programal

1:00:17

cryptography research so a lot of

1:00:19

our focus has been in that

1:00:21

direction because that's the community and

1:00:24

ecosystem we've been a part of

1:00:26

and it's been very synergistic in

1:00:28

that way in that we're able

1:00:30

to take technology that's developed in-house

1:00:32

and then find you know like

1:00:34

product facing uses of it. And

1:00:36

I think yeah as long as

1:00:38

we're funded that way it will

1:00:41

always be a priority to explore

1:00:43

the stuff. Another angle here is

1:00:45

like user first and I think

1:00:47

like something that we have yet

1:00:49

to identify internally and as far

1:00:51

as I can tell also externally

1:00:53

is like when people care about

1:00:55

using the full cryptographic version of

1:00:57

something versus something like secure hardware

1:01:00

ultimately it doesn't really matter what

1:01:02

like purest cryptographers think is correct

1:01:04

if you're trying to build stuff

1:01:06

for people it just matters what

1:01:08

they think. At least if you

1:01:10

want people to use your stuff,

1:01:12

if you don't care, then I

1:01:14

guess, you know, make your perfect

1:01:16

utopia, I guess. And I think

1:01:19

like, ultimately there's tradeoffs to using

1:01:21

this stuff. And I think, I

1:01:23

think there's like a common phrase

1:01:25

that comes up in cryptography. So

1:01:27

girls like, oh, it's just dependent

1:01:29

on math. And like, for some

1:01:31

reason in this world, that's cool

1:01:33

and makes sense. But I think

1:01:35

for a lot of people, this

1:01:38

is just intuition, like, like, like,

1:01:40

like, that can't see your things.

1:01:42

Like I feel like that just

1:01:44

clicks a bit better. Or like

1:01:46

at least I could see it

1:01:48

clicking better. And so yeah, like

1:01:50

one thing we definitely want to

1:01:52

clarify. more in 2025, both for

1:01:55

ourselves and for the external community,

1:01:57

is like, yeah, when do people

1:01:59

care? So running some experiments with

1:02:01

TE is running some stuff with

1:02:03

pure MPC, and just getting some

1:02:05

user data on like, how do

1:02:07

you respond to the copy around

1:02:09

this stuff? Because ultimately. You know,

1:02:11

an average person is not going

1:02:14

to verify a TE attestation or

1:02:16

like look at the verification key

1:02:18

for this proof. They're completely trusting

1:02:20

you and the sort of copy

1:02:22

and explanation that you give them

1:02:24

and like the sort of feeling

1:02:26

the app elicits. So basically, like

1:02:28

my answer is like to be

1:02:30

determined what balance we're going to

1:02:33

use for this and we want

1:02:35

it to just be very user-driven

1:02:37

versus like what we think is

1:02:39

right. And yeah, excited to share

1:02:41

with the community, the findings we

1:02:43

get there. Yeah, super curious to

1:02:45

see results. Like I think a

1:02:47

lot of people are either committing

1:02:49

to one because they believe in

1:02:52

it But I think I've seen

1:02:54

less of a kind of like

1:02:56

Let's try them both and see

1:02:58

what happens Yeah, one way I've

1:03:00

been thinking about this is in

1:03:02

terms of the hierarchy of trust

1:03:04

when it comes to data privacy

1:03:06

and ownership So you can think

1:03:09

about the lowest level of trust

1:03:11

a user has in an application

1:03:13

is one where the application just

1:03:15

like makes no claims to keep

1:03:17

your data private and just like

1:03:19

blatantly will sell your data and

1:03:21

everyone knows about it. The level

1:03:23

that comes above that is one

1:03:25

where the application is just like

1:03:28

trust me bro I know what

1:03:30

I'm doing I'm not gonna like

1:03:32

sell your data or anything. After

1:03:34

that comes regulatory trust. An example

1:03:36

here in a different topic is

1:03:38

like licenses. Like people trust licenses

1:03:40

because you can like eventually bring

1:03:42

someone to a court of law

1:03:44

if they violate the license. And

1:03:47

at the end of day you

1:03:49

like have trust in that system

1:03:51

because of the regulatory mechanism that

1:03:53

underlays it. And then one above

1:03:55

that I would put secure hardware

1:03:57

hardware. where you maybe don't have

1:03:59

to trust the application developer as

1:04:01

much. if they do remote attestation,

1:04:03

if they open source their code

1:04:06

and everything, but now you're trusting

1:04:08

Intel, you're trusting like state actors

1:04:10

to not break into this, maybe

1:04:12

you're trusting a $150,000 microscope or

1:04:14

whatever the cost is for that.

1:04:16

But then the top of that,

1:04:18

or rather the second from the

1:04:20

highest, is cryptography. This is like.

1:04:23

trust in math and the real

1:04:25

beauty of cryptography in the trust

1:04:27

hierarchy is that you can verify

1:04:30

it from first principles like Intel

1:04:32

is is closed source like or

1:04:34

SDX TDX like you know you

1:04:36

can't go into every single layer

1:04:39

of depth and check it yourself

1:04:41

but theoretically you could do the

1:04:43

same with MPC with Fache and

1:04:45

actually go in and make sure

1:04:48

everything checks out and that your

1:04:50

data is safe. But I do.

1:04:52

But yes, yes, yes. Yeah, the more

1:04:54

practical framing is that you trust a

1:04:56

bunch of smart people with a lot

1:04:58

of degrees to actually go in there

1:05:01

and verify this and tell you like,

1:05:03

hey, it's good. Which is also why

1:05:05

cryptography really just has to be paired

1:05:07

with open source if you can't go

1:05:09

in and verify it. It kind of

1:05:11

makes no sense to be going with.

1:05:13

But there is like a sneaky, like,

1:05:16

I don't know, it technically belongs in

1:05:18

the top of the hierarchy, but it's

1:05:20

like somewhat... restricted to a certain design

1:05:22

space, which is just local first. You

1:05:24

can think about this in terms of

1:05:26

LLLM's processing your data. If an LML

1:05:29

wants to process your data and it's

1:05:31

running entirely on your machine, that is

1:05:33

arguably like the highest level of trust.

1:05:36

You know, you can just turn your

1:05:38

computer offline and you can just run

1:05:40

everything locally and that's great. But that's

1:05:42

restricted to single player applications. Like once

1:05:45

you want to get into multi-party computation,

1:05:47

then you have to do either cryptographyography

1:05:49

or secure hardware. So I do think

1:05:51

like the big open question from

1:05:54

a user first perspective is

1:05:56

what applications are best fit

1:05:59

for which level of trust and

1:06:01

how do users actually perceive these things?

1:06:03

And I think to answer your question,

1:06:05

cryptography is really important for very sensitive

1:06:08

data. Anything that might involve like medical

1:06:10

data, anything that people really might not

1:06:12

want to share with the server, I

1:06:15

think that's where cryptography can really shine.

1:06:17

I want to talk about the last

1:06:19

event of the year, I think, for

1:06:22

you. I mean, if we just go

1:06:24

back, this is, I mean, 2024 sounds

1:06:26

like it was just a marathon of

1:06:28

events and activations for the two of

1:06:31

you or for the group. But let's

1:06:33

talk a little bit about DevCon and

1:06:35

what you did there. There's like an

1:06:38

art gallery. I remember you telling me

1:06:40

about this before you did it. How

1:06:42

did it go? Yeah. main new technical

1:06:45

thing we did was a collaboration with

1:06:47

yet another guest on this show, the

1:06:49

folks from Tetsaio, to essentially do a

1:06:51

co-snark experiment. It's actually kind of a

1:06:54

iteration of the ZK-Rapped kind of in

1:06:56

a way where essentially we wanted people

1:06:58

to be able to like prove to

1:07:01

us that they'd hit some number of

1:07:03

taps or connections at the event, again

1:07:05

using NFC cards and this sort of

1:07:08

thing to produce signatures. And we had

1:07:10

a little prize store again. It was

1:07:12

like NFC rings, we had like NFC

1:07:14

bucket hats, we had bracelets, we had

1:07:17

the whole like the whole lineup. NFC

1:07:19

bucket hats, I've never heard that. NFC

1:07:21

bucket hats, yeah, yeah, I think I

1:07:24

have one somewhere. I got one of

1:07:26

the last ones. It's just an NFC,

1:07:28

like a cloth, resistant NFC tag behind

1:07:31

a label. It's actually really nice. And

1:07:33

also setting up that clothing line was

1:07:35

super fun. That was a collaboration with

1:07:38

myself and... Tesla, who is a contractor

1:07:40

on our team. Anyway, we basically want

1:07:42

people to make proofs for us that

1:07:44

they hit some number of connections, and

1:07:47

we want to experiment with delegating that.

1:07:49

Again, actually in the same vein as

1:07:51

like multi-party FHE, where you would love

1:07:54

for, you know, less of the computation

1:07:56

to happen on your own device and

1:07:58

for more of it to be kind

1:08:01

of done somewhere else. And so, yeah,

1:08:03

what people would do is they would

1:08:05

take the signatures they collected, they would

1:08:07

seek or share it into three. three

1:08:10

parts and we did a fun thing

1:08:12

where three different teams hosted the co-snark

1:08:14

servers. PSC hosted one of them, the

1:08:17

PSC Infor team, shout out to them,

1:08:19

cursive hosted one of them. I am

1:08:21

terrible at AWS. I like messed up

1:08:24

the deployment once or twice so I

1:08:26

was the weakest link in the chain,

1:08:28

but we hosted one and then to

1:08:30

say I hosted one. And yeah, like

1:08:33

together, you know, there's some sort of

1:08:35

like... you know theoretically we could all

1:08:37

collude and see your signatures but there's

1:08:40

kind of a reputation level thing which

1:08:42

i think is really interesting of like

1:08:44

we're all privacy focused teams the chance

1:08:47

that we do this is low or

1:08:49

at least it would kind of hurt

1:08:51

our reputations if we did and so

1:08:54

yeah we basically did this whole delegated

1:08:56

proving experiment and it actually like was

1:08:58

like oversubscribed like i think we had

1:09:00

a few tens of thousands of proofs

1:09:03

overall and there's one point where I

1:09:05

think almost almost 20,000 yeah across the

1:09:07

whole event and yeah basically again like

1:09:10

this is my fault where we were

1:09:12

supposed to like replicate the nodes so

1:09:14

that there could be more servers making

1:09:17

proofs And then not only did I

1:09:19

not do this, but also like two

1:09:21

of our nodes broke. So there's only

1:09:23

one node on the cursive side and

1:09:26

then eight for PSC and Tatsayo. And

1:09:28

so we're completely bottlenecked by cursive and

1:09:30

then eventually we fixed it and it

1:09:33

was fine. It cleared out the cue

1:09:35

of proofs that was waiting. But yeah,

1:09:37

one of the coolest things about deploying

1:09:40

things in the real world is that

1:09:42

you get very direct feedback loops. And

1:09:44

the way we had set it up

1:09:46

was that you would make these proofs

1:09:49

to show how many people you've met

1:09:51

and you could get some of this

1:09:53

NFC merch by showing these proofs. And

1:09:56

so we realized that the infrastructure for

1:09:58

generating Coast Narks was not working when

1:10:00

there were tens, if not a hundred,

1:10:03

of angry people standing in front of

1:10:05

our booth asking us. why their number

1:10:07

wasn't going up. Oh no. That was

1:10:09

a fun day. I love that you

1:10:12

you put yourselves in this sort of

1:10:14

line of fire. But it's just so

1:10:16

it's I mean it's also you get

1:10:19

like the visceral feeling I feel like

1:10:21

after that you go back building and

1:10:23

you really you take it in you're

1:10:26

not ignoring that feedback. I feel like

1:10:28

it sticks with you. No I think

1:10:30

we're really motivated by that as a

1:10:33

team like collecting it and also acting

1:10:35

on it. I mean, we do a

1:10:37

lot of just like ivory tower stuff

1:10:39

as well. It's hard not to when

1:10:42

you're trying to figure out what to

1:10:44

do with this technology. But we try

1:10:46

to like be on the ground as

1:10:49

much as we can. Yeah. And we

1:10:51

really like every opportunity we get, we

1:10:53

try to like make the most of

1:10:56

it reflect on it heavily and just

1:10:58

like move forward. I mean, that's a

1:11:00

huge number of events just there that

1:11:02

we covered testified to what you're saying

1:11:05

now. Yeah, again, huge shout out to

1:11:07

Tato folks. Ash, Ash, Lucas. Lucas. The

1:11:09

other Lucas Daniel Roman a bunch of

1:11:12

bunch of amazing folks helped both with

1:11:14

the activation But also they just came

1:11:16

to the booth and like helped volunteer

1:11:19

and stuff. So such lovely folks. Yeah

1:11:21

I Guess that ties into the the

1:11:23

final part of our DevCon experience which

1:11:25

was the cryptographic connections museum that we

1:11:28

launched in collaboration with a bunch of

1:11:30

artists and collaborators And the idea behind

1:11:32

the museum was just to showcase all

1:11:35

the work that has gone into cryptography,

1:11:37

help educate a lot of attendees who

1:11:39

might not know much about cryptography and

1:11:42

its history, and also tell people what's

1:11:44

going on in the future and what

1:11:46

to get excited about. And we basically

1:11:48

got a bunch of artists and cryptographers

1:11:51

and everything in between to showcase different

1:11:53

pieces of work. And I think... the

1:11:55

most exciting part of the whole experience

1:11:58

for me was that But Ash had

1:12:00

set up a display about the history

1:12:02

of MPC. So she had all of

1:12:04

these different fundamental

1:12:06

papers. I remember like the

1:12:08

day before I had gone

1:12:10

to a print shop in

1:12:13

Thailand and gotten like literally

1:12:15

like five, six hundred pages

1:12:17

of papers printed of just

1:12:19

math. And they were so

1:12:21

confused. Like, what is going

1:12:23

on here? And the whole

1:12:25

display was just these papers

1:12:27

and this like history of

1:12:29

MPC and a bunch of

1:12:31

like golden frames Set up

1:12:33

like an academics desk like

1:12:35

a coffee mug pen paper

1:12:37

all that stuff And

1:12:39

throughout the event like

1:12:41

this was Super attractive

1:12:44

to a lot of the museum

1:12:46

I guess and two people and they

1:12:48

both had their own papers, Grott

1:12:51

16 and Beaver Triples, on the

1:12:53

table. And so that was so

1:12:55

exciting seeing them walk over and

1:12:57

get to pick up their own

1:12:59

paper and be like, oh my

1:13:01

God, this is so cool. That's

1:13:03

definitely the highlight of DevCon for

1:13:05

me. Amazing. Nico actually was

1:13:07

also a featured artist. Do

1:13:09

you want to talk about the

1:13:11

thing that you did together? Oh, yeah.

1:13:14

You guys printed out the ZK

1:13:16

jargon decoder. It was supposed to just

1:13:18

be a museum piece and a

1:13:20

prize, but we actually had like

1:13:22

three or four robberies. That was

1:13:24

actually like... Oh, they all got

1:13:26

stolen. Okay, wonderful. Yeah. Pretty much

1:13:28

all of them. I think we

1:13:31

saved one for you, right? I

1:13:33

hope you have yours. Yes. Okay,

1:13:35

great. It's right here. Oh, cool.

1:13:37

Wonderful. Wonderful. The rest got stolen.

1:13:39

So, uh... If you're out there

1:13:41

listening to this, we will find

1:13:43

you. You weren't supposed to take

1:13:45

them. Wild. Okay, so that wrapped

1:13:47

up, I guess, 2024 for you.

1:13:49

When I spoke to you, end

1:13:51

of the year, you did sound.

1:13:53

You were like, I don't want

1:13:56

to travel anymore. Although I think

1:13:58

actually Andrew, you kept traveling. But

1:14:00

we're now, 2025, beginning of the year,

1:14:02

what's the plan? Are we going to

1:14:04

see you doing more activations at events?

1:14:07

Are you actually planning on doing something

1:14:09

else? You were talking about maybe a

1:14:11

shift more towards product. So yeah, tell

1:14:14

us what you're thinking. Yeah, I think

1:14:16

like, specifically technical directions are still being

1:14:18

figured out, but like, definitely, I think

1:14:20

just in terms of like process of

1:14:23

building, I think trying to shift more

1:14:25

to marathon mode, I think last year

1:14:27

was definitely a lot of sprints. of

1:14:29

like something comes up it aligns enough

1:14:32

with our directions and I think we

1:14:34

were just really in deep need of

1:14:36

like data like we just needed to

1:14:39

like try things with people who weren't

1:14:41

already billed sometimes like at a conference

1:14:43

like CK Summit it's actually good to

1:14:45

try it with people who are billed

1:14:48

just to educate more than like necessarily

1:14:50

get user data but something like DevCon

1:14:52

leans much more in the other direction.

1:14:55

And I think this year, probably still

1:14:57

we'll do some events and conferences just

1:14:59

because they're convenient and we have good

1:15:01

connections with them, but I think building

1:15:04

products for specific personas and just trying

1:15:06

to like grow within the cities that

1:15:08

we live in and like just try

1:15:10

to get like further and further away

1:15:13

from like ZK, Ethereum land. Because I

1:15:15

think the sort of confounding factor in

1:15:17

our space, which is both a blessing

1:15:20

and a curse, is that like... People

1:15:22

are down to try technology for technology

1:15:24

sake. Like they're like, oh, cool thing,

1:15:26

in the tech, let me try it.

1:15:29

Versus like, this does something good for

1:15:31

me. And most people operate entirely in

1:15:33

the latter camp of like, oh my

1:15:36

God, I don't want to have another

1:15:38

app on my phone. I just only

1:15:40

wanted if it's going to like add

1:15:42

to my life. So it's a little

1:15:45

bit like high level, but just I

1:15:47

think trying to get more feedback loops

1:15:49

from people, very outside of the space,

1:15:52

very hard. One example that comes to

1:15:54

mind is like professional apps entirely. So

1:15:56

at people, you know, who are like

1:15:58

LinkedIn power users or something, like how

1:16:01

can we make their kind of hiring

1:16:03

job search experience even better and just

1:16:05

trying to build a very like focused

1:16:07

application around that is like one example.

1:16:10

I don't know. and of doing that,

1:16:12

but. Cool. Are you looking for the

1:16:14

product, like the singular product that you

1:16:17

do end up building? Is that sort

1:16:19

of what a lot of this was

1:16:21

headed towards? Or do you actually see

1:16:23

yourself having a suite of different, just

1:16:26

more built-out products? We'll see. I'm excited

1:16:28

to check back in like six months

1:16:30

and let you know, but because we

1:16:33

are in more of a like consumer

1:16:35

and social area, like I think what

1:16:37

tends to be the right process in

1:16:39

this space is. like lots of shots

1:16:42

on goal and you see what sticks

1:16:44

and then go deeper with that. So

1:16:46

I have a sense that that's going

1:16:49

to be at least like the short-term

1:16:51

strategy is just building out lots of

1:16:53

stuff giving it an earnest experiment and

1:16:55

then trying to have a conclusive yes

1:16:58

or no and just continue and understanding

1:17:00

that most of them will be no's

1:17:02

and that is just kind of how

1:17:04

it is. Percursive actually I really liked

1:17:07

the approach like from the start and

1:17:09

Yeah, I want to see what next

1:17:11

year brings. Yeah, totally and on the

1:17:14

flip side We're also always looking to

1:17:16

collaborate with other teams on NBC FHE

1:17:18

research Any sort of technical research that

1:17:20

might align with these directions. We've had

1:17:23

many successful collaborations in the past with

1:17:25

different teams work in the space. So

1:17:27

if you are doing any technical explorations

1:17:30

and you want to maybe see it

1:17:32

integrated into applications, user-facing products, please do

1:17:34

reach out to us. Are you worried

1:17:36

at all about like some of the

1:17:39

patent stuff in the FAG world? Like

1:17:41

does that stress you out because you're

1:17:43

touching FAG a lot? And I'm just

1:17:46

wondering like, are any of the libraries

1:17:48

you're actually using potentially at risk because

1:17:50

of the more extreme patenting and less

1:17:52

open source nature of at least the

1:17:55

way FHA seems to be developing? I

1:17:57

think for the time being actually like.

1:17:59

Multi-party FHE is no longer a key

1:18:01

part of our toolkit. Just like right

1:18:04

now, there's a little bit of, there's

1:18:06

a few bottlenecks, like in particular, like

1:18:08

public key sizes are very large, it's

1:18:10

a little bit slow, it's like a

1:18:12

little bit hard to use, and I

1:18:14

don't foresee that changing in the near

1:18:16

future, or at least we have a lot

1:18:18

of other options that we have to

1:18:20

explore more fully before we come back

1:18:22

to it. So it isn't directly affecting

1:18:24

us. we were talking earlier about how

1:18:26

lovely it is that we're a big

1:18:28

family and this feels like a certain

1:18:30

like you know warring faction just like

1:18:32

decided to leave and make things harder.

1:18:34

And I hope they turn that around

1:18:36

like you know as Andrew said there's

1:18:38

an understanding that we're not fighting each

1:18:40

other we're fighting like the rapidly encroaching

1:18:42

like centralized zero privacy based world and

1:18:44

that's who we need to be focused

1:18:46

against and not against each other so it's

1:18:49

easy to say that in theory but I

1:18:51

understand if you're trying to run a business

1:18:53

it can be easier to be like zero

1:18:55

sum, I don't know. Yeah, I do think

1:18:57

this is a question of

1:18:59

long-term trajectory. If you want developers

1:19:02

to come to your ecosystem,

1:19:04

if you want people to

1:19:06

build upon the crypto systems

1:19:08

you are building, I think

1:19:10

open source is pretty imperative to

1:19:12

that. Now I don't claim to

1:19:14

know the specifics about the licenses

1:19:16

and the tradeoffs there and what

1:19:19

it means for business, but I

1:19:21

do think Generally speaking, the

1:19:23

more open, the better, at least

1:19:26

in this early stage of development.

1:19:28

Because the question is not which

1:19:30

FHE system is going to win.

1:19:33

It's whether FHE even stands a

1:19:35

chance in the first place. Yeah,

1:19:37

good call. Well, thank you both

1:19:39

for coming on the show,

1:19:42

sharing with us the journey

1:19:44

of cursive, all of the

1:19:46

activations and experiments, and the

1:19:48

direction you see yourselves going.

1:19:50

Thanks for being here. the two spotlights at

1:19:52

CK 11 and 12, like that's been so

1:19:54

great for us. And also just, you know,

1:19:57

to I think build awareness and education for

1:19:59

this type. has has been inspiring for

1:20:01

us us. It's really to see what

1:20:03

cryptographers think of this. think

1:20:05

of Thanks, Cool. for joining this one. for

1:20:07

Thanks for having me on. one. I

1:20:10

want to say a big thank

1:20:12

you to the podcast team, to say a big

1:20:14

thank you to the And to our listeners, thanks

1:20:16

for listening. and Tanya, and

1:20:19

to our listeners. Thanks for listening.