0:01

Welcome to Zero Knowledge,

0:04

I'm your host Anna

0:07

Rose. In this podcast

0:09

we will be exploring

0:12

the latest in Zero

0:14

Knowledge Research and

0:17

the decentralized web, as

0:19

well as new paradigms

0:22

that promise to change

0:25

the way we interact

0:27

and transact online. founder of Rare

0:30

Skills. We talk about onboarding developers

0:32

into ZK. The approach they have

0:34

taken with their ZK book and

0:36

boot camp on ZK. How to

0:38

strip down a complicated topic into

0:40

its parts, what to start with,

0:42

what can be introduced later, the

0:45

choices of what to teach and

0:47

which tools to use, as well

0:49

as how to best teach people

0:51

effectively to teach them enough so

0:53

that they can teach themselves going forward.

0:55

Now before we kick off, I do want

0:58

to just let you know about the ZK

1:00

Summit. ZK Summit 13 is coming up. It's

1:02

happening on May 12th in Toronto. This is

1:04

the second time we bring the ZK

1:07

Summit to North America, the first being

1:09

all the way back in 2019 when

1:11

we did one in San Francisco. If

1:13

you've never been to a ZK Summit,

1:15

I definitely recommend checking it out. This

1:18

is the spot to find out about

1:20

the latest research, the newest applications, to

1:22

find out who are the most important

1:24

players in ZK today. It's also a

1:26

wonderful way to get to know through

1:29

the larger ZK community. The application

1:31

to attend is now open. If you'd like

1:33

to apply to speak, you can do so

1:35

in the same form. The speaker application

1:38

deadline is March 15th. Generally,

1:40

I recommend applying early. We

1:42

do expect this event to sell

1:44

out, and tickets are limited. Speaker

1:46

slots are even more limited. So do

1:48

get your application in early, and we'll

1:50

get in touch with you very shortly. Links

1:53

as always are in the show notes and

1:55

I hope to see you there. Now Tanya will

1:57

share a little bit about this week's

1:59

sponsor. You might already know

2:01

of the ZK Jobs Board, a

2:04

place where teams can share open

2:06

roles with our community. But when

2:08

it comes to key hires, finding

2:10

the right fit in our niche

2:12

space can be difficult. That's where

2:14

MissingLink comes in. They're a talent

2:16

team built for the Web3 era,

2:19

helping projects across the ecosystem connect

2:21

with the right candidates at the

2:23

right time. They've worked with names

2:25

you'll recognize, Ethereum Foundation, Matter Labs,

2:27

Lido, Mina, Web3 Foundation, and many

2:29

more. Filling critical roles that drive

2:31

these teams forward. Whether you're an

2:34

established project looking to fill a

2:36

senior leadership role, or a startup

2:38

searching for specialized talent to refine

2:40

your product market fit, MissingLink can

2:42

help. They've done it for some

2:44

of the biggest teams in the

2:47

space, helping them and now you

2:49

find the right people fast. For

2:51

more details, check out the show

2:53

notes and visit their website at

2:55

Missing -Link .io. So thanks again, MissingLink.

2:57

And now, here's our episode. Today,

3:03

I'm here with Jeffrey Scholes, founder of

3:05

Rare Skills. Welcome to the show. Thanks

3:07

for having me. Today, we are going

3:09

to be diving into ZK education, something

3:11

that is near and dear to my

3:13

own heart. As you may know, there's

3:15

a project called ZK Hack. It's a

3:17

sister project to this show, to the

3:19

Zero Knowledge podcast. And over at ZK

3:21

Hack, we have the ZK Whiteboard sessions.

3:23

We have study groups. And we do

3:25

sort of a form of informal education.

3:27

But there is no curriculum. It's not

3:29

really designed with end goals and site.

3:31

But this has made me a little

3:33

bit curious about generally ZK education. And

3:35

I think this is something we're going

3:37

to be talking about in this episode,

3:39

Rare Skills. Jeffrey, what you've been working

3:41

on is very much that. So I

3:43

really want to explore how you think

3:45

about ZK education. Now,

3:48

I hadn't been really aware of your group

3:50

until 2023, I

3:52

think, when there was

3:54

the ZK book that came

3:56

out. I am curious,

3:58

like, what were you doing

4:00

before that? what led you to even

4:02

start working on ZK education, were you educating before

4:04

that? So why do we start a little bit

4:06

with a backstory of rare skills? Where did you

4:09

guys get your start? So rare skills didn't start

4:11

off with ZK. We started off with solidity

4:13

because I had been already publishing some material

4:15

on it under my own personal name on

4:17

medium and I could created a eudomy course

4:19

on it, which I'll talk about later. But

4:21

when you're in Web 3, you know, ZK

4:23

comes up a lot, and there was a

4:25

lot of, I was personally curious about it,

4:28

there were a lot of people who wanted

4:30

to learn about it. In particular, I had

4:32

one guy, I had been talking about, hey,

4:34

we should have a ZK boot camp, we

4:36

should break this down for people. I should

4:38

share what I've learned on the topic. Luckily,

4:40

there was one very pushy person who

4:42

insisted on it, and we created a

4:44

ZK boot camp, which initially started off

4:47

with just three and that the lecture

4:49

notes became the ZK book. And it

4:51

was quite popular. So that's what led

4:53

to us continuing to develop the ZK

4:55

book and keep running ZK boot camps

4:57

and just keep investing in education in

4:59

that in general. Cool. Before the ZK book,

5:02

though, were you doing something else in

5:04

terms of like education? Was there like

5:06

another field? With rare skills, we were

5:08

only doing solidity related things before that.

5:10

So there wasn't as much material on

5:12

that back then. So that was kind

5:14

of the niche we were feeling we

5:16

were feeling in. ZK is a, besides people

5:18

wanting it, it's a very interesting

5:20

subject. And it definitely had and

5:22

still kind of has a very

5:24

strong mystique around it, so that

5:27

made it even more interesting.

5:29

Well, before we dive into just generally

5:31

this course and rare skills, we should

5:33

also learn a little bit about you,

5:35

like where are you coming from to

5:37

be even kind of teaching this stuff

5:39

in the first place. I got extremely

5:42

lucky, I would say, because my... In

5:44

graduate school I studied computational theory, which

5:46

is a lot closer to ZK than

5:48

you would think. So I mean I

5:51

did study cryptography and things like that.

5:53

But ZK is largely about proving that

5:55

you carried out a computation correctly, right?

5:57

But what does a computation even mean?

6:00

Right and what's what's the language of a computation?

6:02

So I think that's something that a lot of

6:04

people get confused about with decay because they look

6:06

at circum and think like, I'm trying to prove

6:08

I did addition and multiplication. Well, most real computations

6:11

are not addition and multiplication. So that's not actually

6:13

how circum is being used. You're building a set

6:15

of constraints that when they're satisfied model, model the

6:17

algorithm. So dealing with stuff around circuits and constraints

6:20

is already stuff I had. dove pretty deeply into

6:22

from an academic perspective. So a lot of the

6:24

stuff that goes into ZK, I had a lot

6:26

of the prerequisites for already. There was just a

6:29

huge amount of luck with that. I don't know

6:31

how else to say that. So that was, I

6:33

graduated in 2015. And I was doing blockchain on

6:35

and off since then. I was aware of, I

6:38

think Z-cash definitely caught my attention back in the

6:40

day, but I did not understand how it works.

6:42

So I worked for Yahoo full-time as a engineering

6:44

manager. That's actually where I would say a good

6:47

chunk of education came in just because I really

6:49

believed in upskilling my team. It's much easier to

6:51

upskill good people than it is to. It's risky

6:53

to hire people, right? You don't know if they

6:55

have a good motivation and you don't know if

6:58

they're talented or talented or not. If they're talented

7:00

or not. Engineers like it if they're talented or

7:02

not. Engineers like it if they're talented or not.

7:04

Engineers like it if they're talented or not. Engineers

7:07

like it if you up-engineers if you up-a if

7:09

you up-a if you up-a if you up-a if

7:11

you up-up-up-kate if you up-kate if you up-kate if

7:13

you up-kate if you up-k That also helps with

7:16

retention because it's hyper competitive to retain a team

7:18

and at the more high levels of tech. So

7:20

I wouldn't say that like launched an educational career

7:22

there, but I moved to, so I was living

7:25

in America at the time, I moved to Asia

7:27

in 2021 and couldn't, I was working for Yahoo

7:29

and I couldn't keep doing that as a manager

7:31

because your job is basically to sit in meetings.

7:34

And so I got started in blockchain, worked on

7:36

a project that didn't end up launching, but this

7:38

was in 2021 when the gas prices on Ethereum

7:40

were really horrible. Yeah. I spent a lot of

7:42

time trying to figure out gas optimization, which there

7:45

was no material on it back then. So I

7:47

wrote a lot of medium articles on it and

7:49

actually created a eutomy course which became a bestseller

7:51

within like two weeks. Wow. And that was kind

7:54

of a, I definitely recognized that it's not natural

7:56

to be that successful that quickly. I hope that

7:58

was a strong signal that education was probably and

8:00

maybe a natural strength that I should lean

8:03

into. Interesting. At the time 2021,

8:05

2022, in the Ethereum land, what

8:07

was there for education? What could

8:09

people already do? Like there was

8:11

the cryptography 101 from Dan Bonnet,

8:13

but that's not Ethereum focus. There

8:15

was the East Global events where

8:17

you could build things, and you

8:19

would have had workshops and stuff

8:21

like that. But yeah, what did

8:23

exist? Most of my learning

8:25

was either reading, was reading

8:27

the yellow paper and then

8:29

writing code that would kind of

8:32

test if I understood it

8:34

correctly. Yeah, there really wasn't

8:36

much education, at least not

8:39

beyond like, okay, here's how you,

8:41

you know, build a New York

8:43

20 contract or something like that.

8:46

Were there no solidity boot camps

8:48

and stuff like that? But the

8:50

introductory stuff, it wasn't. It's not

8:52

that hard to learn, right? But

8:54

if you're really trying to, at

8:56

least back then trying to get

8:58

into the intricacies of how the

9:01

EVM prices things, then there was

9:03

definitely no material for that. Got it.

9:05

Okay, so that's where you started to realize

9:07

this might be a path. Yeah. When did

9:09

you found Rare Skills? September 2022.

9:12

And what was the first thing you did

9:14

with it? Because of the materials I

9:16

had been publishing already. I kind of

9:18

already had a personal brand in an

9:20

audience. So it wasn't that hard to

9:22

get... classes going. This whole like

9:24

post post post like a brain teaser

9:26

that's solidity based or a CTF on

9:28

Twitter. I like to think that we

9:31

innovated that because I didn't see anyone

9:33

doing that before. And that because nobody

9:35

else was doing that that brought a

9:37

lot of attention to us and some

9:39

we didn't call them KOLs back then

9:41

but like engineering KOLs were thought leaders.

9:43

Yeah, thought leaders. We're engaging with the

9:46

problems and of course that boosted our

9:48

reach significantly. Okay. Because they want to

9:50

show hey look by audience I can

9:52

solve this problem that you may have

9:54

seen already and then that massively increased their

9:56

reach. So I guess that was a Twitter

9:59

marketing strategy that ended up working really

10:01

well. This is 2022. And it was

10:03

a solidity-based CTF kind of thing? Yeah,

10:05

all of them were. Okay. So of

10:07

course, you know, when you're running a

10:09

company, you always publish blogs because of

10:11

SEO and whatever. But that actually kind

10:14

of turned into the core. feature of

10:16

the company over time. In fact, I

10:18

think of us more of as a

10:20

content company than like a training company,

10:22

especially that we're now starting to do

10:24

partnerships with these other platforms and creating

10:26

content for them. So that's starting to

10:29

take up a big part of our

10:31

business relative to training. And we really

10:33

put a lot of thought into, okay,

10:35

are we writing something that's really providing

10:37

value for people because it's... Easy like

10:39

okay, I'm gonna look at what people

10:41

are googling for I'm gonna you know

10:44

match the headers and things like that

10:46

in order to target it Well initially

10:48

when we were writing about stuff that

10:50

obviously I started off that way And

10:52

we were writing about stuff that wasn't

10:54

covered and people really appreciate that because

10:56

people were googling for things that there

10:59

were announcers for So I think that's

11:01

one thing that kind of kicked us

11:03

off as a content company but over

11:05

time I would say we really put

11:07

a lot of effort into making sure

11:09

that the writing was and what goes

11:12

into making good writing, is ultimately people

11:14

get value from writing because you save

11:16

them time. That's. really fundamentally what it

11:18

boils down to because you don't need

11:20

education just read the source code yeah

11:22

yeah but if it's hard to read

11:24

that or if it's if it's because

11:27

like you look at ZK research or

11:29

whatever a lot of early ZK research

11:31

was all these papers academic papers which

11:33

didn't really handhold very well and they

11:35

didn't define things and so it's relying

11:37

on a lot of like assumed understanding

11:39

and someone trying to learn ZK from

11:42

scratch they were lacking those early works

11:44

almost to lead them to this, or

11:46

those defining just definitions. I mean, more

11:48

recently, Nico, who's sometimes co-host of this

11:50

show, created the ZK jargon decoder. And

11:52

this was just such a game changer

11:54

for people that like, it's just definitions

11:57

of words that are used all the

11:59

time. But if you are trying to

12:01

learn ZK from a, like a research paper,

12:03

that is not going to be defined for

12:05

you. It's going to be assumed that you

12:08

know it. Exactly. And I'm guessing

12:10

in, like, generally in blockchain, solidity,

12:12

a lot of this stuff is

12:14

happening in the same way. Yep. You

12:16

have to be very conscientious of, like, you

12:18

know, in games, there's like this skill tree

12:20

where you learn a skill and it unlocks

12:22

other skills, blah, blah, blah, blah, right? I

12:25

would say for good education. Like for example,

12:27

let's say you're talking about how to big

12:29

bread or something like that, people are not

12:31

really concerned about like, what's the air pressure

12:33

like in the local temperature and is the

12:35

bread rising enough? Did I even mix the

12:38

ratio correctly or something like that? So it's

12:40

really easy when you're deep in a subject

12:42

to kind of, you start to have an

12:44

intuitive knowledge about what the fundamentals are.

12:47

Yes. But the person you're talking to

12:49

doesn't share that context. Is it you really

12:51

have to be opinionated about what you

12:53

think the reader already knows? The other thing that I

12:55

see a lot of ZK papers do wrong is when

12:57

you're learning something you knew you need to learn it,

13:00

see it from a few angles, and you need

13:02

to be able to relate it to something that

13:04

you already know. For example, I see some ZK

13:06

education, we define a group as a set with

13:08

a binary operator with, you know, these four properties

13:10

identities, identities, inverse, blah, blah, blah, existence of an

13:13

inverse and so forth. That's the first time someone's

13:15

seeing that there's no way that they can make

13:17

any sense out of any sense out of sense

13:19

out of that. So the trouble is, once you

13:21

already know, like we say what a group is,

13:23

it's kind of hard to not know it. And

13:26

it becomes so obvious to you that you can't

13:28

explain it to another person. That's I

13:30

think one thing that can make writing

13:32

well a challenge is that you have

13:34

to translate your intuition into the English

13:37

language. I'm not like a brain scientist

13:39

or something like that, but I have

13:41

a strong suspicion that intuition is not

13:43

inherently verbal. So trying to make that

13:46

translation is... challenging and it takes time

13:48

and I don't think it can be

13:50

done in one shot. So it takes

13:52

a few iterations to receive. Okay, this

13:55

article actually reflects the knowledge that's encoded

13:57

in my brain and is adjacent to

13:59

what the... what I think the reader

14:01

already knows. I think it's

14:03

also, there's another. point maybe

14:05

here is that like the information that

14:08

you have it's also things are changing

14:10

really rapidly around you maybe with solidity

14:12

less so but with ZK it is

14:14

like since 2022 it's been changing quite

14:17

rapidly and so even if there is

14:19

an intuition on like how to do

14:21

this right or what the sort of

14:23

trajectory of research is you could be

14:26

surprised so you have to be sort

14:28

of nimble at the same time right

14:30

yes Actually, that's one thing that we

14:32

still only teach for our boot camp,

14:35

we only teach the Groth 16 algorithm.

14:37

Because what you describe is right, like

14:39

the stuff that's actually used is quite

14:41

complex and changing very fast. But

14:44

I definitely believe that what we

14:46

call advancements in any particular field

14:48

is really just a rearrangement of

14:50

the fundamentals. Because at the end

14:52

of the day, most ZK algorithms

14:55

are polynomials, elliptic curves if they're

14:57

snarks and some kind of commitment.

14:59

Ash based. Ash based. Yeah, if, right, maybe

15:01

it's hash, there could be hash

15:03

as if it's, um, stark. So

15:06

you're just kind of starting with

15:08

the same building blocks and shuffling

15:10

them around, right? So, okay, like

15:12

permutation argument is like, we can

15:14

use it for luckups really cool,

15:17

but like. If you boil it down

15:19

to its things, someone just did some

15:21

algebra to proof that a certain property

15:23

applies. And you are taking advantage of

15:25

the sports zipper lemma, which kind of

15:28

appears everywhere. So if we. just really

15:30

focus on what those atoms are that

15:32

are being used everywhere. Because the way

15:35

I think to teach a complicated subject

15:37

is to really break down what are

15:39

the requisite skills that are involved in

15:41

that and then bring it all together

15:44

at the end. So that's why our

15:46

book is structured. We spend an entire

15:48

chapter is talking about elliptic curves because

15:50

if you're seeing it for the first

15:52

time you need to spend the entire

15:55

article seeing it from a few angles

15:57

looking at a lot of examples at

15:59

the end. To understand like a polynomial

16:01

commitment scheme, you need to understand polynomials

16:03

and you need to understand elliptic curves

16:05

if you're doing something snark-based. So like

16:08

if someone's seeing the Swartz-Zippel-Lema for the

16:10

first time and elliptic curves point arithmetic

16:12

for the first time, and that's just

16:14

way too overwhelming unless this person has

16:16

a lot of background in math. So

16:18

I think it's really key to really

16:21

drill down the... What are the components

16:23

of the knowledge? If something is too

16:25

difficult, that means you need to further

16:27

break it into components. And I think

16:29

that can be done way more times than

16:32

a lot of educators realize, right? Interesting. Yeah.

16:34

So like with groups, OK, groups depend on

16:36

sets. binary operators just a subset of the

16:38

Cartesian product, right? So you can keep like

16:40

merges, you know, you can keep recursively breaking

16:43

things apart. And you should try to do

16:45

that. You should err on the side of

16:47

doing that more than not, I would say. Let's

16:49

talk about how the ZK book came

16:51

together. You released it in 2023, but

16:54

I'm assuming this took you a lot

16:56

longer than that. In 2022, so you've

16:58

already sort of given us

17:00

a bit of a hint about

17:03

the book. So you've chosen, you

17:05

chose at that time to focus

17:07

on Gros 16, which I think

17:09

was the right choice. Still, I

17:11

believe it's the most used proving

17:14

system. So it has quite a

17:16

few variations. There's different implementations. for

17:18

Gros 16. So engineers using one

17:20

may, as far as I understand,

17:22

like it's not a totally standardized

17:24

proving system, but it's the most

17:26

common. And I guess depending on

17:29

what level you're using, like you're

17:31

playing with it, it might be

17:33

a good starting point. So you

17:35

chose in 2022 to do Gros

17:37

16, but tell me like, where

17:39

do you start teaching something like Gros

17:41

16? And actually I am curious about

17:43

the CK. At what level are you

17:45

teaching it? Is it for an engineer

17:47

or is it for a cryptographer? Is

17:49

it like theoretical or is it like

17:51

hands-on? So in terms of picking Groth 16,

17:54

well one thing I learned from was tornado cash

17:56

was kind of a code base that I would

17:58

wrestle with until I really understand. And there

18:00

was a lot of reference implementation, so

18:02

that helps a lot. But the target

18:04

audience is a senior engineer who has

18:06

no experience with cryptography. So the North

18:08

Star is very clear. We want to

18:10

code Groth 16 from scratch and understand

18:12

what we're doing. In terms of how

18:14

to break it down, let's say we're

18:16

teaching people like how a hash function

18:18

works. Well, a hash function is really

18:20

just a bunch of x-hores, bit shifts,

18:22

maybe a permutation somewhere or something like

18:24

that. It's really just a matter of

18:26

what are all of the unfamiliar components

18:29

that people are dealing with in Brock

18:31

Groth 16 and then break that down,

18:33

you know, breaking it down to the

18:35

atomics. Okay, so there's pairings, there's elliptic

18:37

curves, there's QAPs and R1CS. Well, the

18:39

paper doesn't talk about rank one constraint

18:41

systems, but a lot of people do,

18:43

so we talk about that. Then there's

18:45

a trusted setup. Oh yeah. Then there's

18:47

the verification code, which if you look

18:49

at it and you know, tornado cache

18:51

or whatever CERCOM spits out, it's somewhat

18:53

straightforward, but you want to understand everything

18:55

that's going on to there, which is

18:57

largely calling the pre-compiles on Ethereum for

18:59

elliptic curve addition multiplication and pairing. So

19:01

those are things you really want to

19:03

understand, how do you interface with those

19:05

pre-compiles? And why are we even using

19:07

them in the first place? So originally...

19:09

Because the ZK book has actually gone

19:11

through a second edition. So in the

19:13

first edition, it was, okay, we start.

19:15

a rank-one constraint system, rank-one constraint system

19:17

becomes a QAP, and then the QAP,

19:19

you can be run on a trusted

19:21

setup, because the QAP is just one

19:23

big polynomial equation at the end. Can

19:25

you say again what QAP is? Oh,

19:27

sorry, quadratic arithmeticic program. Okay. Thank you.

19:29

I'm always used to just, see, this

19:31

is the thing I'm talking about. It's

19:34

in your head. I work with this,

19:36

and I actually did. I didn't recognize.

19:38

Is that accurate? Yeah, totally. And then

19:40

how to run that on an elliptic

19:42

curve? It evaluated to some elliptic curve

19:44

points. Now, a quadratic arithmetic program boils

19:46

down to three terms at the answer

19:48

you're left with three elliptic curve points.

19:50

If you pair the first two, you

19:52

should end up. with the third one

19:54

when you pair it with the generator

19:56

of the other group for that one.

19:58

OK. But if someone just gives you

20:00

three elliptic curve points and say, hey,

20:02

look, one pairs to the other, that's

20:04

not good enough. You could have just

20:06

given me any three points, whatever. So

20:08

we use that as a motivation for

20:10

why does Groth 16 have these other

20:12

terms like alpha, beta, gamma, and delta.

20:14

And now to understand elliptic curves, because

20:16

we actually taught elliptic curve cryptography at

20:18

least a little bit in our solidity

20:20

boot camp, and something I observed, this

20:22

is something I would tell everybody going

20:24

into it. Nobody ever learns elliptic curves

20:26

after a week. It never happens. And

20:28

I've actually have hard evidence to back

20:30

that up. So I was thinking, so

20:32

if someone's reading this from a book

20:34

perspective, they're not going to learn it

20:36

after one chapter. I had to really

20:38

think about what's the right way to

20:41

package elliptic curves so that they're understandable

20:43

because fundamentally they're actually not that complex.

20:45

It's just the simple algebra equation that

20:47

you can draw a line through and

20:49

then solve for intersex. But trying to

20:51

do algebra in your head like that

20:53

is a nightmare. And that's when I

20:55

realized presenting this as a group is

20:57

a lot simpler because even though groups

20:59

are, most engineers have not done group

21:01

theory before. But it lets you treat

21:03

these things kind of like a polymorphism

21:05

in object-oriented programming or like an interface

21:07

where you don't have to know what's

21:09

going on under the hood. You just

21:11

say, it's going to implement these functions

21:13

which have these properties and you don't

21:15

have to worry about what's going on

21:17

under the hood. But in order to

21:19

understand that interface you want to understand

21:21

what a group is. And then since

21:23

most people don't have an encountered groups

21:25

before, they need to relate it to

21:27

sets, which I think most engineers at

21:29

least have a vague idea of what

21:31

a set is. This is going from

21:33

sets to groups, and then from groups

21:35

to elliptic curves was kind of, here

21:37

is how I can keep breaking it

21:39

down until I'm close to what I

21:41

think the engineer knows already. And then

21:43

that happens over the course of four

21:46

or five chapters or so, which, if

21:48

you try to teach it in one

21:50

chapter, it's just not going to work.

21:52

And I've gone back and forth with

21:54

people where I have unlimited information bandwidth

21:56

and I can't do it. in one

21:58

setting. So there's trying to do it

22:00

in one chapter is just silly. So

22:02

that is how it evolved that way.

22:04

And I'm honestly, this is something I'm

22:06

really proud of because there's a lot

22:08

of CK tutorials that have come out

22:10

since we published that were, he sets

22:12

groups and like, no, we, we, uh...

22:14

You pioneered that. That was our idea

22:16

to break it down that way. And

22:18

then with pairings, pairings are very scary,

22:20

right? So the decision that I've made

22:22

there is we're not going to teach

22:24

how pairings work under the hood because

22:26

even Vitalica has a hard time understanding

22:28

them. But if we can treat them

22:30

like a black box, like engineers use

22:33

hash functions without knowing how they work

22:35

no problem. So how can I try

22:37

to present it in that manner? Not

22:40

knowing how a hash function works doesn't

22:42

mean you don't know its properties, you

22:44

don't know its properties, right? built up

22:46

a lot of intuition around groups. It's

22:49

a case not really a group, but

22:51

can we relate it to that in

22:53

a way where it doesn't feel like

22:56

such a huge leap to people? And

22:58

that is, I would say a

23:00

general teaching philosophy, sorry I'm

23:02

going on for a long

23:04

time, I'll wrap this up

23:07

here, is if something's hard

23:09

to explain, let's say you're

23:11

trying to teach topic why that

23:13

they might. be able to like rock

23:15

the concept. Yeah. If someone already understands

23:17

homomorphic encryption, for example, then elliptic

23:20

curves don't seem as strange because

23:22

they do have some of the

23:25

properties there. So if you're able

23:27

to relate to that, then people can

23:29

grasp it more quickly. So just really

23:31

try to avoid excessively massive leaps and

23:33

just bring people incrementally closer to what

23:35

they're looking for, which is why the

23:37

CK book is somewhat long. It's about...

23:39

When we first published it was about

23:41

50,000 or 60,000 words or something like

23:43

that. But you could describe graph 16

23:46

on one page. It's not at the

23:48

end of the day a complex algorithm.

23:50

The thing it just relies on things

23:52

that are unfamiliar to most people. Most

23:54

people don't know pairings in the

23:56

elliptic curves. And the notion of

23:58

a quadratic arithmetic program. Why are you

24:00

even using that? Even though it can

24:03

be expressed mathematically and a high schooler

24:05

could understand, but why? So explaining those

24:07

wise. So that it's not just, trust

24:09

me bro, write the code. That's not

24:11

what people are really, that's not what

24:13

people are really wanting, right? That whole

24:15

thought process is how the ZK book

24:18

evolved into its current form. The second

24:20

edition was, I realized there was a

24:22

lot of stuff around polynomials that was

24:24

not broken down enough. So I felt,

24:26

okay, short simple limits, so easy to

24:28

understand. I can just leave that

24:30

as like a mini section inside

24:33

of the discussion about quadiotic arithmeticic

24:35

programs. Not getting the quadratic arithmetic

24:37

program as quickly as I would

24:40

think they would. Okay, you evolved it.

24:42

Yeah, okay, we need to spend more

24:44

time thinking about how can we

24:46

represent vectors as polynomials because that's

24:48

basically what a quadratic arithmetic program

24:50

is doing. We can't just say,

24:52

oh, the transformation step is you

24:54

do Lagrange interpolation on the columns.

24:57

Let's just spend some time thinking

24:59

about what does it mean to

25:01

encode data as a polynomial even

25:03

in the first place? And then

25:05

what properties can you rely on?

25:07

Can you add them together and

25:09

then some property is retained? of

25:11

helping people get their hands dirty with

25:13

something before they moved on with it. I

25:15

mean, one thing that we do, for example,

25:18

when we're teaching the ZK boot camp is

25:20

when people are introduced to elliptic curves, I

25:22

have them implement the ECDSA algorithm, which is

25:24

very little to do with ZK. But why?

25:26

Because people need time to wrestle with elliptic

25:28

curves before they're comfortable with it, and if

25:30

you're stuck on ECDSA, there's a ton of

25:32

resources online that will help you, and nowadays,

25:34

chat-chatchat-GPT can probably one-GPT can probably one shot

25:36

it. So that now you're getting an environment

25:39

where you're able to get feedback because if

25:41

you jump right into using elliptic curves the

25:43

way Groth 16 does, you're not going

25:45

to find many resources and when you're

25:47

getting stuck, you're going to be very

25:49

frustrated. But if you've already built up

25:51

some fluency with elliptic curves, then you

25:53

go to the next stage of the

25:55

Groth 16 algorithm, it's not so bad. You're

25:57

not dealing with two things at the same time.

26:00

learned a lot from the actual boot camp

26:02

and then incorporated it into version two, this

26:04

sounds a lot like something that's been happening

26:06

since 2021 over on the ZKHAC Discord. And

26:09

to a listener, like, I'm going to mention

26:11

this a lot, ZKHAC, and this sort of

26:13

the educational portal that I would say, like,

26:15

we have, very different from rare skills. I

26:18

would say, way less structured. It's much more

26:20

kind of makeshift. We do study groups. We

26:22

do study groups. We kind of like we'll

26:25

have a little six-week. study group on

26:27

a paper and then wrap it up

26:29

and then do a different one. But

26:31

in that group there's Justin Thaler who's

26:33

been doing a study group and those

26:35

are long. He does like multiple months

26:38

study groups where he goes through his

26:40

book and he goes through it like

26:42

chapter by chapter with a group. Every

26:44

time he does it he's learning and

26:46

he's been on the show to talk

26:48

about it He's actually learned what's not

26:50

working in some explanation that he's put

26:52

in here or what's changed and so

26:54

he's updating it and this sounds somewhat

26:56

similar like that feedback to a book

26:58

I think in a space like this

27:00

is so important I think in the

27:02

past you'd have textbooks created and they'd sort

27:04

of sit there and they'd dominate a way

27:07

of thinking for a long time But it's

27:09

not a luxury that we have today like

27:11

things definitely need to evolve Yeah, no having that

27:13

feedback mechanism in there is nice because the

27:15

we launched the boot camp slightly before the

27:18

book about two months before So that I

27:20

mean the book the book I had been

27:22

working on for a long time because those

27:25

were my notes for the lecture initially

27:27

But yeah, it definitely informed that you

27:29

should not talk about topics in this

27:31

order or this explanation is just not

27:33

going to work for someone new to

27:36

the topic and I think Even something

27:38

I learned is I had to put a

27:40

lot of effort into learning growth 16, right?

27:42

And the temptation can be is like, I

27:44

want to share with you everything that I

27:46

learned. But in reality, the value is I

27:49

need to trim it off. Like for

27:51

example, a most salient example is, you

27:53

know, abstract algebra, we have groups, rings

27:55

and fields, right? So naturally, okay, I'll

27:57

talk about those things, but then I really...

28:00

after a few iterations on the boot camp, you

28:02

don't really need to know what a ring is

28:04

in order to really understand how

28:06

Roth 16 works. That's just giving people

28:08

another hurdle to go over, which is

28:10

not 100% necessary. Now should they learn

28:12

that eventually, of course, but... What we're

28:15

aiming for is what's the straightest line

28:17

to get from not knowing it to

28:19

being able to implement it and have

28:21

a really good understanding of what you're

28:23

doing. So there was a pretty significant

28:25

subtractive element that came in after having

28:27

learned that. And even when I'm editing

28:30

other people's research for our publications on

28:32

rare skills, it's just a very common

28:34

temptation. I put so much effort into

28:36

understanding this. I just want to tell you

28:38

about it, but it turns out you don't

28:40

actually need it. So which is frustrating. But

28:42

I think that's a, if you want to

28:45

produce good writing, it's just something that you

28:47

have to accept. I feel like there's a

28:49

parallel here to like gaming and how

28:51

really good games on board you well.

28:53

And like one of the things, and

28:55

it depends a little bit on the

28:57

type of learner, I know there's people

28:59

who really need to learn everything before

29:01

they do anything that exists. Like they

29:03

really want to detail, like they want

29:05

to know everything. I personally don't work

29:07

that way. I have a tendency to

29:09

like yellow into something and then figure

29:11

it out after. And games that are

29:13

built that way, where it's like you

29:15

can already very early on do something.

29:17

It's so much more fun. And like obviously

29:20

over time you can up your skills

29:22

and you can like learn a lot

29:24

more about this world or the whatever

29:26

the game infrastructure is. But at the

29:28

beginning being able to do something is

29:30

really key. And I think that's what

29:32

you're talking about here, which If you

29:34

wanted to build secure ZK systems in

29:36

the real world, you do need. But

29:38

if you are just trying to do

29:40

something, you don't. Yep. No, I think

29:42

that's very important. People need

29:44

feedback right away about like, they need

29:46

to see results to keep their motivation

29:48

up. I would say 98% of the

29:50

engineers that I've dealt with to match

29:53

the description of, I don't want to

29:55

learn everything before I get started. I

29:57

would say 98% of people will lose

29:59

motivation. if there's too much theory and

30:01

that before anything happens. So I do

30:03

try to in the ZK boot camp

30:06

and even in the book too. Hey,

30:08

here's a dummy ZK proof that we

30:10

can make as early as possible. Now,

30:12

it has very serious limitations. It's not

30:14

succinct. There's some edge cases where you

30:16

can leak information, but technically it's a

30:18

ZK proof. Oh, okay. That is so

30:20

important. And then when people, people are

30:23

more motivated to learn the theory after

30:25

they've. seen. Oh, okay, I can see.

30:27

I kind of already have a framework

30:29

for where to put things in my

30:31

brain. What tools are you using with

30:33

these engineers? So you have the book

30:35

itself, but like, is there a DSL

30:37

that you've chosen? You sort of mentioned

30:39

CERCOM. Like, is that what you're building

30:42

this stuff in? The ZK boot camp,

30:44

talking about the back end. So assuming

30:46

you already have CERCOM, how do you

30:48

how do you prove that? So that

30:50

compiles to a rank one constraint one

30:52

constraint system, right? Now how do you

30:54

create a proof for that? So this

30:56

is lower on the stack then. You're

30:59

not using CERCOM. You're like working behind

31:01

CERCOM. Would you use anything like Arkworks

31:03

or also not? Well, okay. So we're

31:05

basically implementing Arkworks in Python, essentially. That's

31:07

what the... Not obviously not a fully

31:09

production version, but the whole thing like,

31:11

okay, you're given a rank one constraint

31:13

system now you need to run the

31:15

polynomial interpretation and evaluate the polynomials and

31:18

elliptic curve points to the commitment so

31:20

forth and so on. That's the stuff

31:22

that we teach in the class and

31:24

that's all done in Python. Rust is

31:26

way more important to learn if you

31:28

want to work in ZK, but most

31:30

people are not that fluent in Rust,

31:32

so if people are fighting the Rust

31:35

compiler and learning a new concept at

31:37

the same time, that's just not going

31:39

to work. We get so much of

31:41

this with ZK Hack does the CTF

31:43

around bugs in ZK protocols and we've

31:45

heard especially in the last edition that

31:47

more people were annoyed by having to

31:49

use rust than anything else. Like the

31:51

winners even, they did well, but they

31:54

were like, rust is my problem. Yeah,

31:56

rust is definitely, it needs to be

31:58

studied on it. own, not in parallel

32:00

with something else. This is something I've

32:02

gotten feedback on a lot, because when

32:04

people try to learn CK oftentimes, they,

32:06

oh, I'm going to try to learn

32:08

how to program in SRCOM, which is

32:11

a reasonable thing to do if you

32:13

were just coming in from the outside,

32:15

but you don't really understand that you're

32:17

writing constraints, you're not writing imperative code,

32:19

and people try to force their... imperative

32:21

code mindset onto circumn, it just ends

32:23

up being weird. And they don't understand.

32:25

Okay, why can I not have two

32:28

multiplications in one constraint? That seems like

32:30

the most random requirement ever. And, but

32:32

if you understand that, okay, well, behind

32:34

the scenes, there's a rank one constraint

32:36

system which only has one multiplication per

32:38

constraint, but why is that? Well, because

32:41

with a pairing, you can only do

32:43

one quote-unquote multiplication. So that just helps

32:45

things make a lot more sense. And

32:48

it initially seems random. Why is everything

32:50

happening in a finite field? Or modular

32:52

arithmetic. That just seems like we can

32:55

overflow and have a security bug. Why

32:57

are we doing that? Because elliptic curves

32:59

use finite fields. That's why. Or they

33:02

form an additive group, rather, I should

33:04

say. And I think we're kind of

33:06

in the time right now where if

33:08

you wanted to use solidity in like

33:10

2016, you... kind of had to know

33:12

how the EVM works. The abstraction was

33:14

extremely leaky. Right, and it's better than

33:17

programming in assembly, but not a whole

33:19

lot is hidden from you and the

33:21

abstractions are not very good. I would

33:23

say right now with, it's definitely made

33:25

a huge amount of improvements recently, but

33:27

if you really want to understand ZK

33:29

and write with ZK and write with

33:31

ZK DS cells, then you need to

33:33

know what's going on behind the scenes.

33:35

Interesting. So you would kind of compare

33:37

the ZK DSels today to Solidity 2016,

33:39

I mean, there's lots of people who

33:41

don't actually pay that much attention to

33:44

how the EVM to how the EVM like

33:46

solidity plus all of the

33:48

tooling that's been built up

33:50

over, what is it, nine years? Nine

33:52

years? Wow, yeah. But then, yeah, the

33:54

ZK, I mean, ZK, DSLs generally as

33:57

a topic really only became a thing

33:59

in 20. 2021, 2022. I do have

34:01

a question about that, though, because you're

34:03

saying sort of like you're rebuilding arc

34:05

works in Python to interface potentially with

34:07

CERCOM. There are these other DSLs that

34:09

also kind of live at different parts

34:11

in the stack. Like have you explored

34:13

for any of your boot camps, any

34:15

other languages? And I realize like you're

34:17

still going to stick to Python and

34:19

like recreating this thing, but yeah, like,

34:21

or do you interface with any other

34:24

languages? No, I mean, well, some people

34:26

use solidity just to work with the

34:28

pre-com piles, right? Because if you want

34:30

to, most people who are taking the

34:32

ZK boot camp come from a smart

34:34

contract background. So, and it's helpful to

34:36

reference real things. And my favorite thing

34:38

to reference is the tornado cash verifier,

34:40

with at least not recognizing how to

34:42

interface with the pre-com piles. So, we

34:44

will launch a CERCOM boot camp in

34:46

March, so coming up soon. So that's

34:48

where we teach the back end as

34:50

an abstraction. And how do you actually

34:52

implement the circuits as part of an

34:54

application? So how do you go from

34:56

only having addition and multiplication to verifying

34:58

a ZKVM or a traditional cryptographic hash?

35:00

That's the end goal. So that's how

35:02

I like to design the courses. What's

35:04

the outcome? The very tangible outcome and

35:07

what's the straightest path to get there.

35:09

So for the ZK boot camp that

35:11

we have, the tangible outcome is I

35:13

want to implement Groth 16 from scratch.

35:15

Now what's the straightest the straightest line

35:17

to get there. without skipping things and

35:19

making sure people have the prerequisites that

35:21

we talked about earlier. So for CERCom,

35:23

it's how can I create a ZKVM

35:25

from scratch and also a cryptographic hash

35:27

function? Not the CK-friendly ones. I mean,

35:29

like the ones that seem very mysterious

35:31

for CK proofs. And what background knowledge

35:33

needs to go into that in order

35:35

to actually code that up and understand

35:37

what you're doing. And then take the

35:39

most direct path there. So in that

35:41

case, I don't, being comprehensive is not

35:43

a goal of ours. It's very much

35:45

like, for someone who's new, how do

35:48

you keep them motivated and get them

35:50

the most direct pack there? And then

35:52

they can, there's a lot of ZK

35:54

education at the relatively high level. you're

35:56

good enough to read the papers, then

35:58

like, you don't need us anymore, right?

36:00

So we don't need to cover everything.

36:02

So, but we need to get you

36:04

to that point first. So, I mean,

36:06

I feel comfortable and a lot of

36:08

other people who graduate from our course

36:10

feel comfortable just picking up a paper

36:12

and reading it. So, that's really why

36:14

we want to get to people. We don't

36:16

want to at the end of the day be

36:18

a substitute for the papers. Right, because it's too

36:21

much work to just translate everything. But people get

36:23

to that point by, I know extremely well how

36:25

this particular algorithm works. Now it's easy for me

36:27

to learn adjacent things. Like if you already know

36:29

Python, it's easy to learn Ruby. But learning Python

36:31

for the first time is quite hard. So we

36:33

did that for CERCOM. People ask me like, man,

36:35

that's an outdated language. Why are you doing that?

36:37

Well, I was about to, sure. I was about

36:39

to ask you, why not nor? Right.

36:42

Because I think it provides the,

36:44

are you familiar with Huff by

36:46

any chance? I only know it

36:48

because Zach from Aztec talks about

36:51

it. Okay. But I don't, I

36:53

don't actually know what it

36:55

is. Sure. Basically, it's the

36:57

ZK EVM. So the op

36:59

codes in, in the EVM.

37:01

It's almost one, it's an

37:03

assembly language that's almost one

37:05

to one. So if you

37:07

wanted to write raw op codes

37:09

in solidity, you'd write 0X52,

37:11

you know, 0X60, blah, blah,

37:13

blah, blah. So they turn

37:15

those into the mnemonics,

37:18

so push M store, blah, blah,

37:20

blah. But then there's some nice

37:22

sugar coating on top of it,

37:25

like if you want to do

37:27

a jump, right, undisputably the most

37:29

effective way to teach the EVM

37:31

because people get their hands dirty with

37:34

it and the very roughest edges are

37:36

sharp are smoothed off like having to

37:38

memorize a jump destination or something like

37:40

that but you still deal with what's

37:42

going on with the stack and what's

37:44

going on in memory you still have

37:46

to tangle directly with that. So I

37:49

would say CERCOM has that advantage

37:51

where you're dealing really closely with

37:53

what the constraints are but The very

37:55

roughest edges are taken off where you're

37:57

not literally writing a rank-win-constraint system by

37:59

hand. So it takes off just

38:01

the right amount of roughness, but you're

38:04

still tangling with the core thing. Plus,

38:06

there's a huge ecosystem behind it. That's

38:08

true. And lots of library, like it's

38:10

been around the longest. Yeah. So you

38:13

can look at other established code bases

38:15

and figure out what's going on, which

38:17

is not the case for pretty much

38:19

every other DSL. Like I almost, I

38:22

was considering, I don't know how to

38:24

say it, Gunark from consensus? Gunark? Gunark?

38:26

Gunark? Yeah. I think that's a really

38:28

nicely designed. It's like Sircom, but better.

38:31

But better. I mean, sure, but better.

38:33

I'm better. Lanier, I don't know what

38:35

to reference for that. If

39:14

you pick a random CERCOM library from

39:16

get hub, I'm pretty sure it's going

39:18

to work with the same CERCOM compiler

39:20

you install from the CERCOM website. Okay.

39:22

So, I mean, there is an older

39:24

version of CERCOM, but I really actually

39:26

have a hard time finding those code

39:28

bases anymore. So most of it is

39:30

up to date, right, in terms of

39:32

it works. And there's a lot more

39:34

institutional knowledge around how to audit it.

39:36

Which is not saying much, there's not

39:38

many easy chaoticters, but still way better

39:40

than the other ones. Circum has some

39:42

things that I don't like and I

39:44

think you talked to anyone who's developed

39:47

with it They'll tell you they don't

39:49

like it But once you can get

39:51

over those things then it's you're really

39:53

dealing directly with constraints and so with

39:55

a Like how do you even design

39:57

a stateful computation with constraints? That's something

39:59

that really trips people up. Like I

40:01

can understand, I'm trying to prove I

40:03

did a multiplication correctly, right, you know,

40:05

that just maps directly the multiplication. But

40:07

how do I prove that I hashed

40:09

something correctly? Because that's using a

40:11

bunch of very nonlinear operations. Now we need

40:13

to again break it down. Well, how do

40:15

you even represent a staple computation in the

40:17

first place? Let's say a factorial with an

40:20

unknown number of steps in it. So, you

40:22

know, one times two times three times four,

40:24

blah, blah, blah, blah, blah, blah, blah, blah,

40:26

blah. But it's going to stop. Well, the

40:28

breakdown is you need to have an upper

40:30

bound on the number of times you would

40:32

compute the factorial. Let's say you're not

40:34

going to do anything more than 100 factorial.

40:37

So your constraints are I'm going to

40:39

keep multiplying and then you have one

40:41

factorial, two factorial, three factorial, blah,

40:43

blah. At each step. But now, I want to

40:45

know 89 factorial. Well, because it's an arithmetic, you

40:48

still have to go all the way up to

40:50

100, but then you need to pick something at

40:52

the index at the index 89. Well, how

40:54

do you quote unquote pick things when you

40:56

only have addition and multiplication? Well, you can

40:58

do an inner product with a binary mask.

41:00

So you just multiply everything by zero, one

41:02

where you want it, and zero in the

41:04

other places. So that's gonna zero out the

41:06

other things and you're only left with 89

41:09

factorial. And all of that I described as

41:11

done with only addition and multiplication. Okay, cool.

41:13

We have now we have a stateful computation,

41:15

but what if we're doing more than just

41:17

a single multiplication? Well, well, we can add

41:19

another dimension to it. Now we have like

41:21

registers, like registers, right. That's now the

41:24

state transition between each of those is

41:26

not a straightforward as a multiplication But

41:28

if we keep breaking it down then

41:30

you know we can represent it with

41:32

our very limited language And this

41:35

is something that circum gives you

41:37

yes Well, so the challenge is

41:39

circum only gives you addition multiplication

41:41

and the constraint right? But the

41:43

idea is there's design patterns which

41:45

are That exists like how do you select

41:47

from an array? How do you do a

41:49

stateful computation? How do you represent a

41:52

bit, let's say you're trying to

41:54

actually represent a bit shift. Well, you

41:56

have to do that outside the circuit and

41:58

then prove that the bit shift. was done

42:00

correctly because at the end of the

42:02

day, a bit shift unless you overflow

42:04

is multiplication by power of two, right?

42:06

So you can prove that the bit

42:08

shift was done correctly. You can't directly

42:10

compute it, but outside of circuit, you

42:12

compute the bit shift, put the answer

42:14

back and then constrain that it was

42:16

done correctly. So we're going to publish

42:18

a new part of the ZK book,

42:20

which talks about how to design these

42:22

things with a chapter for that is

42:24

called compute then constrain. Now the challenge

42:26

is I know this pattern exists, but

42:29

I, there's no word for it that's

42:31

existing out there. Interesting. So while, um,

42:33

designing this circum education thing, I had

42:35

to kind of invent a lot of,

42:37

um, terminology for things that other people

42:39

are, I didn't invent the idea. I've

42:41

definitely seen other people doing them and

42:43

doing it repeatedly enough to the point

42:45

where I think it's a design pattern.

42:47

Um, but the design pattern doesn't really

42:49

quite have a name. Yeah. I'm excited

42:51

about that because we're, I'm not aware

42:53

of any, um, resource like that out

42:55

there. You talked a little earlier about

42:57

like content and creating content. This is

42:59

a bootcamp. What, how do you actually

43:01

host a boot? Like what is it?

43:03

Is it like a group that meets

43:05

regularly? Do you do like six weeks?

43:07

Do you do eight weeks? Sure. What

43:09

is it? Um, so the bootcamps are

43:11

just video calls with five or 10

43:13

other people and myself, there's two instructors

43:16

at the company, myself and another person

43:18

named Chowau. He lives in Brazil. I

43:20

live in Asia. So that's great. We

43:22

have both time zones covered. We teach

43:24

the class, but I try to avoid

43:26

just giving people a video lecture because

43:28

that's, um, and I could just record

43:30

myself and then send that, right? I

43:32

really try to make it interactive for

43:34

the instructors asking people questions. I've found

43:36

that to be the level of effectiveness

43:38

learning by someone asking you questions that

43:40

push your knowledge boundary versus someone telling

43:42

you something is not even in the

43:44

same galaxy. It's way more effective to

43:46

where someone is discovering things on their

43:48

own, but they're not really discovering things

43:50

on their own because they're going on

43:52

a guided quest, right? So that's the

43:54

information sticks more. And you don't walk

43:56

out with a lecture like, oh, everything

43:58

made sense there, uh, but I can't

44:00

do anything in here, right? Everyone who's

44:02

gone to college has had that experience

44:05

where I made so much sense in

44:07

the lecture and I can't do the

44:09

whole work. I can't remember anything. Yes.

44:11

So you, yeah, when, when people are

44:13

paying for it, they don't want it.

44:15

Right? They want, I want to actually learn the stuff, which the actual learning

44:17

experience is more painful, but so that is what we do in the boot

44:19

camp, and then we give a homework which really enforces the stuff. But if

44:21

you're looking to LearnZK, you're listening to this, you don't have to take our

44:23

boot camp. The book covers the same stuff, and it's free, and you can

44:25

go through it on your own, and a lot of people have. thing for

44:27

the ZK boot camp is if you really want to accelerate your learning for

44:29

some people time is more valuable than money so it's that

44:31

may be a good deal for people like

44:33

that who are really like I need to

44:36

learn the ZK thing and it needs to

44:38

be done by a certain date or something

44:40

like that. Interesting. So but we teach the

44:42

same the same subject everything that

44:45

we charge we also provide for

44:47

free in terms of content. So

44:49

how many have you actually done? We've

44:51

done at least 15. Nice. That's

44:53

more than I thought, I thought

44:55

you'd have done three, but okay.

44:57

Cool. It's been somewhat popular, so.

45:00

That's nice. The people who've

45:02

gone through these boot camps, do you

45:04

know if they've gone on to work

45:07

in CK? And if so, kind of

45:09

at what level? Oh, for sure.

45:11

Oh, this is what I'm most

45:13

proud of. Yeah. So you know,

45:15

the guy, David Wang, who authored

45:17

real-world real-world cryptography. So he's been

45:20

on the show. Okay, great. We've

45:22

had at least two people become

45:24

interns at his company. One guy

45:26

is working full-time now. So that's actually,

45:28

yeah, that was going recent. We've had

45:30

another, and he's still working at Matter

45:33

Labs, the company behind ZK Cink. So

45:35

he was, he is a security engineer

45:37

and is now actually doing stuff like

45:39

writing constraint solvers to make sure that

45:42

the constraints don't have other cases in

45:44

them that are insecure, right? Amazing. Those

45:46

are probably some of the more cool

45:48

examples of people who've gone on to

45:50

work in work in industry. If you

45:53

want to do actual ZK research, you'd

45:55

kind of need a PhD. And people

45:57

are not really open to that. But

45:59

with. Security, if you can, hey, I can

46:02

find bugs and things, then people are

46:04

more lenient about that in terms of

46:06

hiring you. And then for people who

46:08

want to work in ZTA, that's something

46:10

I tell them, you should consider that

46:12

because we've had quite a bit of

46:14

success with getting on boarding people to

46:16

ZK Security. I don't mean the company,

46:18

I mean, in general. Well, also the

46:20

company. But seemingly, there's a funnel directly

46:22

there. So there's definitely quite a bit

46:24

of need for people who understand how

46:26

to find bugs in this sort of

46:28

stuff. And yeah, so those are the

46:30

professional things. But most people who take

46:32

the course are not looking to get

46:34

a dizzy K. Actually, it's mostly taken

46:36

by like tech leads and protocol engineers

46:38

who just want to kind of hedge

46:40

their career in case things go in

46:42

a different direction. Oh, interesting. Yeah, no,

46:44

no. We've had some high profile people

46:46

take the course, like the head of

46:48

research at Ripple and the core engineer

46:50

at Phantom, which is Sonic now. And,

46:52

um. Yeah, I mean, I won't put

46:54

the individual names on Bless, but you

46:56

can see some famous people, if you

46:58

look at our testimonials page, we've taken

47:00

it. So it's really cool for me,

47:02

because some of the people who, like

47:04

one guy, so consensus came out with

47:06

this tool a long time ago, Mythril,

47:08

for doing static analysis on source code.

47:10

And I partly taught myself solidity back

47:12

in the day by writing some solidity

47:14

code and then running it through static

47:16

analyzers and seeing. What mistakes is it

47:18

pointing out? Now, obviously there can be

47:20

a false positive, but they patterned match

47:22

this kind of a mistake for a

47:24

reason, so let me try to understand

47:26

it. So the guy who built that

47:28

tool back in the day and then

47:30

he was acquired by consensus and they

47:32

maintained it for a while, I indirectly

47:35

learned quite a bit of solidity from

47:37

him. Nice. I wonder if you've been

47:39

following the emergence of the ZK VMs

47:41

and some of those teams, because in

47:43

a way the proposal they're making is

47:45

like, you don't need to know ZK

47:47

at all anymore. My advice for people

47:49

is run a non-trivial computation, like something

47:51

slightly more complex than a fibonacci sequence,

47:53

maybe involves a few hash functions. For

47:55

that kind of stuff, if you have

47:57

like a prover network and you're trying

47:59

to build a blockchain and hardware is

48:01

not really a problem, then I think

48:03

that's great. But the marketing definitely incentive...

48:05

is presenting it as something where everything

48:07

is fast and you don't need to

48:09

worry about it, which probably will be

48:11

true in the future. But just run

48:13

some code on your laptop now. Just

48:15

clone their thing and run the unit

48:17

test. It's going to take half a

48:19

day to get through that stuff. And

48:21

none of these programs are particularly complicated.

48:23

You feel like it's not there yet?

48:25

Definitely not. But I think someday it

48:27

will be, right? But then when that

48:29

day comes, somebody has to be developing

48:31

those tools and somebody has to be

48:33

auditinging them. I think the very successful

48:35

engineers know that you don't need to

48:37

know mindset is kind of a loser's

48:39

one. So that's just to be blunt,

48:41

right? So from an economic standpoint, yes,

48:43

you don't want your engineers burdened down

48:45

by additional work. You want the DevX

48:47

to be as smooth as possible, blah,

48:49

blah, blah, blah, if you want to

48:51

scale. But you don't want to make

48:53

it too easy. Well, I guess there's

48:55

drawbacks to that. But I mean, like,

48:57

let's say you just react for react

48:59

for example. barely even knows what they're

49:01

doing can create a react app, right?

49:03

Because you can copy, paste things. It's

49:05

fairly smooth, aside from node modules, taking

49:07

forever to download, everything works pretty well,

49:10

right? But if you want to be

49:12

a senior engineer, right, then you have

49:14

to know, and you want to be

49:16

designing, like, architecting things and managing a

49:18

team, you need to know. Right, but

49:20

you're causing the page to refresh every

49:22

second, like that's not good. So abstractions

49:24

are never perfect. So if you want

49:26

to be good at your job, you

49:28

have to know what's going on behind

49:30

the scenes so that you can avoid

49:32

the pitfalls. I mean, even solidity still

49:34

has leaky abstractions, right? You can't have

49:36

unbounded a razor, the gas cost can

49:38

go nuts. And re-enterency is still not

49:40

abstracted away, for example. So I think

49:42

more power to people for developing this

49:44

stuff, but I'm not concerned about it

49:46

making what we're teaching obsolete, because there's

49:48

going to always need to be at

49:50

least a few people who even... theoretically

49:52

don't need to know it will benefit

49:54

a lot from knowing it. Nice. Have

49:56

you been, and this is going to

49:58

be a little bit of a plug

50:00

for some of the... study groups we've

50:02

done, but have you been paying attention

50:04

to the other ZK resources out there?

50:06

For example, the Moon Math Manual. Justin

50:08

Thaler's book, there's also like the Snarg

50:11

book, which is much more in the

50:13

research side of things, but it came

50:15

out I think last year. So yeah,

50:17

I'm just curious if like you produced

50:20

the ZK book, are you ever like

50:22

drawing on references from other books or

50:24

are you really sticking into your material?

50:26

Are you aware of it? Yeah, I am aware

50:29

of the Moon Math Manual, and I've

50:31

read some parts of it, and Justin

50:33

Thaler, I've heard a lot of good

50:35

things about him. I haven't, I skimmed

50:37

his book at one point, but I

50:40

haven't really read it. The Snarg book

50:42

actually did not know about until this

50:44

call. So, like, is there other

50:46

material that you're recommending your

50:48

students see, basically? That's a

50:50

good question. I'm hoping you're

50:53

going to tell them the

50:55

CK whiteboard sessions, but I understand

50:57

if not. Usually, okay, so

50:59

when people are asking for

51:01

a recommendation and material, they

51:03

want to say, hey, you

51:05

taught how Groth 16 works.

51:07

I want to know how

51:09

Stark or Plunk works. Yeah.

51:11

And I usually tell

51:13

those people, you know, Vitalica is written

51:15

on it. There's other good blog posts

51:17

out there. I can show you those

51:19

blog posts and you can. you know,

51:21

just read a few and you'll get

51:24

that. You already know how Groth 16

51:26

works, so this is not going to

51:28

be that hard for you. So the

51:30

Moon Math Manual, I have, it's quite

51:32

a mathematical thing, definitely, especially compared to

51:34

R. So I think if people read

51:36

the Moon Math Manual after they read

51:38

the ZK book, it's probably easier to

51:40

do it that way. And I've heard

51:42

a lot of good things about

51:45

Justin Thaylor's thing, from my understanding

51:47

and from my brief skim as

51:49

more of a, like theoretically focused,

51:51

which I think is. good to

51:53

do after you've built a working

51:56

example of it like what we

51:58

teach. In terms of inspiration. There

52:00

was this Pinocchio series on

52:02

medium or some archive thing

52:05

that appeared a long time.

52:07

So this was called Why

52:09

and How ZK Snark works?

52:11

Yeah, by Maxim Petkus. Okay.

52:13

So that was a very helpful

52:15

resource for me when I was

52:18

getting into ZK. So he

52:20

taught Pinocchio and I really felt

52:22

like that was not... I felt like

52:25

Roth 16 was a better thing

52:27

to teach because it's actually used

52:29

and it's actually simpler. Now, understanding

52:31

the intuition behind Groth 16 is

52:33

harder probably than Pinocchio, but there's

52:35

fewer moving parts, so there's less

52:37

to, there's less to teach. But

52:39

if you're wanting to learn some

52:41

of the later stuff, I mean, what I

52:43

generally tell people is, there's a danger in

52:45

just knowing things for the fun of it.

52:48

Right, which is fine. As long as you're

52:50

conscious that this is a hot, that I'm

52:52

learning things for the fun of it. But

52:54

as an engineer or an auditor, you need

52:56

to actually produce something, right? That's why we

52:58

try to set our North Star for the

53:01

courses. Like you need to produce the Groth

53:03

16 algorithm. You need to create the ZKVM.

53:05

And then use any resource that's going to

53:07

help you get to that destination. So whether

53:09

that's reading the source code, reading the

53:11

paper, reading other people's blog posts about

53:14

it. And now I do think. Understanding

53:16

other things at a high level can be

53:19

helpful. Like I watch some of the whiteboard,

53:21

the ZK whiteboard stuff. And it's good to

53:23

know, okay, at least structurally this is how

53:25

things are because if I understand the high

53:27

level, it's easy to start with a high

53:29

level idea and then work your way down.

53:32

Or understand the high level idea and then

53:34

learn something that's adjacent to it. Yeah,

53:36

my advice for people, just in general for

53:38

people who are just going out there and

53:40

learning, try to filter it through. Am I

53:43

learning a skill that somebody else will pay

53:45

for? Not knowledge. Like even in academia, you're

53:47

not paid on the basis of being smart

53:49

or whatever. You're paid on the basis of

53:51

publishing papers at the end of the day.

53:54

So you want to make sure that when

53:56

you're learning something, what additional skill

53:58

are tangible outcomes? am I getting

54:00

from this? Not just, oh, I feel good

54:03

about myself because I understand how something works.

54:05

That can be a dangerous feedback mechanism. Now

54:07

if you're doing it for fun, that's okay,

54:09

right? But just be aware that you're

54:11

doing it for fun. What I hear from you is

54:13

this idea of just like adding value.

54:16

Like make sure, like you're doing that

54:18

in a way with the course. You're

54:20

making sure that there's like a value

54:22

add. It's not just. here's a video,

54:24

enjoy, but it is like hands-on, kind

54:26

of keeping people accountable, bringing them together,

54:28

and making sure that people leave with

54:31

the understanding they're supposed to have. I

54:33

like that. And actually, so it's interesting,

54:35

like... I'm going to bring back CK Hack,

54:37

but on our side we've done the content

54:40

production, but it is the study groups where

54:42

we've gone through the content, which in my

54:44

opinion is like the closest thing we've got

54:46

to it, which are these study groups, but

54:49

like I said, quite informal. What I hear

54:51

from you is there's sort of this You're

54:53

trying to create value, you want to make

54:55

sure that people walk away with value. I'm

54:58

also learning from you, by the way. I

55:00

am slightly like trying to learn how you're

55:02

doing this because I think, yeah, we're always

55:04

looking to make our side a bit better

55:07

too, but creating these environments where

55:09

there's enough of a back and forth

55:11

that it keeps people involved. Like another

55:13

thing you mentioned was shaving off those

55:16

hard edges or those like... They could

55:18

be really interesting rabbit holes, but they're going

55:20

to slow you down situations. Making sure that

55:22

there's like a streamline from where you're starting

55:24

with very little info into like building something

55:26

real. You've kind of talked about choosing a

55:29

proving system that also like is quite accepted.

55:31

There is a lot of information about it

55:33

and not getting like what I hear from

55:35

that too is you're not getting caught up.

55:37

or like confused by all that's out there.

55:39

Because I think that's one of the challenges

55:42

as well is like if you just look

55:44

around there's so much you could dive into.

55:46

But each one of those things will have

55:48

nuances and if you don't know the

55:50

basics it's going to be really hard

55:52

for you to navigate between them and

55:54

it could confuse you. Like I'm sure

55:56

it confuses like as a non-developer talking

55:58

about all this stuff. like confused myself on

56:01

that. So this is interesting. This is like

56:03

some some cool feedback. I think for anyone

56:05

who's listening who's like wants to learn, if

56:07

they have sort of a journey they'd like

56:10

to take, these might be good tips. And

56:12

then I guess they could also join you

56:14

guys. Are you doing another boot camps at

56:16

some point soon? Yeah, we. We tend to

56:19

run the ZK boot camp every month. So

56:21

it should. Okay, that's why you do so

56:23

many. Got it. I don't know the exact

56:25

start date of the next one, but

56:27

it's on our Rare Skills.io/ZK dash boot

56:30

camp. The start date will be there.

56:32

We should be one in March. I

56:34

think this is fantastic for our audience

56:36

to know. Yeah, thank you so much

56:38

for sharing this with us. Now, I. I do want

56:41

to ask you one more thing, which

56:43

is kind of off topic before we

56:45

sign off. One of the reasons we

56:47

actually reached out to you initially was

56:49

a viral tweet you shared about quantum

56:51

computing, totally different topic. But then when

56:54

I realized it was you from Rare

56:56

Skills and I knew about this book,

56:58

I actually invited you on to just

57:00

talk about that. But I do want

57:02

to just mention that tweet. This is

57:04

right around the time that the quantum

57:06

computing blog post came out from Google

57:08

and like sort of an announcement and

57:11

you broke it down in a nice

57:13

way. It does still speak to like

57:15

technical education and all of that. But

57:17

let's talk a little bit about that

57:19

because also looking at it before

57:21

this interview, that thing is like

57:23

megaviral. It's like five million views

57:25

or something. And that's not your

57:28

field. So tell me a little bit about

57:30

putting that together. For sure. Every two years

57:32

these big companies try to come out and

57:34

scare people with like look at our big

57:36

breakthrough in quantum computing when reality is just

57:38

like pump their stock or something like that.

57:40

So the timing was of like people were

57:43

like oh my gosh there's quantum supremacy what's

57:45

going on? And so since there's already a

57:47

lot of heightened emotion around the topic and

57:49

then the response being in response to something

57:51

that was already very emotional it's kind of

57:53

the right situation to for the stars to

57:56

align. So there is that I actually don't

57:58

think I wrote that threat that threat. that

58:00

well. If I were to go back

58:02

and do it again, there's actually some

58:04

jumps. The progression in a few cases

58:06

is not like, oh, I should change

58:08

the order of how I discuss this

58:10

thing in a few places. So I

58:12

think just as a meta comment, you

58:15

can be not that good of an

58:17

educator, but if you're talking about something

58:19

where there's nothing on the topic, people

58:21

will think you're good. But if you're

58:23

writing on something where there's

58:25

a lot of information already.

58:27

Your explanation needs to be

58:29

bang on and it needs to

58:31

have beautiful animations and blah blah

58:33

blah blah blah are people just

58:36

not going to care. So I

58:38

do think that it's so those

58:40

are the two ways you can

58:42

be considered a good educator either

58:44

explain well-known things exceptionally well and

58:46

be very entertaining in the process

58:48

or talk about stuff that nobody

58:50

has a clue about. Okay. So it's not

58:52

your best work is what you're saying. It's

58:54

not my best work by a long shot.

58:56

Again, this goes back to like why is

58:58

ZK kind of difficult to learn? Because the

59:01

model of computation is weird to people, right?

59:03

You're not actually writing an algorithm in

59:05

ZK, you're writing constraints that model an

59:07

algorithm. And quantum computing is kind of

59:09

like that, where you're taking advantage of

59:12

the fact that, you know, entanglement can

59:14

carry a lot of information between places

59:16

simultaneously. So, and then you can probabilistically

59:18

arrive at the answer. So quantum computing

59:20

does not work where you get the

59:23

correct answer each time. You have to

59:25

run it. a few times and then

59:27

probabilistically you can figure out that your

59:29

answer is right. Weird. And the other challenge

59:31

with quantum computing is, like with a Sudoku

59:34

puzzle, for example, you know that you got

59:36

the answer, you have to do a search

59:38

to find the answer, but then you can

59:40

verify that the search was correct because

59:42

you just check all the rose columns

59:45

and sub-squares, right? The thing they were

59:47

doing with Google was not a problem that

59:49

you could check. And when they say, oh,

59:51

this was, you know, 10 to the 25

59:53

times faster, well, that's based on extrapolations, not

59:56

based on, no one actually ran the computer

59:58

for 10 to the 25 times. So quantum

1:00:00

gates are kind of like logic gates.

1:00:02

That's why I open the thread that

1:00:05

way. Right. So hopefully, if you do.

1:00:07

I'll add the link to this tweet.

1:00:09

And even though it's not your best

1:00:11

work, we all know it's, I could,

1:00:14

I could preface that too. Yeah. Tweet

1:00:16

from Jeffrey, not his best work. But

1:00:18

yeah, I'll definitely, I'll add it in

1:00:20

the show notes for anyone who wants

1:00:23

to see it. Like for me, reading

1:00:25

it was, it was actually, like, like,

1:00:27

this is a topic that I'd like

1:00:29

to cover a little bit on the

1:00:32

show going forward. You can actually use

1:00:34

quantum computing in the cloud with IBM.

1:00:36

You'll get like five cubits or something

1:00:38

like that and it's like $95 a

1:00:41

minute to use that thing. So if

1:00:43

you have money to blow, you can,

1:00:45

the player under the quantum computer. And

1:00:47

there are quantum languages for it, which

1:00:50

it's not as scary of a field

1:00:52

as people think it might be. I

1:00:54

mean, obviously if you're going to be

1:00:56

a researcher in it, then yeah, that's

1:00:59

tough stuff. But what they did in

1:01:01

their experiment is they wired up a

1:01:03

bunch of random gates together, but the

1:01:05

distribution of the output is expected to

1:01:08

follow a certain pattern. And then the

1:01:10

distribution did, in fact, match the pattern.

1:01:12

They were like, oh look, we managed

1:01:14

to, we got it right. Yeah, we

1:01:17

got it right, but like, but this

1:01:19

is not very useful, right? So, now

1:01:21

they do have some breakthroughs in there,

1:01:23

I'm not knocking them, but obviously, they

1:01:26

carefully chose the context and the phrasing

1:01:28

so that it would maximize the hype,

1:01:30

which we do that in Web 32,

1:01:32

so I can't blame them. But you

1:01:35

can't throw stones in a glass house,

1:01:37

but it was, but I get very

1:01:39

annoyed when people to your. advantage right

1:01:41

so I think Taroon also had he

1:01:44

had in our end of year episode

1:01:46

he'd also cast doubts on on how

1:01:48

how honest the marketing had been there

1:01:50

yeah when you when you read their

1:01:53

actual thing they're very honest or they're

1:01:55

at least they're not as like being

1:01:57

as blunt as I am but like

1:01:59

But to understand the honesty, you need

1:02:02

to have the technical, you need to

1:02:04

know like what are random gates and what

1:02:06

was actually being done. So when people

1:02:08

talk about quantum computing, a lot of

1:02:11

it just sounds like chat deputy wrote

1:02:13

it. Oh, you know, there's entanglement, therefore

1:02:15

it's faster. We have zero one bits,

1:02:17

but now we can have more than

1:02:19

zero or one. Okay, that's cool. But

1:02:21

how does that translate into faster computation?

1:02:24

And how does that tell me what

1:02:26

Google was doing? Okay, at least has

1:02:28

some idea of what's actually happening. And

1:02:30

then speak about it in terms of

1:02:32

that at least someone who has an

1:02:35

introduction to computer science would

1:02:37

understand. Something that you just sort

1:02:39

of highlighted though that the similarity in

1:02:41

ZK where it's like you're trying to

1:02:44

use existing paradigms of computer programming and

1:02:46

you're adding it into ZK circuits in

1:02:48

one case, but also into a quantum

1:02:50

context. Like that's actually I hadn't ever

1:02:52

sort of made that connection point that

1:02:55

like there may actually be just like

1:02:57

brand new with paradigms like in ZK

1:02:59

as well potentially like why are we

1:03:01

just trying to rewrite existing things? I

1:03:03

think well this is actually because I

1:03:06

studied remember I studied computational theory so

1:03:08

fundamentally everything boils down to either carrying

1:03:10

out the computation or proving the computation

1:03:13

was was correct. Or another thing is

1:03:15

why can quantum computers theoretically give us

1:03:17

a speed up when it's just quantum

1:03:20

state collapsing? Well, because we can translate

1:03:22

a real problem theoretically into something that

1:03:24

collapses and then that answer comes out

1:03:26

of there and tells us something that

1:03:29

tells us something that tells us something

1:03:31

that's useful. So once you just kind

1:03:33

of get that there's really only a

1:03:36

couple modalities of computation and then what

1:03:38

we see in ZK or quantum computing

1:03:40

is just a different manifestation of it.

1:03:42

This is one thing why I really

1:03:44

encourage people to learn the fundamentals. We've

1:03:46

known everything about computation since Alan Turing

1:03:48

and Lambda Calculus, but everything boils down

1:03:51

to the basics. A lot of it

1:03:53

is just engineering to get the basics

1:03:55

running faster. Like we might, like, think

1:03:57

about the advantages from, you know, for

1:03:59

these. EK algorithms that come out. A

1:04:01

lot of them are not fundamental breakthroughs.

1:04:04

They're just very, and they're frankly more

1:04:06

of an engineering thing, which matters a

1:04:08

lot, right? Because invidia is not doing

1:04:10

anything innovative. It's all engineering, in terms

1:04:13

of fundamental, what we think of as

1:04:15

fundamental science, or TSMC. We just figured

1:04:17

out how to like get the UV

1:04:19

lights to, making things smaller and

1:04:22

smaller, sharper and sharper. more efficient

1:04:24

using less energy, making less heat.

1:04:26

Exactly. Yeah. Or even with AI. I

1:04:28

think a lot of the models that

1:04:30

we use as like for chats could

1:04:33

theoretically be coded in about 900 lines

1:04:35

of code or something like that. They're

1:04:37

not that complex. What's complex is how

1:04:39

to get all this data to be

1:04:41

run through a data center and you're

1:04:44

not running into bandwidth issues. Although there

1:04:46

are like I think there there is some edge

1:04:48

like some edge where there's new techniques at least

1:04:50

which I think is more in the research front

1:04:52

which are changing kind of the ways this thing

1:04:54

works. Yeah I definitely don't want to say

1:04:56

that there's no scientific advancement happening but

1:04:58

a lot of what we perceive a

1:05:01

scientific advancement is really engineering advancement and

1:05:03

engineering is oftentimes based on science that's

1:05:05

very very old so what I recommend

1:05:07

this is one thing that I think

1:05:09

that's helped me a lot just because

1:05:12

I studied the foundations of computing very

1:05:14

very closely web three is easy

1:05:16

mode for me so well solidities

1:05:18

like it's never been described as like

1:05:21

hard more that like hard to get

1:05:23

it right right like it's like it's

1:05:25

easy it's maybe too easy but it's

1:05:27

therefore easy to also incorporate bugs

1:05:29

and people who maybe aren't

1:05:32

as who aren't as familiar

1:05:34

with the EVM have historically

1:05:36

just made mistakes because of the

1:05:38

way of how easy it is. I

1:05:40

always found the solidity story kind of

1:05:42

an interesting one. It's like too close

1:05:44

to JavaScript or something. It's too familiar,

1:05:46

but it's not exactly the same thing.

1:05:49

But I guess ZK is very different

1:05:51

though. ZK is like not familiar for

1:05:53

people. So they are kind of learning

1:05:55

these paradigms, especially if you're working at

1:05:57

the level you guys are. They're learning

1:05:59

these. in a way for the first time.

1:06:01

Yeah, so I would just say

1:06:03

in general, it's very important to

1:06:05

have the fundamentals in place because

1:06:07

a lot of what we call

1:06:09

new knowledge is really just a

1:06:11

rearrangement of the fundamentals. So if

1:06:13

you really solidly know your elliptic

1:06:16

curves, you really solidly know your

1:06:18

polynomials and your linear algebra, or

1:06:20

at least some parts of it,

1:06:22

and then. People who are creating

1:06:24

content about this, and we have

1:06:26

study groups, we have courses, we

1:06:28

have books, and blog posts, you

1:06:30

have ZK Whiteboard, these will help you

1:06:32

stay caught up with things. But if you...

1:06:34

really want to have mastery on the subject

1:06:36

and not be dependent on other people to

1:06:38

break it down for you, then what people

1:06:41

think of as a skill of reading papers

1:06:43

is like, oh, I know all the latest

1:06:45

information. It's not, you're really fluent in the

1:06:47

vocabulary that they use. What they think is

1:06:50

obvious to everybody that they don't even explain.

1:06:52

Those are the things that you really want

1:06:54

to be fluent and if you want to

1:06:56

stay caught up with people on your own

1:06:59

and not doing it through other platforms. That's,

1:07:01

I really think is what the... core skill

1:07:03

that people miss. It's not chasing

1:07:05

whatever the latest hot thing is.

1:07:07

Yeah, cool. Jeffrey, thank you so much

1:07:10

for coming on the show and sharing

1:07:12

with us basically the rare skills story,

1:07:14

what you guys are offering, how you're

1:07:16

thinking about it. Also, thanks for the

1:07:18

little... kind of insight into your thoughts

1:07:20

on quantum computing and that tweet. Yeah,

1:07:23

thank you so much. Yeah, I love

1:07:25

talking about this stuff. Thanks for having me.

1:07:27

Yeah, it was also great to meet you.

1:07:29

I feel like we work in parallel streams,

1:07:31

but had never actually come into contact before.

1:07:33

So I'm glad that.