1
00:00:00,500 --> 00:00:04,900
This is epicenter episode 356 
with guest, tarun chitra. 

2
00:00:19,300 --> 00:00:21,800
I am service equity and you're 
listening to epicenter the 

3
00:00:21,800 --> 00:00:24,400
podcast where we interview 
crypto, Founders, Builders and 

4
00:00:24,400 --> 00:00:26,600
thought leaders. 
On this show, we dive deep. 

5
00:00:26,600 --> 00:00:29,500
Learn how things work, at a 
technical level, and we fly high

6
00:00:29,500 --> 00:00:32,299
to understand Visionary, 
Concepts, and long-term trends. 

7
00:00:32,900 --> 00:00:35,300
If you like the show and you'd 
like to support us, the best way

8
00:00:35,300 --> 00:00:37,900
to do that is to leave a review 
on a podcast. 

9
00:00:37,900 --> 00:00:41,200
It helps people find the show. 
It helps people know that we are

10
00:00:41,200 --> 00:00:43,800
one of the best. 
Cashner crypto space, and we're 

11
00:00:43,800 --> 00:00:45,400
always happy to read your 
reviews. 

12
00:00:45,600 --> 00:00:48,600
So if you're on a Mac or iOS 
device, the easiest way to do 

13
00:00:48,600 --> 00:00:51,700
that is to go to epicenter of 
rocks /, Apple. 

14
00:00:52,800 --> 00:00:54,300
Today, our guest is tarun 
chitra. 

15
00:00:54,300 --> 00:00:56,900
He's the founder and CEO of 
Gauntlet Gauntlet is a 

16
00:00:56,908 --> 00:01:01,000
simulation platform for building
Financial models of blockchain, 

17
00:01:01,000 --> 00:01:03,800
protocols and applications. 
So when you're building a 

18
00:01:03,800 --> 00:01:07,300
blockchain protocol, these days,
it's become expected to have the

19
00:01:07,300 --> 00:01:09,400
code audited. 
This is like a, you know, an 

20
00:01:09,400 --> 00:01:12,000
essential thing now. 
And it's a good thing, of 

21
00:01:12,000 --> 00:01:15,100
course, and security audit will 
look at the code to make sure 

22
00:01:15,100 --> 00:01:18,700
there's no bugs and usually 
security audit goes into 

23
00:01:18,700 --> 00:01:22,500
mechanism design, but call it 
goes even further and they do 

24
00:01:22,600 --> 00:01:27,000
Shinzon the mechanism and 
different layers of the stack so

25
00:01:27,000 --> 00:01:29,600
they use machine learning 
methods to simulate different 

26
00:01:29,600 --> 00:01:33,900
environments with various user 
behaviors and to see how the 

27
00:01:33,900 --> 00:01:35,600
system holds up under those 
conditions. 

28
00:01:35,600 --> 00:01:38,500
So they perform analysis on 
things like the core mechanism 

29
00:01:38,500 --> 00:01:42,200
to test for liveness and block 
propagation. 

30
00:01:42,200 --> 00:01:46,400
And they can also do higher 
level analysis and test entire 

31
00:01:46,400 --> 00:01:49,500
markets to see how the market 
will react to a certain 

32
00:01:49,500 --> 00:01:51,600
protocol. 
So this was a really cool 

33
00:01:51,600 --> 00:01:54,800
interview because it's heavy on 
machine learning and statistical

34
00:01:54,800 --> 00:01:58,000
analysis, which is not an area 
that I'm immensely comfortable 

35
00:01:58,000 --> 00:01:59,700
with. 
But I find it really fascinating

36
00:01:59,900 --> 00:02:01,600
nonetheless. 
And when it's applied to 

37
00:02:01,607 --> 00:02:03,400
blockchains, I think there's a 
potential here for this to 

38
00:02:03,400 --> 00:02:07,800
become much less security audits
something that becomes expected 

39
00:02:07,800 --> 00:02:10,900
by users by investors and by the
community for new protocols. 

40
00:02:10,900 --> 00:02:14,000
And you know, it's probably a 
good thing because as we've seen

41
00:02:14,000 --> 00:02:17,700
recently, these protocols can 
lock in a lot of value. 

42
00:02:18,100 --> 00:02:20,400
They're also doing research 
around transaction fees which 

43
00:02:20,400 --> 00:02:22,500
apparently is an entirely new 
space. 

44
00:02:22,900 --> 00:02:27,100
In the area of mechanism design,
a lot of the mechanism design 

45
00:02:27,100 --> 00:02:30,400
research apparently focuses 
around things like auctions for 

46
00:02:30,400 --> 00:02:31,800
like Google ads and things like 
that. 

47
00:02:31,800 --> 00:02:34,800
And of course this is an area in
which like you don't really need

48
00:02:34,800 --> 00:02:37,400
to consider transaction fees 
because you know, transactions 

49
00:02:37,400 --> 00:02:40,400
are abundant and free whereas 
with blockchains that's of 

50
00:02:40,400 --> 00:02:43,900
course not the case. 
And so given the the current 

51
00:02:43,900 --> 00:02:47,500
surge of transaction fees. 
For example, in ethereum is an 

52
00:02:47,500 --> 00:02:50,100
area that's going to become 
increasingly, interesting and 

53
00:02:50,100 --> 00:02:54,300
increasingly relevant as bull. 
It's have, you know, these, 

54
00:02:54,300 --> 00:02:57,100
these effects on the transaction
fees in a network. 

55
00:02:57,400 --> 00:03:00,100
So at the beginning, we talked a
lot about chip manufacturing, 

56
00:03:00,100 --> 00:03:02,900
which is since to room was 
previously in that sector, he's 

57
00:03:02,900 --> 00:03:04,400
got some interesting stories 
there. 

58
00:03:04,900 --> 00:03:08,400
And we also briefly very briefly
touched on Al Gore and which is 

59
00:03:08,400 --> 00:03:11,400
convenient because they are 
sponsoring this episode. 

60
00:03:11,400 --> 00:03:14,400
So, if you haven't heard our 
episodes on Al Gore and I would 

61
00:03:14,400 --> 00:03:16,900
encourage you to go back and 
listen to them, we did one in 

62
00:03:16,900 --> 00:03:20,200
January with Steve coconut house
and Sylvia McCallie and one way 

63
00:03:20,200 --> 00:03:24,400
back in 2017 with sales. 
Oh, this was before El Gran was 

64
00:03:24,400 --> 00:03:26,000
even company. 
He just written the white paper 

65
00:03:26,000 --> 00:03:28,000
at this point. 
Anyway, they're doing really 

66
00:03:28,000 --> 00:03:31,100
cool stuff to improve developer 
experience specifically around 

67
00:03:31,100 --> 00:03:34,800
building defy applications and 
I'll tell you all about that a 

68
00:03:34,800 --> 00:03:38,100
little bit later on during the 
interview, but for now, I give 

69
00:03:38,100 --> 00:03:40,400
you our conversation with tarun 
chitra. 

70
00:03:42,700 --> 00:03:47,500
We are today with turn treat ra.
He is the CEO and founder of 

71
00:03:47,500 --> 00:03:50,800
Gauntlet Gauntlet network. 
Is the domain name? 

72
00:03:50,800 --> 00:03:54,200
Is it actually a network or is 
it sort of just a company? 

73
00:03:55,500 --> 00:03:58,500
So I tried to get all the 
gauntlet domain names and 

74
00:03:58,500 --> 00:04:02,100
fortunately the only one that 
was really open was dot Network.

75
00:04:02,300 --> 00:04:04,100
I think. 
Now we're going to try to see if

76
00:04:04,100 --> 00:04:07,100
there's a TLD dr. 
Jenn and by gamma dr. 

77
00:04:07,100 --> 00:04:12,400
Jenn but we have gotten without 
five but basically in 2017 at 18

78
00:04:12,400 --> 00:04:16,700
I bought the dog network name 
and it kind of stuck and we had 

79
00:04:16,700 --> 00:04:19,899
to incorporate with some name so
I chose Networks. 

80
00:04:20,899 --> 00:04:23,800
It's a great to have you on. 
You have a really interesting 

81
00:04:23,800 --> 00:04:27,600
background that gives you a bit 
of a different perspective from 

82
00:04:27,600 --> 00:04:30,200
a lot of the people in the 
crypto space or especially at 

83
00:04:30,207 --> 00:04:33,400
least in the defy space. 
Could you tell us a little bit 

84
00:04:33,500 --> 00:04:35,300
about it? 
What were you doing before you 

85
00:04:35,300 --> 00:04:40,100
got in involved with crypto? 
After I graduated college, I 

86
00:04:40,100 --> 00:04:43,900
worked at this billionaires. 
Research Institute was called 

87
00:04:43,900 --> 00:04:48,500
disha research. 
And this person who worked in 

88
00:04:48,500 --> 00:04:54,300
trading, he wanted to spend his 
money on building a 6. 

89
00:04:54,600 --> 00:04:57,100
So application-specific 
integrated circuit. 

90
00:04:57,100 --> 00:05:01,800
So, these custom Hardware 
devices for doing computational 

91
00:05:01,800 --> 00:05:04,600
biology and Drug Discovery and 
physics research. 

92
00:05:05,500 --> 00:05:09,000
This is the same company as like
the Esau Investment Group or is 

93
00:05:09,000 --> 00:05:11,400
this a separate? 
It's the same thing. 

94
00:05:11,400 --> 00:05:16,000
Basically one of the branches of
the Investment Group was working

95
00:05:16,000 --> 00:05:18,400
on this building ethics for 
physics research. 

96
00:05:18,400 --> 00:05:21,700
So, basically, the story of 
David not too kind of 

97
00:05:21,700 --> 00:05:25,000
recursively, add some stories is
in the 1980s. 

98
00:05:25,200 --> 00:05:29,100
He was an assistant professor, I
think he didn't get tenure and 

99
00:05:29,100 --> 00:05:31,300
so you know, if you think about 
people, you know who are really 

100
00:05:31,300 --> 00:05:34,200
smart but like they don't get 
tenure for some reason or 

101
00:05:34,200 --> 00:05:38,400
another in his case, it was he 
was really working on non Von 

102
00:05:38,400 --> 00:05:43,700
Neumann computers. 
And 1986 was the time when intel

103
00:05:43,700 --> 00:05:47,800
was kind of like, about to hit 
moon at that time. 

104
00:05:47,800 --> 00:05:50,900
No one wants to hire someone 
who's doing kind of custom 

105
00:05:50,900 --> 00:05:53,200
computer architecture. 
Everyone's like, no Intel's 

106
00:05:53,200 --> 00:05:54,900
going to win. 
So we're going to like hire 

107
00:05:54,900 --> 00:05:57,700
people who are going to make 
Intel hit Moore's Law. 

108
00:05:58,100 --> 00:06:02,500
And so then he went and working 
financed somehow secretly hidden

109
00:06:02,500 --> 00:06:05,300
underneath him working in 
finance and trying to make money

110
00:06:05,900 --> 00:06:08,400
with this idea that he likes 
still want to work on non Von 

111
00:06:08,400 --> 00:06:11,200
Neumann computers. 
So, Find out when computers 

112
00:06:11,200 --> 00:06:14,700
means that like a normal 
computer architecture, kind of 

113
00:06:14,700 --> 00:06:19,300
separates memory and compute 
potentially separates those two 

114
00:06:19,600 --> 00:06:22,200
Von Neumann, computer 
architectures, kind of our what 

115
00:06:22,200 --> 00:06:25,300
you have in your computer, or 
your phone right now, for the 

116
00:06:25,300 --> 00:06:29,400
most part, where compute and 
data are processed in the same 

117
00:06:29,400 --> 00:06:32,500
sort of Pipeline. 
And so I think in the 80s people

118
00:06:32,500 --> 00:06:36,100
didn't know what would win like 
what Hardware would be in your 

119
00:06:36,100 --> 00:06:39,600
devices that you own. 
He was kind of always interested

120
00:06:39,600 --> 00:06:41,400
in that. 
And one of the applications of 

121
00:06:41,407 --> 00:06:44,500
these kind of very esoteric, 
architectures is building 

122
00:06:44,500 --> 00:06:48,800
supercomputer is that we're 
really good at solving physics 

123
00:06:48,800 --> 00:06:51,900
problems and also solving sort 
of like computational biology 

124
00:06:51,900 --> 00:06:56,100
problems at large. 
And he kind of, in the back of 

125
00:06:56,100 --> 00:06:59,400
his mind, he was always like, 
hey, if I become rich enough, 

126
00:06:59,400 --> 00:07:02,400
I'm just going to spend all my 
money on building these custom 

127
00:07:02,400 --> 00:07:05,400
computer architectures for 
something useful for society. 

128
00:07:05,700 --> 00:07:08,900
And so once he became a 
billionaire, he started trying 

129
00:07:08,900 --> 00:07:13,100
to prove theorems About whether 
it was possible to do better 

130
00:07:13,200 --> 00:07:17,400
than these like Intel to sell 
architectures and he proved this

131
00:07:17,400 --> 00:07:21,000
theorem insert 2004, which is 
kind of great because it uses 

132
00:07:21,000 --> 00:07:25,100
only high school math. 
To make this point that you can 

133
00:07:25,100 --> 00:07:28,200
actually do significantly better
like very much better in the 

134
00:07:28,200 --> 00:07:30,600
sense that Intel style 
processors. 

135
00:07:30,900 --> 00:07:35,300
If you try to parallelize this 
computation will always take a 

136
00:07:35,300 --> 00:07:38,100
finite amount of time even if 
you had an infinite number of 

137
00:07:38,100 --> 00:07:42,200
processors, but there, Exists 
kind of crazy architectures that

138
00:07:42,200 --> 00:07:44,800
would take, you know, basically 
zero time as the number of 

139
00:07:44,800 --> 00:07:49,000
processors goes to Infinity. 
So we actually once he kind of 

140
00:07:49,000 --> 00:07:51,700
prove this theorem I guess you 
know once you're a billionaire 

141
00:07:51,700 --> 00:07:54,700
you're like I'm going to build 
Hardware to prove that my 

142
00:07:54,700 --> 00:07:58,300
theorem is correct. 
So that's the Genesis story. 

143
00:07:58,600 --> 00:08:01,900
I work there. 
And in 2011 there weren't many 

144
00:08:01,900 --> 00:08:04,400
people building, a sex. 
Most of the people building a 

145
00:08:04,400 --> 00:08:06,500
six. 
Were certainly not Bitcoin 

146
00:08:06,500 --> 00:08:09,000
miners. 
They were mainly telecom 

147
00:08:09,000 --> 00:08:11,600
companies. 
So One of the things that's 

148
00:08:11,600 --> 00:08:15,500
interesting, is that the only 
people who really needed custom 

149
00:08:15,500 --> 00:08:19,500
Hardware, were people doing 
really low latency fast Fourier 

150
00:08:19,500 --> 00:08:23,800
transforms ffts, and most of the
people doing that were in 

151
00:08:23,800 --> 00:08:25,500
Telecom. 
There's no neural Nets. 

152
00:08:25,900 --> 00:08:28,300
There is no self driving cars. 
There's none of that type of 

153
00:08:28,300 --> 00:08:32,400
stuff when we were talking to 
suppliers because we're not 

154
00:08:32,400 --> 00:08:34,700
someone who's like Chinese, 
Samsung or Apple building a 

155
00:08:34,700 --> 00:08:38,100
whole infrastructure on it. 
We wanted kind of like pay them 

156
00:08:38,100 --> 00:08:42,299
for their excess Supply So it's 
like we want to build 10,000 

157
00:08:42,299 --> 00:08:46,300
chips Intel and Apple and 
Samsung or each building a 

158
00:08:46,308 --> 00:08:49,400
billion chips. 
But the factory is that you that

159
00:08:49,400 --> 00:08:52,600
build that stuff they have 
excess Supply sometimes like the

160
00:08:52,700 --> 00:08:56,300
one Factory might have an excess
10,000 and they're like oh well 

161
00:08:56,400 --> 00:08:58,700
can we sign someone who wants to
buy that? 

162
00:08:58,800 --> 00:09:01,900
And so the way that excess 
Supply gets old, it's a sort of 

163
00:09:01,900 --> 00:09:05,600
gets auctioned off to different 
Market participants and at that 

164
00:09:05,600 --> 00:09:09,200
time we were literally the only 
ones buying this type of space 

165
00:09:09,200 --> 00:09:13,200
is like us And like random 
Telecom company in Japan, in 

166
00:09:13,200 --> 00:09:15,900
2011 and 12. 
One of the companies that sort 

167
00:09:15,900 --> 00:09:19,400
of became Avalon started buying 
a bunch of this chip space. 

168
00:09:19,400 --> 00:09:22,900
And so we were like talking our 
supplier and like, hey, we sent 

169
00:09:22,900 --> 00:09:27,400
you, you 25 million dollars when
our chips coming and then we got

170
00:09:27,400 --> 00:09:29,300
ghosted. 
This is getting ghosted, 

171
00:09:29,300 --> 00:09:31,700
Pretender. 
So that was weird. 

172
00:09:31,800 --> 00:09:35,500
And at some level we were like, 
what the hell? 

173
00:09:35,500 --> 00:09:36,700
We just gave you a bunch of 
money. 

174
00:09:36,700 --> 00:09:39,300
You just disappeared like and 
then they came back eventually 

175
00:09:39,300 --> 00:09:42,800
and were like Sorry, we're going
to do your chips, you know, the 

176
00:09:42,800 --> 00:09:45,200
next batch house, the 10% 
discount sound. 

177
00:09:45,200 --> 00:09:48,000
And that was the first time I've
really thought seriously about 

178
00:09:48,000 --> 00:09:50,100
Bitcoin. 
After that, I kind of started 

179
00:09:50,100 --> 00:09:54,100
mining How did that go into 
mining was so you found out the 

180
00:09:54,100 --> 00:09:57,500
people who like took your order.
We're basically Bitcoin miners 

181
00:09:57,500 --> 00:10:02,800
who are like a hardware Fab. 
Let's say they have 10,000 spots

182
00:10:03,000 --> 00:10:05,000
that like imagine a physical 
wafer. 

183
00:10:05,500 --> 00:10:09,300
Now, cut up the wafer into 
10,000, like, 1 centimeter by 1.

184
00:10:09,300 --> 00:10:14,600
Centimeter little units. 
Let's say apple takes up 9,000 

185
00:10:14,600 --> 00:10:16,800
of them. 
So there's 1000 left and what 

186
00:10:16,800 --> 00:10:19,100
they do is they auction off the 
physical space. 

187
00:10:19,100 --> 00:10:22,600
Imagine block space. 
But physically auctioned off. 

188
00:10:22,900 --> 00:10:25,200
Usually, it's kind of a fair 
auction. 

189
00:10:25,600 --> 00:10:29,800
Like, you're like, okay, I put 
in a bid, I wanted, you know, 

190
00:10:29,800 --> 00:10:33,100
8,000 of those spots for $100 
each. 

191
00:10:33,300 --> 00:10:37,200
So we put in our bid and then we
got told, hey, you want and then

192
00:10:37,300 --> 00:10:40,000
the person disappeared and so, 
once you win, you put, you have 

193
00:10:40,000 --> 00:10:42,800
to put money in escrow. 
So you post the money that 

194
00:10:42,800 --> 00:10:46,100
you're supposed to post 
collateral and then you're 

195
00:10:46,100 --> 00:10:49,100
supposed to eventually get paid 
or eventually, get your hardware

196
00:10:49,100 --> 00:10:51,800
and then they take your money. 
They just Kind of kept it in 

197
00:10:51,800 --> 00:10:54,800
escrow for like six months way 
longer than it was supposed to 

198
00:10:54,808 --> 00:10:57,900
be and we were like, what the 
hell and so what happened was 

199
00:10:57,900 --> 00:11:01,200
this Bitcoin miner went to the 
supplier because the Bitcoin 

200
00:11:01,200 --> 00:11:03,800
miner was in Taiwan and the 
supplier is in Taiwan and they 

201
00:11:03,800 --> 00:11:07,600
like just I guess they knew each
other and they're like look I 

202
00:11:07,608 --> 00:11:11,500
know you close this auction but 
we need to make these miners by 

203
00:11:11,800 --> 00:11:14,000
X date and we will pay you 
anything. 

204
00:11:15,200 --> 00:11:17,500
It's kind of tangential to this 
conversation, but people don't 

205
00:11:17,500 --> 00:11:20,400
really realize this strategic 
importance of Chip 

206
00:11:20,400 --> 00:11:22,000
manufacturers, because there 
aren't that many. 

207
00:11:22,000 --> 00:11:24,600
I think you probably know this 
more a lot better than I do. 

208
00:11:24,600 --> 00:11:28,000
But like Intel, for example, is 
a chip designer and a chip Fab 

209
00:11:28,000 --> 00:11:32,400
and in the last 20 years, or so 
there's been a shift towards, 

210
00:11:32,400 --> 00:11:35,700
you know, like more the chip Fab
model and like the designers and

211
00:11:35,700 --> 00:11:38,900
the Fabs are now separated. 
And so, like companies like 

212
00:11:38,900 --> 00:11:43,000
tsmc, which is like this big 
chip Fab in Taiwan, I've kind of

213
00:11:43,008 --> 00:11:47,200
like one market And apple, and 
all these companies get their 

214
00:11:47,208 --> 00:11:50,100
chips made there. 
And the question here, that's 

215
00:11:50,100 --> 00:11:53,300
kind of interesting in the 
current geopolitical context and

216
00:11:53,300 --> 00:11:56,100
this kind of example that you 
have kind of exemplifies this 

217
00:11:56,100 --> 00:12:00,400
very thing is that these chip 
manufacturers are an Arm's Reach

218
00:12:00,400 --> 00:12:04,600
away from China and they're very
far away from, you know, the us 

219
00:12:04,600 --> 00:12:06,100
or Europe and other Western 
countries. 

220
00:12:06,108 --> 00:12:11,800
And there are strategic to you 
know critical military and 

221
00:12:11,800 --> 00:12:15,700
Industrial infrastructure in 
West the amount of Of chips 

222
00:12:15,700 --> 00:12:19,000
that, you know, the u.s. can 
produce on its own without the 

223
00:12:19,000 --> 00:12:21,400
suppliers. 
It's like insanely small 

224
00:12:21,400 --> 00:12:24,300
compared to just like what tsmc 
you can put out or like, 

225
00:12:24,300 --> 00:12:25,400
Samsung. 
For example. 

226
00:12:25,900 --> 00:12:28,100
What your thoughts on that? 
Like, knowing this ecosystem? 

227
00:12:28,100 --> 00:12:31,900
A lot better than I do. 
I think the u.s. is actually 

228
00:12:31,900 --> 00:12:34,100
completely uncompetitive at this
point. 

229
00:12:34,200 --> 00:12:38,300
There's globalfoundries, which 
has this kind of big Fab in 

230
00:12:38,500 --> 00:12:40,400
Upstate New York and Long 
Island. 

231
00:12:40,500 --> 00:12:43,900
I guess now they're kind of like
split into two and then there is

232
00:12:43,900 --> 00:12:46,200
like the Intel's in Has their 
own Fab. 

233
00:12:46,200 --> 00:12:49,500
So they have one in Arizona and 
the US government, especially in

234
00:12:49,500 --> 00:12:53,700
the current kind of strong arm 
Administration, type of nonsense

235
00:12:53,800 --> 00:12:56,700
is trying to be like, hey, you 
have to, like build your chips 

236
00:12:56,700 --> 00:12:58,800
in the u.s. if you want to sell 
them here or something. 

237
00:12:59,400 --> 00:13:02,800
That's not a very good point of 
Leverage in the long run. 

238
00:13:02,800 --> 00:13:04,900
Right. 
Because one of the more 

239
00:13:04,900 --> 00:13:07,400
impressive things about Moore's,
Law is Moore's. 

240
00:13:07,400 --> 00:13:10,500
Law actually is a 
self-fulfilling prophecy Gordon 

241
00:13:10,500 --> 00:13:14,500
Moore said this kind of 
apocryphal thing of like oh 

242
00:13:14,600 --> 00:13:18,100
every eight 18 months, your chip
frequencies going to double it 

243
00:13:18,100 --> 00:13:20,800
turned into its own kind of War,
right? 

244
00:13:20,800 --> 00:13:25,200
So like, every 18 months, these 
companies would have liked the 

245
00:13:25,200 --> 00:13:27,800
chip designers Benchmark 
themselves on how close they 

246
00:13:27,800 --> 00:13:30,200
were. 
And then once they had a design 

247
00:13:30,200 --> 00:13:34,400
that could achieve, that sort of
doubling rate, then they would 

248
00:13:34,400 --> 00:13:37,400
go through this entire process 
of finding suppliers who would 

249
00:13:37,400 --> 00:13:40,700
like be able to do that. 
And the suppliers also had to 

250
00:13:40,700 --> 00:13:44,500
follow the like, hey, we need to
double every 18 months kind of 

251
00:13:44,500 --> 00:13:46,700
rule. 
And you had this cycle of like 

252
00:13:46,900 --> 00:13:51,000
chip supplier gives you design 
suppliers who are like, oh man, 

253
00:13:51,000 --> 00:13:53,800
we need to Source. 
This really rare material to 

254
00:13:53,800 --> 00:13:57,200
like, make this happen. 
Like we're going to spend all of

255
00:13:57,200 --> 00:14:00,800
our money trying to do. 
That leads to successful 

256
00:14:00,800 --> 00:14:03,100
Moore's. 
Law thing, leads to lots of 

257
00:14:03,100 --> 00:14:05,500
chips. 
Sold leads to chip designer, 

258
00:14:05,500 --> 00:14:09,600
forcing supplier to do this, and
there's an ecosystem effect, 

259
00:14:09,600 --> 00:14:14,300
kind of, not unlike cars. 
We're like the car. 

260
00:14:15,100 --> 00:14:18,400
Sure isn't really the True 
end-all-be-all Manufacturer. 

261
00:14:18,800 --> 00:14:22,900
There's this whole network of 
suppliers who is necessary for 

262
00:14:22,900 --> 00:14:25,500
it to kind of you to get the 
final product. 

263
00:14:25,500 --> 00:14:28,700
And the chip designer is the 
Intel's of the world, plus the 

264
00:14:28,700 --> 00:14:32,700
suppliers who are making the 
little subcomponents had to work

265
00:14:32,700 --> 00:14:36,700
together cooperatively for this 
like, very long time period in 

266
00:14:36,700 --> 00:14:39,600
order to achieve the current 
kind of Sabbath. 

267
00:14:40,100 --> 00:14:44,300
And the supplier is not just the
Fabs themselves are all in Asia 

268
00:14:44,300 --> 00:14:46,800
it, right? 
Like Entire supply chain is 

269
00:14:46,800 --> 00:14:49,200
completely in Asia. 
There's literally nothing in the

270
00:14:49,200 --> 00:14:53,000
u.s. 
I think it's a farce when the US

271
00:14:53,000 --> 00:14:55,100
was like we're going to take 
back all this me. 

272
00:14:55,200 --> 00:14:59,000
Like, it's a 30-year effort of 
building out multiple 

273
00:14:59,000 --> 00:15:02,100
Industries, right? 
Like one of the things that's 

274
00:15:02,100 --> 00:15:04,000
very important to getting to sub
10. 

275
00:15:04,000 --> 00:15:07,800
Nanometers chips is something 
called Extreme UV. 

276
00:15:08,000 --> 00:15:10,600
So, it's building these really 
crazy lasers. 

277
00:15:11,300 --> 00:15:13,100
I do the same movies that 
prevent the coronavirus. 

278
00:15:13,100 --> 00:15:15,000
Exactly. 
It. 

279
00:15:15,100 --> 00:15:19,400
Like, these really crazy lasers 
that are very like hard to build

280
00:15:19,800 --> 00:15:23,500
and Intel has claimed that. 
Hey, they've been working on it 

281
00:15:23,500 --> 00:15:26,800
for 20 years of like, we can 
build these like really crazy 

282
00:15:26,800 --> 00:15:28,600
lasers. 
The reason you need these 

283
00:15:28,600 --> 00:15:31,400
lasers, is that when you have a 
chip, you have a piece of 

284
00:15:31,500 --> 00:15:34,500
silicon and then you build, 
what's called a mask. 

285
00:15:34,900 --> 00:15:38,200
So the mask covers a piece of 
silicon, and then you shine some

286
00:15:38,200 --> 00:15:41,200
type of electromagnetic 
radiation on there and etches 

287
00:15:41,200 --> 00:15:44,500
it, like, cut it etch-a-sketches
out, your circuit design. 

288
00:15:44,800 --> 00:15:48,800
But This whole industry of these
like extreme UV lasers in order 

289
00:15:48,800 --> 00:15:52,700
to get like the size of the 
little thing to be really small.

290
00:15:52,700 --> 00:15:55,000
So you can pack more transistors
on a chip. 

291
00:15:55,100 --> 00:15:56,500
You need to build these custom 
lasers. 

292
00:15:56,500 --> 00:16:00,200
The only place in the world that
you can make the kind of crazy 

293
00:16:00,200 --> 00:16:04,300
glass that you need. 
For the laser is in sort of 

294
00:16:04,500 --> 00:16:06,400
Mongolia inner Mongolia and 
China. 

295
00:16:06,600 --> 00:16:10,000
And there's just like little 
tidbits, like that, like, oh, 

296
00:16:10,000 --> 00:16:13,600
well, we need this type of glass
for this type of thing, or we 

297
00:16:13,600 --> 00:16:16,700
need this type of silicon, or we
Need this Rare Earth material. 

298
00:16:17,100 --> 00:16:19,600
Those are all things that you 
need to build. 

299
00:16:19,600 --> 00:16:22,100
If you want to like, vertically 
integrate, the chip stack. 

300
00:16:22,100 --> 00:16:25,500
And I think the geopolitical 
thing is like, well, Asia has 

301
00:16:25,500 --> 00:16:29,000
spent 20 or 30 years, building 
the whole supply chain around 

302
00:16:29,000 --> 00:16:31,500
this industry, and you're not 
just going to like, uproot the 

303
00:16:31,500 --> 00:16:33,100
whole tree. 
That's like saying that my 

304
00:16:33,100 --> 00:16:36,300
Quarry, his a rhizome of like 
this industry has been like 

305
00:16:36,600 --> 00:16:38,600
migrate in two seconds. 
I just don't think that's 

306
00:16:38,600 --> 00:16:42,200
possible. 
I think maybe like tsmc has been

307
00:16:42,300 --> 00:16:45,900
due to open a Fab in the US and 
like some time or there's some 

308
00:16:45,900 --> 00:16:48,400
kind of thing like that but like
the fat itself isn't able to 

309
00:16:48,400 --> 00:16:50,600
produce like these you know, ten
them internship. 

310
00:16:50,600 --> 00:16:52,800
I mean, I know very little about
this but I just from what I know

311
00:16:52,800 --> 00:16:56,300
it seems like a very kind of 
interesting thing that most 

312
00:16:56,300 --> 00:16:59,200
people don't realize the 
geopolitical impacts for sure 

313
00:16:59,200 --> 00:17:02,100
that's like this Olive Branch 
that was given to Trump because 

314
00:17:02,100 --> 00:17:05,000
he's like, I want to have chip 
Manufacturing in the US and it's

315
00:17:05,000 --> 00:17:07,500
like, that's not happening by 
the same thing. 

316
00:17:07,500 --> 00:17:09,700
Happened to Boeing, I know this 
is really off. 

317
00:17:09,800 --> 00:17:11,800
Off topic. 
But part of the reason we had 

318
00:17:11,800 --> 00:17:15,800
this whole 77 Fiasco, is that 
Boeing tried to decentralize its

319
00:17:15,800 --> 00:17:19,700
supply chain and then like they 
stopped having control over like

320
00:17:19,700 --> 00:17:22,599
batteries and then batteries 
exploded, whereas they used to 

321
00:17:22,599 --> 00:17:27,500
make your own batteries before. 
Back in January, we interviewed 

322
00:17:27,500 --> 00:17:29,900
Steve coconut house and Sylvia 
McCallie of Al Gore. 

323
00:17:29,900 --> 00:17:32,000
And, and during our 
conversation, we talked about 

324
00:17:32,000 --> 00:17:35,100
how algorithms unique design, 
makes it easy for developers to 

325
00:17:35,100 --> 00:17:38,100
build sophisticated applications
on their platform. 

326
00:17:38,500 --> 00:17:40,800
So what's great about Al Gore 
and Beyond the fact that it's 

327
00:17:40,800 --> 00:17:44,700
fast, it's secure its scales and
it has instant finality is the 

328
00:17:44,700 --> 00:17:46,700
fact that they've designed a 
layer one protocol with 

329
00:17:46,700 --> 00:17:49,100
Primitives that are 
purpose-built for defy. 

330
00:17:49,400 --> 00:17:51,400
So what that means is that 
they've taken some of the most 

331
00:17:51,400 --> 00:17:54,300
common things that people do 
with smart contracts and they've

332
00:17:54,300 --> 00:17:57,200
embedded them right in System 
right in the layer 1. 

333
00:17:57,400 --> 00:18:00,400
So things like issuing tokens 
Atomic transfers. 

334
00:18:00,600 --> 00:18:02,700
These are built into the layer 1
and smart. 

335
00:18:02,700 --> 00:18:05,200
Contracts are first class 
citizens on all Grant. 

336
00:18:05,500 --> 00:18:08,300
So with these essential building
blocks at your disposal, you can

337
00:18:08,300 --> 00:18:10,800
build fast and secure, defy apps
in. 

338
00:18:10,800 --> 00:18:14,300
No time to learn more about what
Al Gore and brings to the table 

339
00:18:14,300 --> 00:18:16,700
and how to get started. 
I would encourage you to check 

340
00:18:16,700 --> 00:18:20,900
out Al Grand.com / at the center
that lets them know that you 

341
00:18:20,900 --> 00:18:24,100
heard about it from us and it 
takes you, where you need to go 

342
00:18:24,100 --> 00:18:27,300
to learn about their Tech. 
And what that we'd like to thank

343
00:18:27,300 --> 00:18:29,000
algren for supporting the 
podcast. 

344
00:18:31,400 --> 00:18:33,700
I think our audience would also 
like to hear about Gauntlet and 

345
00:18:33,700 --> 00:18:36,700
what you guys are doing. 
So I guess I didn't even explain

346
00:18:36,700 --> 00:18:41,300
how Hardware gotten to crypto 
Bitcoin, miners, front-run us. 

347
00:18:41,400 --> 00:18:45,000
I started Mining and then in 
2013, I sold all my Bitcoin 

348
00:18:45,000 --> 00:18:47,600
because I was like, this is 
going to blow up. 

349
00:18:47,600 --> 00:18:49,100
This is going to be a Ponzi 
scheme. 

350
00:18:49,400 --> 00:18:54,400
Very dumb idea, obviously in 
retrospect, but I started really

351
00:18:54,400 --> 00:18:57,700
pay more attention to the papers
because we worked in distributed

352
00:18:57,700 --> 00:19:00,000
systems. 
We were building this Type of 

353
00:19:00,200 --> 00:19:02,600
when you're building these 
ethics, we we built this data 

354
00:19:02,600 --> 00:19:06,000
center to kind of run like 
millions of these machines. 

355
00:19:06,000 --> 00:19:08,800
So we had to kind of think about
this type of stuff. 

356
00:19:09,600 --> 00:19:12,600
I started really getting 
convinced that there was 

357
00:19:12,600 --> 00:19:15,400
something novel here when the 
ghost paper came out. 

358
00:19:15,400 --> 00:19:18,900
So ghost just kind of this Fork 
Choice rule that was in the 

359
00:19:18,900 --> 00:19:22,900
early versions of the theory on 
that kind of promised you that 

360
00:19:22,900 --> 00:19:26,300
you could handle like faster 
block production times. 

361
00:19:26,600 --> 00:19:30,600
If you chose a different 
Fortress Rule and ghost was one 

362
00:19:30,600 --> 00:19:34,000
of the first papers that 
thoughtfully thought about the 

363
00:19:34,200 --> 00:19:36,600
incentive design and also the 
networking. 

364
00:19:36,600 --> 00:19:40,000
And also, the sort of like basic
like architecture is like, if I 

365
00:19:40,000 --> 00:19:41,500
wrote this code, how would I 
write? 

366
00:19:41,700 --> 00:19:45,000
And that was when I was like, 
wow, there's something serious 

367
00:19:45,000 --> 00:19:47,400
here. 
It's not just like, oh, haha, 

368
00:19:47,400 --> 00:19:50,600
like a bunch of people on the 
internet like made tried to 

369
00:19:50,800 --> 00:19:53,700
usurp Leslie. 
Lamport spax us. 

370
00:19:54,100 --> 00:19:56,900
It was like, oh, there's 
actually some novel thing here. 

371
00:19:58,000 --> 00:20:01,000
So you know, I think before that
I was like very, you know, maybe

372
00:20:01,000 --> 00:20:03,800
Bitcoin maximalist, I think the 
ghost paper was one of the first

373
00:20:03,800 --> 00:20:07,700
papers that I was like oh wow, 
there's like cool ideas that the

374
00:20:07,700 --> 00:20:10,700
bitcoiners are not paying 
attention to, for sure. 

375
00:20:10,800 --> 00:20:14,100
It also made me realize like, oh
man, the design space of this 

376
00:20:14,100 --> 00:20:17,000
thing is like, bigger than 
anything that Humanity has ever 

377
00:20:17,000 --> 00:20:19,700
had. 
Like you have to like this 

378
00:20:19,700 --> 00:20:23,000
combine so many things to say a 
simple result, like that's kind 

379
00:20:23,000 --> 00:20:25,000
of insane, right? 
Like, you know, in other fields 

380
00:20:25,000 --> 00:20:26,900
you don't have to do as many 
things like that. 

381
00:20:27,300 --> 00:20:30,200
And then I worked in high 
frequency trading afterwards and

382
00:20:30,200 --> 00:20:33,400
there, we actually would do this
type of thing where we would 

383
00:20:33,400 --> 00:20:36,800
make models of our trading 
strategies and then we make 

384
00:20:36,800 --> 00:20:39,500
models of other people's trading
strategies and we'd have them. 

385
00:20:39,500 --> 00:20:43,500
Run kind of think like alphago 
style where they would like play

386
00:20:43,500 --> 00:20:46,200
against each other and you try 
to optimize your strategy and 

387
00:20:46,200 --> 00:20:48,600
that was around the time. 
I think the algorithm paper came

388
00:20:48,600 --> 00:20:51,900
out and I remember reading the 
algorithm paper being like, this

389
00:20:51,900 --> 00:20:55,200
is amazing from a cryptography 
standpoint in the sense of like 

390
00:20:55,200 --> 00:20:58,500
wow, like I you can actually 
generate It is verifiable random

391
00:20:58,500 --> 00:21:00,300
functions. 
I didn't study cryptography. 

392
00:21:00,500 --> 00:21:02,000
I had to go read the classical 
paper. 

393
00:21:02,000 --> 00:21:05,800
So the time because I didn't 
really know that existed but at 

394
00:21:05,800 --> 00:21:08,500
the same time as like, this 
seems like a little bit like a 

395
00:21:08,500 --> 00:21:11,700
derivative more than it seems 
kind of like proof-of-work, like

396
00:21:12,100 --> 00:21:14,800
a one-way function. 
Like burning energy, is a true. 

397
00:21:14,800 --> 00:21:18,300
Is like Nature's only one-way 
function that we know of like a 

398
00:21:18,308 --> 00:21:21,000
perfectly 1 Min function. 
Whereas like, in cryptography, 

399
00:21:21,000 --> 00:21:23,600
we tried you tried to, like, 
emulate that, but it's never 

400
00:21:23,600 --> 00:21:25,600
perfect. 
And so it's kind of I was a 

401
00:21:25,600 --> 00:21:28,900
little bit like surprised that 
That there's this whole proof of

402
00:21:28,900 --> 00:21:30,800
stake thing, but like people 
didn't really think about the 

403
00:21:30,800 --> 00:21:33,700
financial aspects and then 2017 
happened. 

404
00:21:33,700 --> 00:21:38,200
And then I kind of started being
like, hey, maybe this is a real 

405
00:21:38,200 --> 00:21:41,600
deal and I writing simulations 
for fun based on the type of 

406
00:21:41,600 --> 00:21:45,700
things we're doing in finance 
and then in 2018, I kind of kept

407
00:21:45,700 --> 00:21:47,700
talking to a bunch of layer 1. 
Protocols, because I was 

408
00:21:47,700 --> 00:21:51,700
curious, if anyone was doing 
this financial modeling, I quit 

409
00:21:51,700 --> 00:21:55,200
trading and then started 
Consulting for later, ones and 

410
00:21:55,200 --> 00:21:59,000
then the big badly bread. 
To like by my consultancy. 

411
00:21:59,000 --> 00:22:01,700
And that was when I was like, 
you know what, I think there is 

412
00:22:01,700 --> 00:22:06,000
enough room that there are a lot
of people who probably need 

413
00:22:06,000 --> 00:22:09,600
financial and Actuarial modeling
for the stuff and I met my 

414
00:22:09,600 --> 00:22:12,200
co-founder along the Wake. 
He also was in trading for a 

415
00:22:12,200 --> 00:22:15,500
while and then he actually 
worked on designing like 

416
00:22:15,500 --> 00:22:19,400
incentives for drivers at Uber. 
So we were both like, yeah, you 

417
00:22:19,400 --> 00:22:21,800
know, like I think there's like 
a way to make this rigorous. 

418
00:22:21,800 --> 00:22:25,300
So we started initially focusing
on proof of stake, especially 

419
00:22:25,300 --> 00:22:28,300
because I think that was the 
Genesis of Of my kind of 

420
00:22:28,300 --> 00:22:32,300
interest in really committing 
100% of my time to this. 

421
00:22:32,700 --> 00:22:34,700
And then over time, it became 
much more clear. 

422
00:22:34,700 --> 00:22:39,500
That Defy is really the place 
where there's crazy amount of 

423
00:22:39,600 --> 00:22:43,800
financial incentive modeling for
multiple agents that exist. 

424
00:22:43,900 --> 00:22:48,300
And there's just this open space
of both research as well as 

425
00:22:48,300 --> 00:22:50,600
like, actually deploying it to 
production. 

426
00:22:50,600 --> 00:22:56,500
And, you know, the 2010 to 12:00
shift from, in AI from like it's

427
00:22:56,500 --> 00:22:59,400
like half a That makes for a 
watched up from the 90s and 

428
00:22:59,400 --> 00:23:03,200
half, like people who are just 
making random stuff and like 

429
00:23:03,200 --> 00:23:06,600
calling it, like sentient. 
But we don't know if it works 

430
00:23:07,200 --> 00:23:10,500
was really this thing where like
these kind of hooligans turned 

431
00:23:10,500 --> 00:23:12,900
into, like the people who are 
correct. 

432
00:23:12,900 --> 00:23:15,900
I really feel like that's 
starting to happen right now in 

433
00:23:15,900 --> 00:23:17,700
crypto. 
That's a very long-winded 

434
00:23:17,700 --> 00:23:21,300
explanation of how I got here. 
And so Gauntlet really is about 

435
00:23:21,300 --> 00:23:25,200
taking these tools from Finance,
Actuarial modeling agent-based 

436
00:23:25,200 --> 00:23:29,700
simulation and Fooling them 
towards the kind of new problems

437
00:23:29,700 --> 00:23:34,200
in incentive design and Krypton.
Could you give us an example of 

438
00:23:34,200 --> 00:23:38,000
like so water is the sort of 
things that you would model. 

439
00:23:38,000 --> 00:23:42,900
Like, let's say I came to you 
with a new proof of stake 

440
00:23:43,000 --> 00:23:45,500
consensus protocol. 
Are you modeling? 

441
00:23:45,500 --> 00:23:48,400
Like are you testing the safety 
and liveness of my consensus 

442
00:23:48,400 --> 00:23:51,900
protocol, is it at some higher 
layer than that? 

443
00:23:51,900 --> 00:23:53,700
Like, what specifically are you 
testing? 

444
00:23:55,000 --> 00:23:57,700
I think it starts in a bunch of 
different levels. 

445
00:23:57,700 --> 00:24:01,000
I think that's certainly the 
first level, one of the things. 

446
00:24:01,000 --> 00:24:04,600
I remember that tip me off when 
I was first reading about proof 

447
00:24:04,600 --> 00:24:06,800
sake. 
Was this idea that there were 

448
00:24:06,800 --> 00:24:09,200
many different synchrony 
assumptions and all of the 

449
00:24:09,200 --> 00:24:12,100
different papers but they were 
quite in equivalent. 

450
00:24:12,200 --> 00:24:17,600
So some people would say you're 
live, if you eventually were 

451
00:24:17,600 --> 00:24:21,700
able to process the block, some 
people would say you're live if 

452
00:24:21,700 --> 00:24:24,700
greater than x percent of nodes 
agreed. 

453
00:24:24,900 --> 00:24:28,400
At a block have been produced. 
And some people would say you're

454
00:24:28,400 --> 00:24:30,600
alive. 
If and this is sort of the way 

455
00:24:30,600 --> 00:24:35,400
the Avalanche paper kind of 
eventually go proved if like in 

456
00:24:35,400 --> 00:24:39,500
the limit of an infinite number 
of blocks of nonzero, fraction 

457
00:24:39,500 --> 00:24:43,700
of them were actually reached by
large percentage of the note. 

458
00:24:43,900 --> 00:24:46,900
Now those all kind of sound 
equivalent but mathematically 

459
00:24:46,900 --> 00:24:48,500
when you're trying to write 
these purse they're not. 

460
00:24:48,700 --> 00:24:51,400
So the types of things I was 
really first interested in 

461
00:24:51,400 --> 00:24:54,700
simulating where things like how
long does it actually take? 

462
00:24:54,800 --> 00:24:58,400
Take on different network, 
topologies for these blocks, to 

463
00:24:58,408 --> 00:25:01,700
actually have disseminated 
enough such that, the network 

464
00:25:01,700 --> 00:25:04,900
reaches consensus. 
And one of the things I was very

465
00:25:05,000 --> 00:25:07,600
realized, you could only kind of
answer by simulation and would 

466
00:25:07,600 --> 00:25:11,200
be very hard to prove is given 
the network topology? 

467
00:25:11,500 --> 00:25:14,300
What is the true? 
Partial synchrony constant? 

468
00:25:14,400 --> 00:25:17,200
And what I mean by that is, 
like, what's the constant at? 

469
00:25:17,200 --> 00:25:20,600
Which, if everyone receives all 
of the blocks within a certain 

470
00:25:20,600 --> 00:25:22,500
time window? 
How long does that time? 

471
00:25:22,500 --> 00:25:26,100
Window have to be for the 
network to Achieve liveness and 

472
00:25:26,100 --> 00:25:28,700
safety. 
And so stimulating that on 

473
00:25:28,700 --> 00:25:32,300
different network topologies 
actually convinced me that even 

474
00:25:32,300 --> 00:25:34,400
Bitcoin has a lot of problems if
the network. 

475
00:25:34,400 --> 00:25:36,800
Topology is like two 
disconnected and so 

476
00:25:36,800 --> 00:25:40,000
mathematicians have sort of ways
of defining what it means to be,

477
00:25:40,000 --> 00:25:43,500
too disconnected for without 
getting into too much detail. 

478
00:25:43,500 --> 00:25:46,800
I think like the spectral gap of
the graph is something that 

479
00:25:46,800 --> 00:25:49,100
measures. 
How long random walks take? 

480
00:25:49,200 --> 00:25:52,600
And so the idea is like if 
someone who's randomly walking 

481
00:25:52,600 --> 00:25:56,100
on your network topology, Gets 
lost because they're too drunk 

482
00:25:56,100 --> 00:25:59,500
than your block, may never reach
everyone. 

483
00:25:59,700 --> 00:26:03,800
And so you kind of like assume 
like hey I put a drunk person on

484
00:26:03,800 --> 00:26:07,000
the network graph and I see how 
long it takes them to reach 

485
00:26:07,000 --> 00:26:09,700
everyone. 
That's kind of the this model of

486
00:26:09,700 --> 00:26:12,800
like time that mathematicians 
use that, I was trying to map 

487
00:26:12,800 --> 00:26:15,800
the model that people have 
formal proofs that land to to 

488
00:26:15,800 --> 00:26:18,700
like what could Prague refers 
and distributed systems people 

489
00:26:18,700 --> 00:26:20,800
were using and simulation was 
the tool for that. 

490
00:26:20,808 --> 00:26:24,500
So we start by assessing kind of
some of these types of She's I 

491
00:26:24,508 --> 00:26:27,600
think safety is not the type of 
thing we assess. 

492
00:26:27,600 --> 00:26:30,600
I think safety is a purely 
cryptographic property but 

493
00:26:30,600 --> 00:26:33,800
liveness of proof of stake, 
protocols is very much a sort of

494
00:26:33,800 --> 00:26:36,100
statistical property. 
It depends on the network 

495
00:26:36,100 --> 00:26:38,000
topology. 
It depends on the Layton sees 

496
00:26:38,600 --> 00:26:41,600
how random they are. 
What the 95th percentile is the 

497
00:26:41,600 --> 00:26:44,900
latency is look like cetera. 
I'll use example of ten, 

498
00:26:44,900 --> 00:26:46,800
America's obviously. 
That's what I'm most familiar 

499
00:26:46,800 --> 00:26:48,300
with. 
You know, we also have this 

500
00:26:48,300 --> 00:26:51,700
whole live in partial synchrony 
how it works is. 

501
00:26:51,700 --> 00:26:55,700
We basically have these round 
Round and each round, we say 

502
00:26:55,700 --> 00:26:59,500
there's a timeout which nodes 
Will Wait currently on most kind

503
00:26:59,500 --> 00:27:01,600
of network that one second by 
default. 

504
00:27:01,600 --> 00:27:05,700
But then if we don't reach 
consensus in that one, second we

505
00:27:05,700 --> 00:27:09,100
go to the next round and we 
increase it by the time out by a

506
00:27:09,100 --> 00:27:11,300
certain amount of. 
So I think we increase it by a 

507
00:27:11,300 --> 00:27:14,500
quarter of a second every time. 
So we do a one second timeout' 

508
00:27:14,500 --> 00:27:18,000
than if that round doesn't work.
We go to a 1.25 second time out.

509
00:27:18,000 --> 00:27:19,800
There we go. 
To 1 Point 5, Second timeout. 

510
00:27:19,900 --> 00:27:23,400
And so these numbers for us we 
just pulled these numbers out of

511
00:27:23,400 --> 00:27:25,700
a hat, you know? 
We did a little bit of testing, 

512
00:27:25,700 --> 00:27:29,000
but if I would select Enderman, 
I would go to you and basically 

513
00:27:29,000 --> 00:27:32,000
say like, hey, help us figure 
out the right numbers. 

514
00:27:32,000 --> 00:27:34,500
We should be putting here 
because if one second is too 

515
00:27:34,500 --> 00:27:38,000
long, then we're wasting time 
that we could be making faster 

516
00:27:38,000 --> 00:27:39,900
blocks. 
Meanwhile, if it's too short, 

517
00:27:40,000 --> 00:27:42,500
we're causing unnecessary rounds
for no reason. 

518
00:27:42,500 --> 00:27:45,400
And so you would basically be 
able to help us parameterize 

519
00:27:45,400 --> 00:27:48,000
that correctly. 
Exactly. 

520
00:27:48,000 --> 00:27:50,800
Yeah, it's like a band with 
firstly in see trade-off of like

521
00:27:50,900 --> 00:27:52,700
how much communication to have 
to do. 

522
00:27:52,800 --> 00:27:56,000
There's an expected number of 
rounds and the distribution of 

523
00:27:56,000 --> 00:27:57,900
the number of rounds. 
In the thing you're talking 

524
00:27:57,900 --> 00:28:01,400
about, imagine you have a 
hundred million blocks, there 

525
00:28:01,400 --> 00:28:03,800
are produced. 
And for each block, I looked at 

526
00:28:03,900 --> 00:28:06,500
the number of rounds, it took 
before the network agrees. 

527
00:28:06,500 --> 00:28:09,700
And I look at that distribution.
Now that distribution is a 

528
00:28:09,700 --> 00:28:13,600
function of these parameters, 
you chose, but the problem is 

529
00:28:13,600 --> 00:28:15,700
that distribution. 
Also depends on the network 

530
00:28:15,700 --> 00:28:18,400
topology. 
It depends on Some lower level 

531
00:28:18,400 --> 00:28:21,400
details of protocol. 
And so, yeah, the type of thing 

532
00:28:21,400 --> 00:28:24,600
we need stress, stress test is 
like, how does that work under 

533
00:28:24,600 --> 00:28:27,500
different models of users? 
Because you can have different 

534
00:28:27,500 --> 00:28:31,800
types of users who effect that 
behavior one type of user might 

535
00:28:31,800 --> 00:28:35,100
be the type of user that drops a
lot of packets because their 

536
00:28:35,200 --> 00:28:38,000
computer goes off a lot. 
They don't care about getting / 

537
00:28:38,000 --> 00:28:39,700
because they don't even know 
they're getting slashed for 

538
00:28:39,700 --> 00:28:43,300
being offline or something. 
Another type of user might be 

539
00:28:43,300 --> 00:28:44,600
one. 
That's malicious, who's 

540
00:28:44,600 --> 00:28:47,000
purposely, forwarding bad 
packets? 

541
00:28:47,200 --> 00:28:51,200
Another type of user might be 
one who is kind of a big block 

542
00:28:51,200 --> 00:28:54,300
producer and like just is like 
trying to get not even just 

543
00:28:54,300 --> 00:28:57,000
honest but it's hyper-rational 
and that they're just trying to 

544
00:28:57,000 --> 00:29:01,300
like flood the network so that 
their block always is first what

545
00:29:01,300 --> 00:29:02,800
the different composition of 
users. 

546
00:29:02,800 --> 00:29:05,600
Also affects this distribution 
of like the expected number of 

547
00:29:05,600 --> 00:29:07,300
rounds. 
It takes and that's kind of 

548
00:29:07,308 --> 00:29:09,500
where we model when it comes to 
per sake. 

549
00:29:09,500 --> 00:29:13,200
But we also model over time. 
We realized that we started with

550
00:29:13,200 --> 00:29:15,700
these networking models, because
that's what people in high 

551
00:29:15,700 --> 00:29:19,400
frequency trading do A lot in 
high frequency trading you 

552
00:29:19,400 --> 00:29:21,400
model. 
Like here's exchange one, hears,

553
00:29:21,400 --> 00:29:23,300
exchange to here's a change 
three. 

554
00:29:23,500 --> 00:29:24,900
Here's all the routers and 
exchange. 

555
00:29:24,900 --> 00:29:28,000
One takes change to. 
If I send a packet, how long 

556
00:29:28,000 --> 00:29:30,300
does it take? 
And if, you know, you kind of 

557
00:29:30,300 --> 00:29:33,100
model the topology and sort of a
similar way you would think 

558
00:29:33,100 --> 00:29:35,600
about modeling validators. 
And then you say, hey, if I have

559
00:29:35,600 --> 00:29:39,100
an adversary who's also thinking
the same way as me are they also

560
00:29:39,100 --> 00:29:41,900
sending the same number of 
packets and will it caught will 

561
00:29:41,900 --> 00:29:45,100
who will reach first. 
It's a similar type of pump. 

562
00:29:46,200 --> 00:29:48,600
It sounds like you're doing 
analysis at like different 

563
00:29:48,600 --> 00:29:52,100
layers of the stack. 
You're doing the mechanism 

564
00:29:52,100 --> 00:29:56,600
analysis of the system's 
themselves in order to looking 

565
00:29:56,600 --> 00:29:59,200
for liveness and availability 
and things like that. 

566
00:29:59,200 --> 00:30:02,900
So this is like the mechanism 
design part and this might take 

567
00:30:02,900 --> 00:30:07,400
place when the team is building 
the system, but you're also 

568
00:30:07,400 --> 00:30:11,300
doing research and Analysis and 
simulations at a higher level up

569
00:30:11,300 --> 00:30:12,500
the stack. 
So I know you like you're also 

570
00:30:12,500 --> 00:30:16,500
doing, say research on Market 
participants Like in the 

571
00:30:16,500 --> 00:30:18,300
compound protocol. 
So this is happening at the 

572
00:30:18,300 --> 00:30:20,100
economic layer at the market 
layer. 

573
00:30:20,200 --> 00:30:22,500
Is that right? 
Yeah. 

574
00:30:22,500 --> 00:30:26,300
So I like to think of when you 
do simulation, I think one of 

575
00:30:26,300 --> 00:30:29,800
the reasons people often times 
think like hey this can never be

576
00:30:29,800 --> 00:30:32,700
real or it doesn't replicate 
reality or how do you know it. 

577
00:30:32,700 --> 00:30:35,800
Replicates reality. 
Is that a lot of people try to 

578
00:30:35,800 --> 00:30:38,900
simulate everything all at once 
and you really need to think of 

579
00:30:38,900 --> 00:30:41,800
it like an onion. 
We're here is a particular 

580
00:30:41,800 --> 00:30:46,000
problem that I'm trying to solve
and here is the particular 

581
00:30:46,000 --> 00:30:48,400
instance of it. 
And here, it kind of the bounds 

582
00:30:48,400 --> 00:30:50,100
of like the worst case when best
case. 

583
00:30:50,200 --> 00:30:53,300
And I'm going to try to That in 
isolation. 

584
00:30:53,800 --> 00:30:56,600
And then I add the next layer of
the onion and I have it interact

585
00:30:56,600 --> 00:30:58,900
with that layer. 
And then I add the next layer of

586
00:30:58,900 --> 00:31:00,900
the onion, and I haven't 
interact with out there. 

587
00:31:01,300 --> 00:31:03,900
I think, if you do it, kind of 
in this incremental way, you can

588
00:31:03,900 --> 00:31:06,900
actually try to reason about the
whole complex system. 

589
00:31:07,200 --> 00:31:10,500
You know, we start with things 
like this layer 1, liveness type

590
00:31:10,500 --> 00:31:13,600
of stuff, but you slowly build 
up to the economic incentives. 

591
00:31:15,000 --> 00:31:18,400
So, how much that complexity 
gets injected into that. 

592
00:31:18,400 --> 00:31:22,200
Once you start thinking of 
things like interoperability 

593
00:31:22,200 --> 00:31:25,300
between blockchains for 
instance, the problem gets 

594
00:31:25,300 --> 00:31:28,300
exponentially more difficult. 
If you start factoring in 

595
00:31:28,400 --> 00:31:30,700
multiple block, chains and 
interactions between all these 

596
00:31:30,700 --> 00:31:33,000
different systems that doesn't 
fit in my brain space. 

597
00:31:34,100 --> 00:31:35,800
It is certainly exponentially 
bigger. 

598
00:31:35,800 --> 00:31:38,100
I mean the you're taking address
space. 

599
00:31:38,100 --> 00:31:41,100
One address space to you've 
doubled the number of bits 

600
00:31:41,300 --> 00:31:45,200
definitely increasing in an 
exponential manner but the Is to

601
00:31:45,200 --> 00:31:49,700
try to like isolate the points 
of complexity that are most 

602
00:31:49,700 --> 00:31:53,300
tangible to think about how 
humans would interact with these

603
00:31:53,300 --> 00:31:57,600
systems because fundamentally 
okay, I went from 128 bits of 

604
00:31:57,600 --> 00:32:02,400
entropy to 256, bits of entropy 
for a 2 Chain interactive system

605
00:32:02,600 --> 00:32:04,600
but humans are still using 
those, right? 

606
00:32:04,600 --> 00:32:08,200
And like methods and interfaces 
that you provide to the human as

607
00:32:08,200 --> 00:32:12,600
a developer also, dictate what 
usage, you're going to get and 

608
00:32:12,600 --> 00:32:16,500
so you try to model things, That
replicate what humans who were 

609
00:32:16,500 --> 00:32:18,500
using those interfaces would 
look like. 

610
00:32:18,800 --> 00:32:22,400
And then you kind of say, okay, 
let's say I have 10 different 

611
00:32:22,400 --> 00:32:26,400
versions of the same human, how 
they use the system, 100 

612
00:32:26,400 --> 00:32:27,900
different versions of the same 
human. 

613
00:32:27,900 --> 00:32:31,600
How would they use a system? 
And you kind of build some sort 

614
00:32:31,600 --> 00:32:35,200
of a bottoms-up approach where 
you try to like identify 

615
00:32:35,200 --> 00:32:39,600
behaviors, figure out which of 
those behaviors are consistent 

616
00:32:39,600 --> 00:32:42,900
among group of people then 
figure out what math describes 

617
00:32:42,900 --> 00:32:45,200
them. 
Like what their utility As what 

618
00:32:45,200 --> 00:32:48,300
value, they're getting out of 
calling this function cross 

619
00:32:48,300 --> 00:32:50,600
Block Chain Transaction. 
What value? 

620
00:32:50,600 --> 00:32:53,000
They're getting out of. 
Hey I'm willing to pay a 

621
00:32:53,000 --> 00:32:56,300
transaction fee that's higher 
than the one on my chain to move

622
00:32:56,300 --> 00:32:59,300
across chain. 
Then after that, what decisions 

623
00:32:59,300 --> 00:33:03,000
they make like given this sort 
of notion of how they can value 

624
00:33:03,300 --> 00:33:07,300
a cross Block Chain Transaction.
What actions can they take one? 

625
00:33:07,300 --> 00:33:10,600
Action is certainly makes class 
blockchain transaction. 

626
00:33:10,600 --> 00:33:14,400
Another one is don't another one
is, is there a way for me to do 

627
00:33:14,400 --> 00:33:17,400
it? 
It on my current blockchain that

628
00:33:17,400 --> 00:33:21,500
gives me 80 percent of the same 
value or 70 percent or 50 

629
00:33:21,500 --> 00:33:24,500
percent breaking it down in kind
of this. 

630
00:33:24,600 --> 00:33:28,100
Hey, there's still a human using
this thing or there's a human 

631
00:33:28,100 --> 00:33:30,300
writing a bot, that's using this
thing. 

632
00:33:30,500 --> 00:33:34,900
There's still this notion of 
like peoples, ux habits are not 

633
00:33:34,900 --> 00:33:37,700
uniformly, random right there, 
not just like a father. 

634
00:33:37,800 --> 00:33:41,300
They're really kind of like 
using these interfaces in a very

635
00:33:41,300 --> 00:33:45,100
concrete way and reasoning about
how different Void. 

636
00:33:45,100 --> 00:33:48,700
User is really how you Mom kind 
of try to start modeling these 

637
00:33:48,700 --> 00:33:52,100
types of things. 
In consensus protocols. 

638
00:33:52,100 --> 00:33:54,600
We usually like think of it like
okay. 

639
00:33:54,700 --> 00:33:57,300
The three types of reactors. 
We have our like Byzantine 

640
00:33:57,600 --> 00:34:01,000
rational and altruistic. 
But so what you're essentially 

641
00:34:01,000 --> 00:34:04,900
implying is that this is like 
way too simplistic and that we 

642
00:34:04,900 --> 00:34:08,500
need to be much more specific. 
It's not just these three 

643
00:34:08,500 --> 00:34:10,699
categories. 
It's way more of a spectrum of 

644
00:34:11,000 --> 00:34:14,000
many more types of users are 
actors. 

645
00:34:14,300 --> 00:34:18,100
So how do you know you've 
modeled all the actors possible 

646
00:34:18,100 --> 00:34:20,400
or like how do you know you 
cover the entire space? 

647
00:34:20,600 --> 00:34:25,100
Extend, how do you like deal 
with like things that were just?

648
00:34:25,100 --> 00:34:27,500
You couldn't predict like 
imagine you wanted to try to 

649
00:34:27,500 --> 00:34:31,699
predict like the distribution of
SN X and how much what it would 

650
00:34:31,699 --> 00:34:33,800
be collateralizing. 
But like, you know what world 

651
00:34:33,800 --> 00:34:36,800
could you have predicted that 
like a million snx would be 

652
00:34:36,800 --> 00:34:39,800
sitting here. 
Farming, yams craziness. 

653
00:34:39,800 --> 00:34:43,199
How can you possibly build all 
of these into your models? 

654
00:34:44,199 --> 00:34:47,800
For sure. 
So I think one thing to remember

655
00:34:47,800 --> 00:34:51,400
from consensus protocols, is 
this bar model is Byzantine 

656
00:34:51,400 --> 00:34:56,300
altruistic rational is very 
unfair in one way, which is that

657
00:34:56,300 --> 00:34:59,800
Byzantine than altruistic are 
like one dimensional thing. 

658
00:34:59,800 --> 00:35:03,200
So in the space of all possible,
strategies, if I represent a 

659
00:35:03,207 --> 00:35:07,500
strategy that a user takes given
an interface, so an interface. 

660
00:35:08,400 --> 00:35:10,800
Let's just say it's a set of 
functions that you can call. 

661
00:35:11,500 --> 00:35:15,100
And let's assume that all users 
have Valuation model. 

662
00:35:15,300 --> 00:35:17,500
And based on the valuation model
evaluation, all the means of 

663
00:35:17,500 --> 00:35:20,500
utility function and in 
traditional economics or kind of

664
00:35:20,500 --> 00:35:22,200
objective function in machine 
learning. 

665
00:35:23,300 --> 00:35:25,900
You have objective function and 
you have decision function. 

666
00:35:25,900 --> 00:35:28,800
So the objective function gives 
you a single number or some so 

667
00:35:28,800 --> 00:35:31,100
numbers, the decision function 
takes those numbers and gives 

668
00:35:31,100 --> 00:35:33,400
you an action like what the user
does. 

669
00:35:33,400 --> 00:35:36,900
And there's an entire you do not
to to drills. 

670
00:35:36,900 --> 00:35:39,900
But, you know, if you read 
philosophy, there's a whole 

671
00:35:39,900 --> 00:35:42,000
argument of whether this is how 
humans acted. 

672
00:35:42,000 --> 00:35:45,700
But ignoring the content type of
stuff. 

673
00:35:46,400 --> 00:35:50,300
That's how almost all models in 
machine learning for like 

674
00:35:50,500 --> 00:35:52,500
alphago and stuff. 
Think about the world, right? 

675
00:35:52,500 --> 00:35:55,200
There's this Kind of value 
function, decision function. 

676
00:35:56,400 --> 00:36:01,000
Now, Byzantine the value 
function is choose a random 

677
00:36:01,000 --> 00:36:04,000
number choose management 
altruistic, the random number, 

678
00:36:04,000 --> 00:36:07,000
it's like choose the same number
and choose this particular 

679
00:36:07,000 --> 00:36:10,000
Behavior. 
Always rational is much more 

680
00:36:10,000 --> 00:36:14,600
like I'm actually observing the 
environment and and like trying 

681
00:36:14,600 --> 00:36:18,300
to figure out a valuation and a 
value of these actions and 

682
00:36:18,300 --> 00:36:21,500
changing over time. 
Fundamentally Byzantine and 

683
00:36:21,500 --> 00:36:23,900
altruistic are actually very 
one-dimensional that way. 

684
00:36:24,000 --> 00:36:26,100
One is purely random and one is 
purely deterministic. 

685
00:36:26,200 --> 00:36:30,300
Rustic, but rational is actually
this adaptive type of adversary.

686
00:36:30,800 --> 00:36:33,900
And it's an infinite dimensional
space of functions, like the set

687
00:36:33,900 --> 00:36:39,100
of functions that can give you 
utilities and values is infinite

688
00:36:39,100 --> 00:36:42,100
dimensional for rational, but 
the other ones are actually Zero

689
00:36:42,100 --> 00:36:44,600
Dimensional like one dimension 
like, you know, there's like a 

690
00:36:44,600 --> 00:36:46,900
single function that everyone 
knows in advance. 

691
00:36:47,800 --> 00:36:50,900
And so the problem is by saying,
Byzantine altruistic rational 

692
00:36:51,200 --> 00:36:54,300
your kind of assuming, hey, 
there are three equal categories

693
00:36:54,700 --> 00:36:57,700
but that's not true because The 
rational category from a 

694
00:36:57,700 --> 00:37:02,100
mathematical perspective is 
significantly larger. 

695
00:37:02,200 --> 00:37:05,700
Like, infinitely larger, and not
just countably infinite larger, 

696
00:37:05,700 --> 00:37:07,400
it's uncountably infinite we 
logically. 

697
00:37:07,400 --> 00:37:11,000
I think there's a lot of kind of
classical functional analysis, 

698
00:37:11,000 --> 00:37:14,200
theorems, that approve, this, 
I'd say, Nash won. 

699
00:37:14,400 --> 00:37:17,300
His Nobel Prize was kind of 
related to proving this. 

700
00:37:18,000 --> 00:37:22,100
But the point is that, like you 
can't actually, you know, when 

701
00:37:22,100 --> 00:37:24,800
you, when you say hey I've 
analyzed this protocol under 

702
00:37:24,800 --> 00:37:30,200
Byzantine altruistic, Floor 
99.99999 percent of people 

703
00:37:30,200 --> 00:37:32,900
who've done that are like hey 
we're just going to say 

704
00:37:32,900 --> 00:37:38,000
rationale is optimizing quasi 
linear utility like Keith Keith 

705
00:37:38,400 --> 00:37:42,700
or hey. 
Rationale is like they they are 

706
00:37:42,700 --> 00:37:45,500
like only caring about a certain
type of thing. 

707
00:37:45,800 --> 00:37:48,600
And the problem is no matter how
you reason about any of these 

708
00:37:48,600 --> 00:37:52,800
systems, you are fundamentally 
imbuing, a notion of what you 

709
00:37:52,800 --> 00:37:57,100
think rational is. 
And you can never perfectly 

710
00:37:57,100 --> 00:38:00,600
simulate these things. 
I will be the first person, and 

711
00:38:00,600 --> 00:38:03,400
the last person to tell you 
that, but I do think you should 

712
00:38:03,400 --> 00:38:07,300
be fully ating the set of 
rational actors with a much 

713
00:38:07,300 --> 00:38:11,400
broader set of views. 
Then what you can do with formal

714
00:38:11,400 --> 00:38:16,200
proofs, I think informal proof. 
The problem is really like it's 

715
00:38:16,200 --> 00:38:20,300
very hard to prescribe a model 
that's you know kind of can 

716
00:38:20,400 --> 00:38:22,300
cover this infinite dimensional 
space. 

717
00:38:23,100 --> 00:38:25,800
And I think cryptographers have 
this willingness to suspend 

718
00:38:25,800 --> 00:38:28,300
disbelief of like, hey, we're 
just going to pretend that the 

719
00:38:28,300 --> 00:38:32,000
rational actor, only does one 
type of rational action, but 

720
00:38:32,000 --> 00:38:35,600
that's just not true, right? 
And Game Theory and algorithm 

721
00:38:35,600 --> 00:38:39,000
that game theory and stuff, have
tons of examples of this 

722
00:38:39,000 --> 00:38:41,800
happening in practice. 
So I think the better better 

723
00:38:41,800 --> 00:38:43,600
answer. 
And, you know, I think this is 

724
00:38:43,600 --> 00:38:46,900
what the biggest algorithmic 
game theory systems that are in 

725
00:38:46,900 --> 00:38:52,500
production like Google ads and 
Facebook do Is they do tons of 

726
00:38:52,500 --> 00:38:55,200
numerical stress, testing of 
different types of users, trying

727
00:38:55,200 --> 00:38:58,600
to commit fraud different types 
of users trying to do XYZ type 

728
00:38:58,600 --> 00:39:00,000
of action. 
And then you run these 

729
00:39:00,000 --> 00:39:03,500
simulations and say like okay 
this parameter for our auction 

730
00:39:03,500 --> 00:39:06,700
is correct or like that's what 
we're going to use tomorrow and 

731
00:39:06,700 --> 00:39:09,400
you keep updating it as you get 
new data. 

732
00:39:09,400 --> 00:39:14,500
So the yam thing happens, okay? 
Snx whales are suddenly into 

733
00:39:14,700 --> 00:39:19,600
into vegetables, like didn't 
predict that, but now that we 

734
00:39:19,600 --> 00:39:22,900
add it to our little That's a 
new type of rational agent. 

735
00:39:22,900 --> 00:39:26,900
And so, then the next time. 
So, the day after the yams 

736
00:39:26,900 --> 00:39:30,100
happen, we can say. 
Hey, look, here is the thing 

737
00:39:30,100 --> 00:39:33,700
that actually causes this crazy 
amount of risk to your system 

738
00:39:33,700 --> 00:39:35,800
because people who are have us 
an X. 

739
00:39:36,300 --> 00:39:39,400
They're already leverage, they 
print printed some s USD and 

740
00:39:39,400 --> 00:39:42,300
then they took their SEO, see 
bought more snx and put in yams.

741
00:39:42,800 --> 00:39:45,600
Now your system, even though you
say it's a 700% 

742
00:39:45,600 --> 00:39:49,600
collateralization ratio, it's 
actually much lower because 

743
00:39:49,600 --> 00:39:53,200
people have been doing this 
chuckling and kind of like sort 

744
00:39:53,200 --> 00:39:57,500
of weird weird sort of financial
engineering that they might not 

745
00:39:57,500 --> 00:40:01,100
even realize they're doing and 
now we have a strategy that 

746
00:40:01,100 --> 00:40:03,200
replicates that. 
So now when someone else wants 

747
00:40:03,200 --> 00:40:06,600
let's say yams add atoms, I 
don't know. 

748
00:40:06,600 --> 00:40:09,500
Let's pretend there's like a 
synthetic atom on ethos that you

749
00:40:09,500 --> 00:40:11,300
can deposit. 
Yeah. 

750
00:40:11,900 --> 00:40:14,400
We can run the same strategy and
say like this is the amount of 

751
00:40:14,400 --> 00:40:15,800
risk, the atom holders are 
taking. 

752
00:40:16,100 --> 00:40:20,000
And so my point is, it's an 
incremental thing where you're 

753
00:40:20,000 --> 00:40:23,400
not going to predict But you're 
going to try to make your best 

754
00:40:23,400 --> 00:40:28,000
guess by building, the biggest 
library of possible things and 

755
00:40:28,000 --> 00:40:31,000
then stress testing against it. 
It's a lot like security 

756
00:40:31,000 --> 00:40:34,100
auditing where you say, either 
there exists, formal 

757
00:40:34,100 --> 00:40:37,200
verification, is this dream? 
That will predict everything or 

758
00:40:37,600 --> 00:40:39,900
here are the set of things. 
I know that could happen and I'm

759
00:40:39,900 --> 00:40:43,700
going to try to carefully look 
through each line of Vicodin and

760
00:40:43,700 --> 00:40:45,600
say like this might happen, this
might happen. 

761
00:40:45,600 --> 00:40:48,000
It's much closer to them that 
make sense. 

762
00:40:48,800 --> 00:40:52,600
And so how does like historical?
All data play a role or do this.

763
00:40:52,600 --> 00:40:55,400
Do you like when you have the 
simulation? 

764
00:40:55,600 --> 00:41:01,100
Do you like run it against 
historical data and like modify 

765
00:41:01,100 --> 00:41:05,600
your simulation models until 
they fit the historical data and

766
00:41:05,600 --> 00:41:09,100
then you start to use them to 
predict or yeah. 

767
00:41:09,300 --> 00:41:11,300
How does that work? 
Yeah. 

768
00:41:11,300 --> 00:41:14,700
Not another kind of 
philosophical dichotomy that 

769
00:41:14,700 --> 00:41:17,900
exists in the Trap. 
Traditional world is the 

770
00:41:17,900 --> 00:41:21,500
difference between the financial
world and the World. 

771
00:41:22,100 --> 00:41:25,400
So in the financial world, 
people really care about what 

772
00:41:25,400 --> 00:41:28,000
are called Point estimates. 
So Point estimates are what your

773
00:41:28,000 --> 00:41:32,800
neural net does, they give you 
an answer, they say like, hey, 

774
00:41:32,800 --> 00:41:35,800
here's a function here, A bunch 
of examples, train, it all those

775
00:41:35,800 --> 00:41:39,000
examples so that it gets the 
right answer and then in the 

776
00:41:39,000 --> 00:41:42,500
future, take a new example and 
give me a guess of like, what 

777
00:41:42,500 --> 00:41:45,100
the output is. 
It doesn't give you any estimate

778
00:41:45,100 --> 00:41:48,100
of what the uncertainty is. 
It doesn't give you any estimate

779
00:41:48,100 --> 00:41:50,700
of like what, how wrong can this
function? 

780
00:41:50,900 --> 00:41:54,300
That's mapping things, be neural
Nets, don't give you that, but 

781
00:41:54,300 --> 00:41:56,700
like sort of more traffic 
officials to just go methods, do

782
00:41:56,700 --> 00:41:58,300
that. 
And so in finance people care 

783
00:41:58,300 --> 00:42:00,600
about Point estimates because 
they're like, I want to maximize

784
00:42:00,600 --> 00:42:04,100
my expected reward or economics 
microeconomics in general. 

785
00:42:04,800 --> 00:42:08,500
In Actuarial studies. 
People care about like, hey, I 

786
00:42:08,500 --> 00:42:10,600
have this life table for this 
insurance. 

787
00:42:10,600 --> 00:42:14,400
I'm underwriting. 
And I care about the variance of

788
00:42:14,400 --> 00:42:17,900
how much I have to pay or like, 
hey, like yeah, sure. 

789
00:42:18,300 --> 00:42:22,600
On average, I only have two Like
two hundred dollars in premiums 

790
00:42:22,600 --> 00:42:26,300
from each person but like 
there's this one dude who has 

791
00:42:26,300 --> 00:42:30,200
asbestos poisoning and heat. 
It's going to cost us a billion 

792
00:42:30,200 --> 00:42:32,400
dollars to cover his health 
insurance. 

793
00:42:32,500 --> 00:42:36,500
I'm making something ingredients
but so in insurance and 

794
00:42:36,500 --> 00:42:38,800
Actuarial studies. 
You care about this kind of like

795
00:42:38,800 --> 00:42:41,600
distributional effect. 
Whereas in finance, you care 

796
00:42:41,600 --> 00:42:45,100
about like the end. 
And you know I would say machine

797
00:42:45,100 --> 00:42:47,900
learning General you care about 
kind of like predicting the 

798
00:42:47,900 --> 00:42:51,000
average Although in finance, you
care about the tail events 

799
00:42:51,000 --> 00:42:54,000
blowing you out, but there's 
kind of this dichotomy. 

800
00:42:54,300 --> 00:42:58,300
And so one way we do this is we 
fit some of the rational actors 

801
00:42:58,300 --> 00:43:00,600
behaviors based on historical 
data. 

802
00:43:00,600 --> 00:43:03,400
So we tried to take the 
historical data and say, hey 

803
00:43:03,500 --> 00:43:07,900
these actions were done by this 
address repeatedly, you know, 

804
00:43:07,900 --> 00:43:09,100
let's say your tag these 
address. 

805
00:43:09,100 --> 00:43:12,100
It wasn't like this is a, you 
know, Dex Trader that does this 

806
00:43:12,100 --> 00:43:15,300
type of action. 
So can we try to infer their 

807
00:43:15,300 --> 00:43:18,100
utility function in further 
utility function? 

808
00:43:18,100 --> 00:43:22,000
Now, that's one of the Libraries
of the one type of user who's 

809
00:43:22,000 --> 00:43:24,900
fit to Circle data. 
Another type of user is one 

810
00:43:24,900 --> 00:43:27,600
where we leave. 
We say hey this user we're going

811
00:43:27,600 --> 00:43:30,300
to parameterize in this way, 
we're going to say hey they have

812
00:43:30,300 --> 00:43:32,800
a value function but we're not 
going to say it's precisely 

813
00:43:32,800 --> 00:43:37,400
these numbers and then we sample
all the numbers we're like hey 

814
00:43:37,600 --> 00:43:41,100
there that we have a parameter 
that says how risky they are and

815
00:43:41,100 --> 00:43:43,700
when it's one there are get a 
complete Gambler. 

816
00:43:43,700 --> 00:43:46,300
They just pull the slot machine 
every time and when it's zero 

817
00:43:46,300 --> 00:43:49,100
it's they're very risk-averse. 
And then we have another 

818
00:43:49,100 --> 00:43:53,600
parameter that says, hey how 
much do they value high growth 

819
00:43:54,100 --> 00:43:55,400
versus? 
How much do they value? 

820
00:43:55,400 --> 00:43:59,100
Kind of like safe growth. 
So like they they're like oh 

821
00:43:59,100 --> 00:44:03,000
like I'm willing to invest in 
the S&P 500 vs. 

822
00:44:03,000 --> 00:44:07,000
Like I'm willing to put all my 
money into Nicola or I don't 

823
00:44:07,000 --> 00:44:09,600
know I don't know what the hot 
like now that thanks to Robin 

824
00:44:09,600 --> 00:44:12,900
Hood, I don't even know what the
like hot stock thing. 

825
00:44:12,900 --> 00:44:16,100
Like, you know, Portnoy stock 
thing is anymore. 

826
00:44:17,000 --> 00:44:20,300
But the idea is you try to say, 
hey, here's how we parameterize.

827
00:44:20,300 --> 00:44:23,400
How this agent thinks about 
risk, and then we search through

828
00:44:23,400 --> 00:44:27,300
the whole parameter space. 
So we say, we're going to grid 

829
00:44:27,300 --> 00:44:30,200
search from 0 to 1 on their risk
level. 

830
00:44:30,500 --> 00:44:34,100
And then show kind of these heat
Maps or like these plots or 

831
00:44:34,100 --> 00:44:38,600
these kind of more descriptive 
statistics about how at each 

832
00:44:38,600 --> 00:44:41,300
parameter, how the system 
behaves. 

833
00:44:42,100 --> 00:44:45,300
So you kind of have to do both. 
That's maybe a long winded 

834
00:44:45,300 --> 00:44:50,000
answer to that where you want 
some historical types of users, 

835
00:44:50,200 --> 00:44:53,400
but you also want to try to make
sure you parameterize a space in

836
00:44:53,400 --> 00:44:55,200
a flexible enough way that you 
can search. 

837
00:44:56,700 --> 00:44:59,100
I want to ask you about 
transaction fees and if you're 

838
00:44:59,100 --> 00:45:02,500
doing any research there and how
important that is and sort of 

839
00:45:02,700 --> 00:45:06,300
mechanism design space. 
Yeah, so I think in traditional 

840
00:45:06,300 --> 00:45:11,400
mechanism design, it it's not 
quite, it's not quite well 

841
00:45:11,400 --> 00:45:16,200
studied and I think, you know, 
we're one of our advisors is Tim

842
00:45:16,200 --> 00:45:20,400
roughgarden and we're constantly
educating him a lot about the 

843
00:45:20,400 --> 00:45:23,200
this type of stuff and he's 
really been like, hey yeah, like

844
00:45:23,200 --> 00:45:25,200
we just didn't really, you know,
we spent the last 20 years 

845
00:45:25,300 --> 00:45:29,100
Years, building auctions for 
Google because that mechanism 

846
00:45:29,100 --> 00:45:31,200
designers and algorithmic Game 
Theory. 

847
00:45:31,200 --> 00:45:33,700
Folks. 
Sorry when you say when you say 

848
00:45:33,700 --> 00:45:36,900
in traditional mechanism design,
you mean transaction fees 

849
00:45:36,900 --> 00:45:40,300
applied to other mechanisms than
blockchains like? 

850
00:45:40,300 --> 00:45:43,000
I mean, what other in other in 
what other places do? 

851
00:45:43,000 --> 00:45:45,900
We see like transactions fees as
part of mechanism design for 

852
00:45:45,900 --> 00:45:46,500
sure? 
Yeah. 

853
00:45:46,500 --> 00:45:52,700
So the biggest I would say 
practical user of mechanism 

854
00:45:52,700 --> 00:45:55,200
design that exists in the world 
is online. 

855
00:45:55,400 --> 00:45:58,000
Actions. 
Okay, what sorry what I mean by 

856
00:45:58,000 --> 00:46:00,900
mechanism design was 
specifically like sorry I was 

857
00:46:00,900 --> 00:46:03,700
talking about like crypto 
cryptocurrency mechanisms like 

858
00:46:03,700 --> 00:46:06,000
the big basically like this 
cryptocurrency design. 

859
00:46:06,100 --> 00:46:08,300
Yeah. 
What I guess what I mean is a 

860
00:46:08,300 --> 00:46:11,200
lot of the math that has been 
invented for traditional 

861
00:46:11,200 --> 00:46:15,100
mechanism design, doesn't 
include transaction fees in the 

862
00:46:15,100 --> 00:46:17,700
way that blockchains use 
transaction fees. 

863
00:46:18,400 --> 00:46:22,200
And what I mean by that is when 
I say buying an ad or I'm 

864
00:46:22,800 --> 00:46:26,300
connecting to a Futures 
Exchange, I don't pay per 

865
00:46:26,300 --> 00:46:28,100
message, I sent to the exchange,
right? 

866
00:46:28,100 --> 00:46:32,500
But in crypto I have to actually
pay per message that I send and 

867
00:46:32,500 --> 00:46:36,300
so that actually changes the 
Dynamics of the lot of the, a 

868
00:46:36,300 --> 00:46:39,500
lot of the math and a lot of the
math that works for a adoptions 

869
00:46:39,500 --> 00:46:43,200
is completely invalid for blocks
to make sense because of this, 

870
00:46:43,700 --> 00:46:46,400
it's a new research space 
essentially what you're saying. 

871
00:46:47,000 --> 00:46:49,800
Yeah, it's 100% new research 
because people don't think about

872
00:46:49,800 --> 00:46:54,800
this paper message aspect of it,
it's assumed that like any user 

873
00:46:54,800 --> 00:46:58,000
can send As many messages as 
they want to Google or Facebook 

874
00:46:58,800 --> 00:47:01,200
and they don't have to pay like 
they're kind of paying in like, 

875
00:47:01,200 --> 00:47:04,400
there's some DDOS prevention, 
but there's not like a like hey 

876
00:47:04,400 --> 00:47:07,100
you actually took to pay for 
spam prevention. 

877
00:47:08,300 --> 00:47:11,100
And so what we do is we spend a 
lot of time modeling this, we 

878
00:47:11,100 --> 00:47:15,800
don't model it say in the way 
that we could probably prove a 

879
00:47:15,808 --> 00:47:20,600
theorem about it but we do try 
to say how should I value if I'm

880
00:47:20,600 --> 00:47:24,600
a minor and I get I have a 
mempool how should I value a 

881
00:47:24,607 --> 00:47:26,500
certain? 
A mutation right? 

882
00:47:26,500 --> 00:47:31,000
Because a mempool is a set of 
unordered set of transactions 

883
00:47:31,400 --> 00:47:36,600
and the the kind of notion of a 
I chose some subset of it and bi

884
00:47:36,600 --> 00:47:39,400
chosen ordering of that subset. 
That's the value that's 

885
00:47:39,400 --> 00:47:41,600
extracted by the minor. 
Right? 

886
00:47:42,900 --> 00:47:47,400
And so we tried to take kind of 
the more machine learning issues

887
00:47:47,600 --> 00:47:51,200
more statistical approach to it 
which traditional mechanism 

888
00:47:51,200 --> 00:47:53,100
designers would say. 
Oh well, like how do you know 

889
00:47:53,100 --> 00:47:55,600
it's optimal? 
We just try to say like hey Any 

890
00:47:55,600 --> 00:47:59,000
local Optimum is good enough, 
which is sort of the machine 

891
00:47:59,000 --> 00:48:02,700
learning a fit of like, what? 
Permutation what subset can you 

892
00:48:02,700 --> 00:48:04,400
pick? 
That will maximize your value 

893
00:48:04,400 --> 00:48:07,200
and then what ordering will 
maximize your oh. 

894
00:48:07,207 --> 00:48:11,500
So we measure that both in terms
of trying to predict 

895
00:48:11,500 --> 00:48:14,400
distributions of delays so 
submit transaction. 

896
00:48:14,700 --> 00:48:16,500
How long is the delay given a 
fee? 

897
00:48:17,100 --> 00:48:20,700
And then we also tried to say 
what permutation is like most 

898
00:48:20,700 --> 00:48:25,100
likely so we, but the problem is
prescribing value functions over

899
00:48:25,300 --> 00:48:29,900
Permutations is very difficult 
because a very large you know 

900
00:48:29,900 --> 00:48:34,500
this is n factorial space so you
have to kind of like come up 

901
00:48:34,500 --> 00:48:37,400
with some heuristics for that 
but roughly speaking that's what

902
00:48:37,400 --> 00:48:40,300
you do. 
The good news is that everyone 

903
00:48:40,300 --> 00:48:43,400
who's writing front-running? 
Bots is still a human and so 

904
00:48:43,400 --> 00:48:46,100
like they write a certain set of
strategies, right? 

905
00:48:46,100 --> 00:48:49,400
Like it's not like it's not like
there's like they're really 

906
00:48:49,400 --> 00:48:52,400
looking at the strategy that's 
like compute the ackermann 

907
00:48:52,400 --> 00:48:55,100
function divided by the maximum 
value. 

908
00:48:55,300 --> 00:48:59,500
It could have been and then use 
that as a random number to flip 

909
00:48:59,500 --> 00:49:02,900
a coin to decide on the 
ordering, right? 

910
00:49:02,900 --> 00:49:05,800
They're not going to choose some
crazy thing whose complexity is 

911
00:49:05,800 --> 00:49:08,900
like super factorial or 
something, right? 

912
00:49:08,900 --> 00:49:12,700
So, So, so far, you know, we've 
been discussing this in the 

913
00:49:12,707 --> 00:49:18,400
context of like simulating an 
existing designed game. 

914
00:49:19,500 --> 00:49:25,100
Do you guys also work on 
designing new games, altogether.

915
00:49:25,200 --> 00:49:30,000
So, in hft, for example, you 
could simulate hft, or you could

916
00:49:30,000 --> 00:49:33,700
solve some of the problems by 
like inventing batch execution. 

917
00:49:33,900 --> 00:49:36,900
So, when it comes to like, you 
know, for example, on a theory 

918
00:49:36,900 --> 00:49:41,000
on this like crazy gasps spikes 
that we've For the past couple 

919
00:49:41,000 --> 00:49:44,600
of weeks, you know we could 
continue to simulate this game 

920
00:49:44,600 --> 00:49:46,300
but it's probably not 
sustainable. 

921
00:49:46,300 --> 00:49:49,400
Like there's probably a good 
chance that the game design 

922
00:49:49,400 --> 00:49:54,200
itself is broken and we need to 
rethink How We Do block space 

923
00:49:54,200 --> 00:49:59,100
auctions in the first place. 
And so, would you be able to use

924
00:49:59,100 --> 00:50:03,800
similar methodologies to 
construct new games or is that, 

925
00:50:03,800 --> 00:50:06,500
or is the construction of new 
game, sort of something that has

926
00:50:06,500 --> 00:50:10,500
to be just intuitive? 
And then this stuff is only Used

927
00:50:10,500 --> 00:50:16,200
to test them out. 
I think it's sort of a there's a

928
00:50:16,200 --> 00:50:19,500
feedback loop, right? 
Of like I have an idea, I run a 

929
00:50:19,508 --> 00:50:22,500
bunch of simulations, I see if 
it works and then I see what 

930
00:50:22,500 --> 00:50:25,600
doesn't work. 
And then I mutate my idea until 

931
00:50:25,600 --> 00:50:30,600
like, I kept some type of 
minimum, like, Optimum solution.

932
00:50:31,700 --> 00:50:34,000
I think a lot of the problem 
with things like designing box 

933
00:50:34,000 --> 00:50:37,800
base auctions is like they. 
There's a really 

934
00:50:37,800 --> 00:50:40,200
well-established Theory, that's 
very attractive. 

935
00:50:40,300 --> 00:50:43,600
People to use which is the 
theory of adoption. 

936
00:50:43,600 --> 00:50:47,500
So a lot of the papers on that I
would say that especially by 

937
00:50:47,500 --> 00:50:51,400
crypto of professors are just 
like cribbing algorithmic games 

938
00:50:51,400 --> 00:50:55,200
are results and saying like hey 
they apply here but I think that

939
00:50:55,200 --> 00:50:58,800
like a lot there is certainly 
some theoretical Innovation. 

940
00:50:58,800 --> 00:51:02,400
You have to make first I think 
you do have to write the 

941
00:51:02,400 --> 00:51:04,900
correct. 
Mathematical framework and 

942
00:51:04,900 --> 00:51:08,600
equations before you can really 
stimulate. 

943
00:51:08,600 --> 00:51:11,700
But I do think simulation tells.
You when you're wrong, it 

944
00:51:11,700 --> 00:51:14,400
doesn't tell you you're right, 
but it definitely tells you when

945
00:51:14,400 --> 00:51:17,800
you're wrong. 
So it's kind of like a property 

946
00:51:17,800 --> 00:51:22,500
test, like, you know, you say 
this model should do X kind of 

947
00:51:22,500 --> 00:51:25,700
like in formal verification 
except add a statistical you 

948
00:51:25,700 --> 00:51:29,800
say, on average, this type of 
block, space thing should do X 

949
00:51:30,600 --> 00:51:32,800
and you use simulation to verify
and then you find. 

950
00:51:32,800 --> 00:51:36,000
Hey, it doesn't work so like I 
must have made the wrong model 

951
00:51:36,100 --> 00:51:37,300
so now I have to change 
something. 

952
00:51:37,300 --> 00:51:40,800
And when I say model here, I 
mean First price auction. 

953
00:51:40,800 --> 00:51:45,300
Second price auction weird like 
auction mechanic for box base 

954
00:51:45,300 --> 00:51:48,100
that you choose, right? 
Like, you somehow have to kind 

955
00:51:48,100 --> 00:51:51,000
of, you know, you can think of 
you should really think of 

956
00:51:51,000 --> 00:51:53,200
simulations, a way of doing 
this, property testing and 

957
00:51:53,200 --> 00:51:59,600
verification. 
So what piece of the crypt or 

958
00:51:59,600 --> 00:52:04,300
economic be old or stack? 
Do you think would most benefit 

959
00:52:04,300 --> 00:52:07,800
right now? 
But today from some of this 

960
00:52:08,300 --> 00:52:11,500
simulation Work. 
So it would it be like the proof

961
00:52:11,500 --> 00:52:13,000
of stake? 
Protocols, is it? 

962
00:52:13,000 --> 00:52:17,200
The fee models, is it some of 
these on chain device stuff? 

963
00:52:17,200 --> 00:52:22,900
Like lending dex's, I used to 
think it was perfect steak 

964
00:52:22,900 --> 00:52:26,300
itself. 
I think, I think the problem for

965
00:52:26,300 --> 00:52:30,300
proof of stake, from a more 
practical standpoint is that 

966
00:52:30,800 --> 00:52:33,400
people are just more 
risk-averse, which is good. 

967
00:52:33,500 --> 00:52:36,800
You should be very risk-averse 
for your base layer but that 

968
00:52:36,800 --> 00:52:42,300
also means you're like way too. 
Slow to like try to like update.

969
00:52:42,300 --> 00:52:45,400
Like you know, simulation should
be used in like we did 

970
00:52:45,400 --> 00:52:47,000
something. 
We observe something. 

971
00:52:47,200 --> 00:52:49,100
We try to predict, what will 
happen, given the new 

972
00:52:49,100 --> 00:52:52,200
observations, and we update, and
you kind of have this feedback 

973
00:52:52,200 --> 00:52:55,800
loop repeatedly applied. 
That's when it works best. 

974
00:52:55,900 --> 00:52:59,500
So, like, that's what happens in
trading, that's what happens in 

975
00:52:59,700 --> 00:53:03,800
chip design when simulation 
tools and other places But I 

976
00:53:03,808 --> 00:53:06,600
think proof of stake is like 
very very slow. 

977
00:53:06,900 --> 00:53:11,000
Like I and and like Defy is 
basically copying proof of stake

978
00:53:11,000 --> 00:53:14,000
except it's replacing proof of 
stake as a with an insurance 

979
00:53:14,000 --> 00:53:17,800
fund type of thing. 
And I think yeah the D5 

980
00:53:17,800 --> 00:53:20,700
parameters are really the 
biggest deal right now for sure.

981
00:53:20,700 --> 00:53:23,000
Because like people are doing 
all the stuff that they said 

982
00:53:23,000 --> 00:53:25,500
they would do improve mistake, 
except they're doing it like 

983
00:53:25,500 --> 00:53:28,400
recklessly. 
So I think in the long run proof

984
00:53:28,400 --> 00:53:32,000
of stake will learn a lot of the
lessons of failure from these 

985
00:53:32,000 --> 00:53:35,200
defy things. 
But yeah, right now, it's just 

986
00:53:35,200 --> 00:53:39,300
so much more, you know, you can 
like make a prediction, someone 

987
00:53:39,300 --> 00:53:43,700
does it, see how it works? 
Use that, as an example to add 

988
00:53:43,700 --> 00:53:47,600
to your simulation and, and 
like, that's is happening in all

989
00:53:47,600 --> 00:53:50,100
in defy, right now. 
I just don't think it's really 

990
00:53:50,100 --> 00:53:54,200
happening, and proof of stake. 
How much does, like, sort of 

991
00:53:54,200 --> 00:53:56,500
governance actually impact a lot
of this stuff. 

992
00:53:56,500 --> 00:53:59,900
So, so, when you do these 
simulations or like this 

993
00:53:59,900 --> 00:54:04,900
mechanism design, you have like 
some You love what like some 

994
00:54:04,900 --> 00:54:10,800
socially Optimum utility is for 
the entire system and if I was 

995
00:54:10,800 --> 00:54:13,900
like a benevolent dictator and I
wanted to maximize like the 

996
00:54:13,900 --> 00:54:18,800
social Optimum for this thing, 
you know, you could figure that 

997
00:54:18,800 --> 00:54:21,100
out, tell me what the best 
mechanism is and I can go to 

998
00:54:21,100 --> 00:54:24,800
Ploy, but now what happens when 
you know there's a governance 

999
00:54:24,800 --> 00:54:27,600
token. 
And so sometimes the holders of 

1000
00:54:27,600 --> 00:54:30,900
the governance token are not 
trying to maximize socially 

1001
00:54:30,900 --> 00:54:32,900
Optimum. 
They're trying to maximize, you 

1002
00:54:32,900 --> 00:54:36,400
know, They themselves are 
rational agents so it seems like

1003
00:54:36,400 --> 00:54:39,800
it becomes this like very weird 
meta thing thing that you have 

1004
00:54:39,800 --> 00:54:43,700
to also account for you have to 
model first you have to model 

1005
00:54:43,700 --> 00:54:47,600
the game that's the mechanism. 
So let's say, you know, it's 

1006
00:54:47,600 --> 00:54:49,500
curved. 
But now you also have to Maxim 

1007
00:54:49,500 --> 00:54:53,500
model like the incentives of the
curve governance to the dowels. 

1008
00:54:53,500 --> 00:54:56,700
Yeah. 
I think the key is to inject 

1009
00:54:56,700 --> 00:54:59,100
simulation into the decision 
making process. 

1010
00:54:59,100 --> 00:55:03,000
So like when someone is 
proposing a vote, you run a 

1011
00:55:03,008 --> 00:55:05,500
bunch of simulations and you 
say, like here are the set of 

1012
00:55:05,500 --> 00:55:09,600
outcome given these types of 
utility functions, put yourself 

1013
00:55:09,600 --> 00:55:12,100
in one of them. 
And if you don't find yourself 

1014
00:55:12,100 --> 00:55:13,700
in one of them then you can 
complain. 

1015
00:55:14,200 --> 00:55:16,700
But I assume you have these sets
of value functions. 

1016
00:55:16,700 --> 00:55:19,300
We've run these simulations 
under different edge cases. 

1017
00:55:19,500 --> 00:55:21,300
And here are the property is at 
hold. 

1018
00:55:21,300 --> 00:55:24,700
And here's a probability they 
hold with like if everyone's a 

1019
00:55:24,800 --> 00:55:26,500
Gambler of the probability of 
the system. 

1020
00:55:26,500 --> 00:55:29,200
Going. 
The zero went from point is wet 

1021
00:55:29,200 --> 00:55:33,400
from previously before this 
vote, one basis point two five 

1022
00:55:33,400 --> 00:55:36,300
percent. 
Like okay, that's something 

1023
00:55:36,300 --> 00:55:38,600
right like and I can give you an
uncertainty estimates. 

1024
00:55:38,600 --> 00:55:41,500
So one of the things that I was 
saying before about like Point 

1025
00:55:41,500 --> 00:55:45,000
estimates Finance machine 
learning verse uncertainty 

1026
00:55:45,000 --> 00:55:48,400
estimates, actuaries, insurance 
statisticians. 

1027
00:55:49,500 --> 00:55:53,200
Is that if you can provide good 
uncertainty estimates, if I tell

1028
00:55:53,200 --> 00:55:57,100
you, it's hey I'm increasing 
from one basis point of a chance

1029
00:55:57,100 --> 00:56:01,300
of s and X going to 0, to 5 
percent of a chance of Aston X 

1030
00:56:01,300 --> 00:56:05,400
going to 0 but five percent plus
or minus 0.2 percent. 

1031
00:56:06,100 --> 00:56:08,400
Then you actually have a lot 
more content like if I can give 

1032
00:56:08,400 --> 00:56:11,400
you more and more confidence 
that like this is an increasing 

1033
00:56:11,400 --> 00:56:17,400
thing, then governance actually 
you can impact governance in a 

1034
00:56:17,400 --> 00:56:21,600
way that's quantitative and not.
This emotional view of the world

1035
00:56:21,600 --> 00:56:24,900
because like at the end of the 
day, it's a new field. 

1036
00:56:25,500 --> 00:56:31,500
People don't really like their 
voting kind of blindly and at 

1037
00:56:31,500 --> 00:56:33,900
least giving some sort of 
uncertainty estimate. 

1038
00:56:34,700 --> 00:56:39,700
Lets people be like, okay, well 
I'm rational, but I'm not a my 

1039
00:56:39,700 --> 00:56:42,000
rational and iterated game, 
right? 

1040
00:56:42,000 --> 00:56:45,100
Like, if I'm taking a single 
step, it's rational for me to 

1041
00:56:45,100 --> 00:56:48,800
say, increase the fees 99% 
because I'm an SN X folder and I

1042
00:56:48,800 --> 00:56:53,400
want All those fees, but this 
kind of iterated. 

1043
00:56:53,400 --> 00:56:57,600
Simulation says, if the game 
lasts long enough, we might go 

1044
00:56:57,600 --> 00:57:03,400
to 0 if I try to collect any 
fees and if that happens, is 

1045
00:57:03,400 --> 00:57:06,200
that really worth it. 
And so it gives people a way to 

1046
00:57:06,200 --> 00:57:09,100
figure out their own valuation 
of how much they want to risk 

1047
00:57:09,100 --> 00:57:11,600
adjust. 
So I think simulation is not 

1048
00:57:11,600 --> 00:57:14,500
going to be able to predict 
perfectly this governance 

1049
00:57:14,500 --> 00:57:17,400
actions but it's going to show 
you the outcomes under what 

1050
00:57:17,400 --> 00:57:19,600
happens when you choose. 
Them. 

1051
00:57:20,000 --> 00:57:22,600
And so it's an integral part of 
giving quantitative 

1052
00:57:22,600 --> 00:57:25,500
justification for these things. 
I did in the normal world, it's 

1053
00:57:25,500 --> 00:57:29,300
much harder actually to impact 
governance in a quantitative 

1054
00:57:29,300 --> 00:57:31,100
way. 
Whereas in crypto, that actually

1055
00:57:31,100 --> 00:57:33,900
feels like it's quite tangible. 
But you're like, giving people 

1056
00:57:34,300 --> 00:57:38,700
risk assessments based on a lot 
of, like, a lot of very clear 

1057
00:57:38,700 --> 00:57:41,800
financial data. 
In a lot of stuff you do in the 

1058
00:57:41,800 --> 00:57:45,400
real world, you have to infer, 
whether the data is real, 

1059
00:57:45,400 --> 00:57:47,900
whether it's accurate. 
Sometimes you're like, well, 

1060
00:57:47,900 --> 00:57:50,800
someone may have been like And 
of injecting noise into the 

1061
00:57:50,800 --> 00:57:53,600
data. 
But the on chain data being 

1062
00:57:53,900 --> 00:57:56,800
something you trust in those 
valid is actually quite 

1063
00:57:56,800 --> 00:58:01,100
important for those I'd like to 
come back to an earlier point, 

1064
00:58:01,100 --> 00:58:04,800
which is, let's imagine that 
someone is like building a new 

1065
00:58:05,200 --> 00:58:07,700
Block Chain as as that happens 
these days. 

1066
00:58:07,700 --> 00:58:11,100
Right? 
And, you know, a lot of times I 

1067
00:58:11,100 --> 00:58:15,100
think teams are kind of focused 
on on, like building the product

1068
00:58:15,100 --> 00:58:17,000
growing and community. 
And then, of course, one of the 

1069
00:58:17,000 --> 00:58:21,100
things that often comes up at 
some point, is doing a security 

1070
00:58:21,100 --> 00:58:24,800
audit and the security audit 
will entail like a bunch of 

1071
00:58:24,808 --> 00:58:27,200
things. 
But there's some like design 

1072
00:58:27,700 --> 00:58:29,700
aspects that are also Part of 
that audit. 

1073
00:58:29,700 --> 00:58:35,100
I'm curious how you consider 
your work to be complementary to

1074
00:58:35,100 --> 00:58:38,700
that, or you should I call that 
replace it? 

1075
00:58:38,700 --> 00:58:42,100
Is it better, or is it a little 
bit somewhere in between like 

1076
00:58:42,100 --> 00:58:45,300
where you put yourself in that 
sort of like early stage 

1077
00:58:45,300 --> 00:58:47,900
research when one is building a 
blockchain? 

1078
00:58:47,900 --> 00:58:52,600
I think it's pretty 
complementary to both normal 

1079
00:58:52,600 --> 00:58:55,600
Audits and formal verification. 
Because I think one of the 

1080
00:58:55,600 --> 00:58:58,700
problems for formal verification
is required. 

1081
00:58:58,800 --> 00:59:01,900
Related to the thing you're 
talking about which is that 

1082
00:59:02,100 --> 00:59:05,900
naively there's an exponential 
State space blow up. 

1083
00:59:05,900 --> 00:59:09,400
Once I start interacting to 
system, like K systems once K 

1084
00:59:09,400 --> 00:59:13,900
systems interact, you have this 
the naive notion of the number 

1085
00:59:13,900 --> 00:59:16,400
of bits. 
You need is blowing linearly in 

1086
00:59:16,408 --> 00:59:20,800
case the space is kind of blown 
up exponentially but simulation 

1087
00:59:20,800 --> 00:59:23,000
is more about like well what's 
the behavior? 

1088
00:59:23,000 --> 00:59:26,200
That's not. 
If I don't have to sample every 

1089
00:59:26,200 --> 00:59:30,200
possible action, if I sample the
most Reactions as well as the 

1090
00:59:30,200 --> 00:59:33,000
ones that are near the most 
likely actions, what happens? 

1091
00:59:33,800 --> 00:59:36,100
And so there are two different 
types of scenarios, right? 

1092
00:59:36,100 --> 00:59:39,500
One is the pure worst case, but 
might take infinitely long to 

1093
00:59:39,500 --> 00:59:42,200
search through this set of 
tests. 

1094
00:59:42,900 --> 00:59:46,800
And the other is, how do I kind 
of use the expected Behavior to 

1095
00:59:46,800 --> 00:59:50,800
estimate risk in a way that is 
intuitive and interpretable to 

1096
00:59:50,800 --> 00:59:55,400
the non-developer? 
And both of those I think are 

1097
00:59:55,400 --> 00:59:58,500
are valid ways of stress testing
but they're very different. 

1098
00:59:58,900 --> 01:00:00,900
And they are extremely 
complimentary. 

1099
01:00:00,900 --> 01:00:04,300
So like when an exchange when 
like the CME the Chicago 

1100
01:00:04,300 --> 01:00:07,200
Mercantile Exchange which is 
like the biggest features Market

1101
01:00:07,200 --> 01:00:11,200
in the world when they build a 
new piece of software, they of 

1102
01:00:11,200 --> 01:00:14,100
course get audited and they do 
kind of traditional 

1103
01:00:14,600 --> 01:00:18,100
cybersecurity as but they also 
do simulation and they stress 

1104
01:00:18,100 --> 01:00:21,200
test like, hey, did we choose 
the right tick size as a 

1105
01:00:21,207 --> 01:00:24,700
parameter. 
Did we like our we resistant to 

1106
01:00:24,800 --> 01:00:27,500
kind of certain types of 
malicious trading strategies 

1107
01:00:27,500 --> 01:00:30,700
that try to like block Now the 
market and there's a whole 

1108
01:00:30,700 --> 01:00:34,000
literature of this like the SEC 
themselves spends a bunch of 

1109
01:00:34,000 --> 01:00:36,800
time. 
Doing these stress tests on 

1110
01:00:36,900 --> 01:00:40,300
Exchange code to meet, show that
they meet compliance. 

1111
01:00:40,400 --> 01:00:43,100
And so they're very 
complimentary but they stress 

1112
01:00:43,100 --> 01:00:47,100
has very different things. 
One is really stressing user 

1113
01:00:47,100 --> 01:00:50,900
Behavior and the other is stress
testing like code behavior. 

1114
01:00:51,700 --> 01:00:56,100
And user behavior is about 
probabilities. 

1115
01:00:56,300 --> 01:00:58,700
Code behavior is about 
determinism period. 

1116
01:00:58,800 --> 01:01:03,300
Feminism, but they're related. 
So in all of our simulation 

1117
01:01:03,300 --> 01:01:06,900
research and I think this is 
what distinguishes us from other

1118
01:01:06,900 --> 01:01:11,900
people, that who's kind of tried
to do this, is we run everything

1119
01:01:11,900 --> 01:01:15,700
against the real code. 
So we build kind of think of 

1120
01:01:15,700 --> 01:01:21,300
open AI gym or like, you know, 
the alphago training program. 

1121
01:01:21,700 --> 01:01:24,800
What happens is people build a 
harness around the real piece of

1122
01:01:24,800 --> 01:01:29,800
code and then the harness has a 
way has a The interfaces that 

1123
01:01:29,800 --> 01:01:31,600
you can model the different 
types of users. 

1124
01:01:31,600 --> 01:01:34,700
And you make a domain specific 
language that you can program 

1125
01:01:34,700 --> 01:01:37,400
the different types of users and
then the users interact with the

1126
01:01:37,400 --> 01:01:39,700
real code. 
And I think I've seen a lot of 

1127
01:01:39,700 --> 01:01:43,500
simulation, especially in 2017, 
I saw a lot of kind of less 

1128
01:01:43,500 --> 01:01:46,900
rigorous simulation stuff that 
kind of is like, hey, well, we 

1129
01:01:46,900 --> 01:01:50,700
think the model of how the Block
Chain itself works, is this and 

1130
01:01:50,700 --> 01:01:53,000
we're going to say this is a 
poisson process and this is at 

1131
01:01:53,000 --> 01:01:56,600
this thing, and this is a dis 
thing and then we have models of

1132
01:01:56,600 --> 01:02:01,100
user interact with them. 
The Um is in reality, like a lot

1133
01:02:01,100 --> 01:02:04,700
of these code things that cause 
problems for formal verification

1134
01:02:04,700 --> 01:02:08,200
or security out, there's also 
will affect the economics in 

1135
01:02:08,300 --> 01:02:11,900
super edge cases. 
So you want to minimize the 

1136
01:02:11,900 --> 01:02:16,200
amount of surface area that you 
seed to your model, you want to 

1137
01:02:16,200 --> 01:02:17,700
say? 
Hey look we're running this 

1138
01:02:17,700 --> 01:02:20,400
against the real code as much as
possible. 

1139
01:02:21,200 --> 01:02:23,600
And this is something people in 
trading do a lot. 

1140
01:02:23,600 --> 01:02:27,800
And I think that I only really 
respected this ones, I saw the 

1141
01:02:27,800 --> 01:02:32,600
difference in trading Between. 
Hey, like this exchange happens 

1142
01:02:32,600 --> 01:02:36,500
to use only 18 bit fixed Point 
integers and like all of a 

1143
01:02:36,500 --> 01:02:39,200
sudden this strategy loses 
money, right? 

1144
01:02:39,200 --> 01:02:42,700
Like that's that's the type of 
detail that you know I think a 

1145
01:02:42,700 --> 01:02:45,700
lot of people who are like oh 
well I just like Learn Python 

1146
01:02:45,700 --> 01:02:50,000
and use Pi torch and like I made
a model of your blockchain don't

1147
01:02:50,100 --> 01:02:53,100
kind of like are missing like 
they've never seen like people 

1148
01:02:53,100 --> 01:02:56,600
lose money because like the you 
went from floating point to 

1149
01:02:56,600 --> 01:03:00,600
18-bit fixed Point integer. 
I randomly and I think that 

1150
01:03:00,600 --> 01:03:03,000
that's why you need to actually 
run this stuff against the real 

1151
01:03:03,000 --> 01:03:05,900
code because you there's just 
like tons of weird developer 

1152
01:03:05,900 --> 01:03:09,200
decisions some random if 
statement somewhere that 

1153
01:03:09,200 --> 01:03:12,100
completely like takes all the 
money out and you don't realize 

1154
01:03:12,100 --> 01:03:15,300
why until you like actually are 
running like tons of. 

1155
01:03:15,300 --> 01:03:18,800
So I it's complementary that I 
just don't see. 

1156
01:03:18,800 --> 01:03:22,000
I think security officers are 
really focused on, like, binary 

1157
01:03:22,100 --> 01:03:24,200
objective functions of like, 
does it property? 

1158
01:03:24,200 --> 01:03:28,700
Testing of like? 
Yes, no and, you know, I think, 

1159
01:03:28,800 --> 01:03:32,000
What we're focused on this is 
kind of like statistical version

1160
01:03:32,000 --> 01:03:35,600
of that have like but we still 
want to run against the real 

1161
01:03:35,600 --> 01:03:37,700
code, right? 
We still you know compiled 

1162
01:03:37,700 --> 01:03:39,600
against whatever Docker image 
you give us. 

1163
01:03:39,600 --> 01:03:42,500
We run it against whatever 
kernel module is you say it 

1164
01:03:42,500 --> 01:03:47,500
should be running against 
because I think you never know 

1165
01:03:47,500 --> 01:03:51,100
when like a some random piece of
the could just change the 

1166
01:03:51,100 --> 01:03:57,200
economics, completely, What kind
of tools do you use to do this? 

1167
01:03:57,200 --> 01:04:01,000
Like, how do you all? 
These are custom in-house built 

1168
01:04:01,000 --> 01:04:03,900
tools or are there. 
Like so I'm, you know, public 

1169
01:04:03,900 --> 01:04:06,900
tools that you kind of used to 
do these sort of simulations, 

1170
01:04:06,900 --> 01:04:10,400
both, you know, with dummy code.
And then also, when you want to 

1171
01:04:10,400 --> 01:04:14,600
test with a real code, Yeah. 
So similar to kind of security 

1172
01:04:14,600 --> 01:04:19,600
Auditors, we kind of have we 
kind of build a lot of our own 

1173
01:04:19,600 --> 01:04:25,000
versions of the virtual machines
themselves and like add in extra

1174
01:04:25,000 --> 01:04:28,600
tracing functionality and extra 
kind of like tracking 

1175
01:04:28,600 --> 01:04:34,100
functionality for like agent 
submits a trade to you the Swap 

1176
01:04:34,100 --> 01:04:37,700
and we track kind of like we're 
through the client that 

1177
01:04:37,700 --> 01:04:41,700
transaction goes and like oh did
it get the halted of the certain

1178
01:04:41,800 --> 01:04:45,300
Point or, oh, did the networking
layer like look malformed at it.

1179
01:04:45,300 --> 01:04:48,500
But we spend a lot of time is 
certainly. 

1180
01:04:48,500 --> 01:04:50,400
Now we're pretty much only 
aetherium. 

1181
01:04:50,400 --> 01:04:56,100
We were doing a lot more other 
chains but honestly it defying 

1182
01:04:56,100 --> 01:04:58,900
aetherium has the most sort of 
Need for this. 

1183
01:04:58,900 --> 01:05:00,900
But so we kind of have written 
our own. 

1184
01:05:00,900 --> 01:05:06,900
We kind of have our own Fork of 
gath where we have optimized, a 

1185
01:05:06,908 --> 01:05:09,000
bunch of things for doing 
simulation. 

1186
01:05:09,000 --> 01:05:11,500
One thing to remember, when 
you're doing simulation is your 

1187
01:05:11,800 --> 01:05:15,300
Knowing the threat model, you're
describing all of the users in 

1188
01:05:15,300 --> 01:05:17,800
the system. 
So by controlling the threat 

1189
01:05:17,800 --> 01:05:21,000
model, you can actually reduce a
lot of the cryptography burden 

1190
01:05:21,900 --> 01:05:24,200
and by doing that you can make 
the performance lot better. 

1191
01:05:24,900 --> 01:05:29,200
And so, we've spent a lot of 
time building this client with 

1192
01:05:29,200 --> 01:05:34,300
extra tracing and kind of ways 
of like, having multiple agents 

1193
01:05:34,300 --> 01:05:38,600
interact with the same node, 
multiple agents kind of work 

1194
01:05:38,600 --> 01:05:41,400
off, the same kind of simulated 
blockchains fate stuff like 

1195
01:05:41,400 --> 01:05:43,400
that. 
At and so yeah. 

1196
01:05:43,400 --> 01:05:48,400
So we a have that and then B, we
have sort of a domain specific 

1197
01:05:48,400 --> 01:05:53,000
language that we is mainly in 
Python because I think from a 

1198
01:05:53,000 --> 01:05:56,200
data science perspective, it's 
just like still too hot as much 

1199
01:05:56,200 --> 01:05:58,600
as I love. 
Julia, I'm sorry. 

1200
01:05:58,600 --> 01:06:02,000
Julia fans. 
It's just still not quite quite 

1201
01:06:02,000 --> 01:06:05,200
there, but it's always one year 
away, right. 

1202
01:06:05,400 --> 01:06:08,200
But I'd sorry. 
Julia is the scientific 

1203
01:06:08,200 --> 01:06:11,400
programming language that's like
way faster than python. 

1204
01:06:11,400 --> 01:06:16,100
It's It's like compiles to rust 
and C++ like supposed to be like

1205
01:06:16,100 --> 01:06:19,600
the, the real deal, but yet if 
you talk to every data scientist

1206
01:06:19,600 --> 01:06:21,900
in the world, they're gonna tell
you they use python or r or 

1207
01:06:21,900 --> 01:06:24,300
something. 
So we have these python 

1208
01:06:24,300 --> 01:06:29,900
bindings, we have this DSL, DSL 
compiles, to some byte code that

1209
01:06:29,900 --> 01:06:34,500
basically gets run against the 
virtual machine directly. 

1210
01:06:34,500 --> 01:06:38,200
So, there's like kind of a layer
in between that take the 

1211
01:06:38,200 --> 01:06:41,100
compiled agent code, and 
interact has it interact with 

1212
01:06:41,100 --> 01:06:46,300
the virtual machine, I think, in
a world, you know, kind of in 

1213
01:06:46,300 --> 01:06:50,000
the same way that it took Trail,
a b forever to open source, a 

1214
01:06:50,000 --> 01:06:53,900
lot of critic. 
I think we Open sourced some of 

1215
01:06:53,900 --> 01:06:55,300
it though. 
It's just going to take us a 

1216
01:06:55,308 --> 01:06:57,500
while. 
But yeah, right right now it's 

1217
01:06:57,500 --> 01:07:01,300
me only that type of stuff. 
A lot of what we use is based on

1218
01:07:02,100 --> 01:07:06,700
a lot of the work that Google 
and Facebook have done on 

1219
01:07:07,400 --> 01:07:12,700
compiling python models to C++. 
That that type of stuff has been

1220
01:07:12,800 --> 01:07:17,000
is like really deep in our in 
our stack for increasing 

1221
01:07:17,000 --> 01:07:20,900
performance. 
So before we wrap up here, I'd 

1222
01:07:20,900 --> 01:07:24,600
like to ask you a little bit 
about Gauntlet the business and 

1223
01:07:24,900 --> 01:07:26,300
what does the current business 
look like? 

1224
01:07:26,300 --> 01:07:29,000
I mean you guys put all these 
reports and all this research 

1225
01:07:29,000 --> 01:07:33,300
and but who do you work for? 
And then also you know what's 

1226
01:07:33,300 --> 01:07:35,900
the sort of roadmap and plans 
looking forward. 

1227
01:07:36,600 --> 01:07:38,600
Yeah. 
So, you know, I think a lot of 

1228
01:07:38,600 --> 01:07:42,800
what we do right now is putting 
out reports working with the 

1229
01:07:42,800 --> 01:07:47,400
protocols themselves kind of 
close to security outdoors 

1230
01:07:47,400 --> 01:07:50,500
although we've been Seeing a lot
more of an active role in 

1231
01:07:50,500 --> 01:07:52,900
governance. 
So we're sort of the 

1232
01:07:53,200 --> 01:08:00,400
third-largest comp older and 
governance by votes and we have 

1233
01:08:00,400 --> 01:08:03,200
a bunch of stuff. 
We're working on right now to 

1234
01:08:03,500 --> 01:08:07,200
try to automate the Actuarial 
predictions. 

1235
01:08:07,200 --> 01:08:10,600
I was telling you about earlier.
So, imagine that there's a 

1236
01:08:10,607 --> 01:08:14,400
governance vote. 
Someone says, hey, we want to 

1237
01:08:14,400 --> 01:08:21,300
change the collateral Factor on 
compound for BTC to this value. 

1238
01:08:22,500 --> 01:08:26,600
We will basically other generate
a bunch of simulations and risk 

1239
01:08:26,600 --> 01:08:29,200
estimates for like what, this, 
what the before. 

1240
01:08:29,200 --> 01:08:34,200
And after of this particular 
vote, look like to our best 

1241
01:08:34,200 --> 01:08:37,500
estimates and then present them 
to the user in a way that's 

1242
01:08:38,399 --> 01:08:40,200
intuitive. 
So you can pick, oh, well, you 

1243
01:08:40,207 --> 01:08:43,700
know, by making this change, we 
decrease the probability of 

1244
01:08:43,700 --> 01:08:46,899
default by this amount. 
But then we also lower the 

1245
01:08:46,899 --> 01:08:52,000
revenue that the cash cash flow 
that the network gets By this 

1246
01:08:52,000 --> 01:08:55,200
amount and then, you know, a 
user who's like, maybe more 

1247
01:08:55,200 --> 01:08:59,200
financially educated but not so 
like in the Weeds on like how 

1248
01:08:59,200 --> 01:09:03,300
the protocol Works can do I go? 
Okay, I kind of get that this is

1249
01:09:03,300 --> 01:09:07,399
what this change does and we're 
also working on sort of what we 

1250
01:09:07,399 --> 01:09:13,500
call other gov which is A way 
where we monitor the markets and

1251
01:09:13,500 --> 01:09:17,899
then Auto submit proposals. 
So we do we have you know sort 

1252
01:09:17,899 --> 01:09:22,200
of some of the proposals are 
more simple but some of them 

1253
01:09:22,300 --> 01:09:25,800
need a little bit of like 
program synthesis where we 

1254
01:09:25,800 --> 01:09:28,100
generate the code for the 
proposal. 

1255
01:09:28,100 --> 01:09:31,600
Based we run a bunch of 
simulations we say hey there's 

1256
01:09:31,899 --> 01:09:35,899
way too much risk in because a 
bunch of yield Farmers decided 

1257
01:09:35,899 --> 01:09:41,399
to like mint too much as USD And
we're going to submit a proposal

1258
01:09:41,399 --> 01:09:45,800
that says, like, increased s USD
minting fee by X. 

1259
01:09:45,899 --> 01:09:49,100
And here's the reasons why, and 
here's the code for doing it. 

1260
01:09:49,500 --> 01:09:53,600
And so the dream is to have the 
sort of automated system that 

1261
01:09:53,600 --> 01:09:57,100
can monitor these things and 
submit proposals to governance 

1262
01:09:57,200 --> 01:10:01,500
in a fully automated fashion and
then the smart contracts pay for

1263
01:10:01,500 --> 01:10:04,700
this sir. 
I don't know if that it's a 

1264
01:10:04,708 --> 01:10:07,100
little bit of the opposite 
business model of most 

1265
01:10:07,100 --> 01:10:11,300
blockages, most blockchains And 
smart contracts want to make 

1266
01:10:11,300 --> 01:10:14,500
their coin worth lat. 
Whereas we want to kind of 

1267
01:10:15,100 --> 01:10:18,400
reduce the tragedy of the 
commons and have be kind of paid

1268
01:10:18,400 --> 01:10:20,500
as a service provider, but it's 
automated. 

1269
01:10:21,700 --> 01:10:25,500
And how do you get the time to 
like, spit out all these papers?

1270
01:10:25,500 --> 01:10:28,700
You know, one of my, you know, I
just remember, I had this idea, 

1271
01:10:28,800 --> 01:10:32,200
like a couple of months ago of 
like, oh, you can combine ideas 

1272
01:10:32,200 --> 01:10:34,400
from stellar and Avalanche your 
Grant. 

1273
01:10:34,400 --> 01:10:38,700
This new consensus protocol, and
then like you like, like, oh me,

1274
01:10:38,900 --> 01:10:41,400
My brother, we wrote this up, 
like, two months ago, for fun. 

1275
01:10:41,400 --> 01:10:43,800
And I'm like, what, like, where 
do you get all the time to, 

1276
01:10:43,808 --> 01:10:46,400
like, write all these papers and
like, is that part of the work 

1277
01:10:46,400 --> 01:10:48,500
you do with? 
You know, for example you wrote 

1278
01:10:48,500 --> 01:10:52,800
this paper on like you know, 
swap or just like in a M&M's in 

1279
01:10:52,800 --> 01:10:55,100
general, is that also work 
you're doing with these 

1280
01:10:55,100 --> 01:10:57,600
companies or is that sort of do 
something you do on the side? 

1281
01:10:59,400 --> 01:11:05,800
That is something we do on the 
side but I think it's very 

1282
01:11:05,800 --> 01:11:10,400
closely related in the sense 
that you know how you were 

1283
01:11:10,400 --> 01:11:15,100
talking earlier about like, can 
you discover new mechanisms by 

1284
01:11:15,100 --> 01:11:18,800
pure sort of simulation methods?
I think that there's kind of 

1285
01:11:18,800 --> 01:11:23,700
this interplay between the 
theory and actual discovery of 

1286
01:11:23,700 --> 01:11:26,400
these things, so you actually 
need to make the theory so that 

1287
01:11:26,400 --> 01:11:29,900
you can simulate it right? 
Like, once you have the theory 

1288
01:11:30,100 --> 01:11:33,100
you can start saying like here's
where the theory breaks and 

1289
01:11:33,100 --> 01:11:35,400
that's where we're going to 
simulate and that's where we're 

1290
01:11:35,400 --> 01:11:38,300
going to do kind of these stress
testing type things. 

1291
01:11:39,100 --> 01:11:45,000
And I feel like right now, the 
way that things look, especially

1292
01:11:45,000 --> 01:11:51,600
in defy, it feels a lot. 
Like the kind of late 2000s and 

1293
01:11:51,608 --> 01:11:55,700
early 2010's in, in machine 
learning, it feels a lot like 

1294
01:11:56,200 --> 01:12:01,700
quantitative Finance in the If 
we're, if you can figure out how

1295
01:12:01,700 --> 01:12:08,200
to make the valuation model that
people use, then you will 

1296
01:12:08,200 --> 01:12:10,700
actually impact the usage of 
these systems. 

1297
01:12:10,700 --> 01:12:14,300
And so that is related in that 
like yes, we use the same models

1298
01:12:14,300 --> 01:12:17,400
and simulation. 
But also, we have more people 

1299
01:12:17,400 --> 01:12:20,500
using these things because like 
they understand these Financial 

1300
01:12:20,500 --> 01:12:22,500
aspects. 
So there's kind of this duel 

1301
01:12:22,500 --> 01:12:27,500
play between like doing research
and convincing people that These

1302
01:12:27,500 --> 01:12:31,600
risk metrics are correct and I 
think I think writing the 

1303
01:12:31,600 --> 01:12:34,400
research is quite crucial to 
that. 

1304
01:12:34,400 --> 01:12:37,900
It's the equivalent of Open 
Source software for this type of

1305
01:12:37,900 --> 01:12:39,900
stuff. 
Cool. 

1306
01:12:39,900 --> 01:12:43,600
So where should people go to 
find out more about Gauntlet and

1307
01:12:43,600 --> 01:12:46,700
your work. 
So our Twitter is at Tom, 

1308
01:12:46,700 --> 01:12:50,700
Lynette work for me. 
My Twitter is at my name at 

1309
01:12:50,700 --> 01:12:57,700
tarun chitra so ta Ru n CH. 
I tra you know we're we publish 

1310
01:12:57,700 --> 01:13:00,400
a lot of stuff but I think we're
going to be coming to a 

1311
01:13:00,400 --> 01:13:03,600
governance vote near you soon. 
So cool. 

1312
01:13:04,200 --> 01:13:06,900
If you're if you're in that 
realm you will you will see us 

1313
01:13:06,900 --> 01:13:08,500
or you've already seen us in 
compound. 

1314
01:13:08,600 --> 01:13:13,000
But I think that's the story. 
Great. 

1315
01:13:13,000 --> 01:13:15,100
Thanks for coming on through. 
Yeah, thanks. 

1316
01:13:17,600 --> 01:13:19,400
It doesn't end here. 
There's more to this 

1317
01:13:19,400 --> 01:13:22,700
conversation and you can hear it
on epicenter premium as a 

1318
01:13:22,700 --> 01:13:26,100
premium subscriber, you'll get 
access to a private RSS feed 

1319
01:13:26,100 --> 01:13:29,100
where you can hear the interview
debrief and get enhanced 

1320
01:13:29,100 --> 01:13:32,500
features like full episode 
transcripts and chapters which 

1321
01:13:32,500 --> 01:13:35,000
allow you to easily skip to 
specific sections of the 

1322
01:13:35,000 --> 01:13:37,500
interview. 
You'll also get exclusive access

1323
01:13:37,500 --> 01:13:40,500
to Roundtable conversation. 
Shins with epicenter hosts and 

1324
01:13:40,500 --> 01:13:42,400
bonus content. 
We put out from time to time. 

1325
01:13:42,800 --> 01:13:46,400
Go to premium epicenter dot TV 
to become a subscriber and 

1326
01:13:46,400 --> 01:13:47,400
support the podcast.