1
00:00:00,240 --> 00:00:02,360
Hi, and welcome to the Neil
Ashton Podcast.

2
00:00:03,000 --> 00:00:05,880
In each episode, we explained
some of the fascinating ways

3
00:00:05,880 --> 00:00:09,080
that science and engineering are
changing the world around us.

4
00:00:09,760 --> 00:00:12,720
We talked to leading engineers
from elite level sports like

5
00:00:12,800 --> 00:00:16,840
cycling and Formula One to some
of the world's top academics to

6
00:00:16,840 --> 00:00:20,960
understand how fluid dynamics,
machine learning, supercomputing

7
00:00:21,360 --> 00:00:22,960
are bringing in a new era of
discovery.

8
00:00:23,920 --> 00:00:27,040
We also hear some of their life
stories, their career advice,

9
00:00:27,640 --> 00:00:30,080
the lessons they've learned on
the way that I hope will be

10
00:00:30,080 --> 00:00:33,800
helpful to you too.
So sit back and enjoy this

11
00:00:33,800 --> 00:00:41,760
episode.
Hi, and welcome back to the Neil

12
00:00:41,760 --> 00:00:44,160
Ashton pod guest.
So my guest today is Professor

13
00:00:44,160 --> 00:00:48,520
Nathan Coots, our true pioneer
at the intersection of machine

14
00:00:48,520 --> 00:00:51,160
learning, dynamical systems and
fluid mechanics.

15
00:00:52,000 --> 00:00:55,680
He, he's really been one of the
household names in this field

16
00:00:55,680 --> 00:01:01,600
and has led to, has been part of
many of the key sort of advances

17
00:01:02,480 --> 00:01:06,560
and now actually has moved on to
play to behind a key role at one

18
00:01:06,560 --> 00:01:10,040
of the major companies in the
space at Autodesk Research.

19
00:01:10,400 --> 00:01:12,400
You know, he, he started his
journey at University of

20
00:01:12,400 --> 00:01:14,320
Washington.
So he earned a degree in physics

21
00:01:14,320 --> 00:01:18,200
and maths there in 1990 and then
did a PhD at Northwestern

22
00:01:18,200 --> 00:01:20,400
University.
He returned back to University

23
00:01:20,400 --> 00:01:23,520
of Washington and was the chair
of the applied mathematics

24
00:01:23,520 --> 00:01:26,760
department and was also Co
directing the AI Institute in

25
00:01:26,760 --> 00:01:30,560
dynamic systems.
He's probably best known for his

26
00:01:30,560 --> 00:01:34,680
work on the dynamic mode
decomposition and DMD and also

27
00:01:34,680 --> 00:01:39,400
was a co-author with at his
time, one of his postdocs,

28
00:01:40,000 --> 00:01:43,260
Stephen Brunton on the Cindy
sparse identification of non

29
00:01:43,260 --> 00:01:48,440
linear dynamics, which really
was part of the breakthrough in

30
00:01:48,440 --> 00:01:52,960
the push in the mid 220 tens
2016.

31
00:01:52,960 --> 00:01:57,760
Well before machine learning for
for science and engineering was

32
00:01:57,760 --> 00:02:01,040
was such a household thing and
then such a you know, a key

33
00:02:01,040 --> 00:02:03,560
area.
So that arguably they worked on

34
00:02:03,560 --> 00:02:08,240
it whilst it was still not clear
how how important this space

35
00:02:08,240 --> 00:02:10,039
would be.
And the work that they did

36
00:02:10,039 --> 00:02:12,360
really has proved to be a
landmark paper.

37
00:02:12,560 --> 00:02:15,280
They also wrote a textbook
Data-driven Science and

38
00:02:15,280 --> 00:02:18,360
Engineering, and that was
between Steve Brunton and and

39
00:02:18,360 --> 00:02:21,960
Nathan.
But you know, and of course,

40
00:02:21,960 --> 00:02:25,360
many other papers, which we'll
link to in the the show notes

41
00:02:26,120 --> 00:02:28,600
that that really has been
influential in this field.

42
00:02:29,080 --> 00:02:32,880
One of the the reasons why he's
so interesting is that he's

43
00:02:32,880 --> 00:02:36,960
isn't just sort of academic.
He actually spent time

44
00:02:37,000 --> 00:02:40,680
sabbatical working alongside
Formula One in with the McLaren

45
00:02:40,680 --> 00:02:43,280
team.
And we talk in this episode

46
00:02:43,280 --> 00:02:48,200
about how that was really
influential for his thinking and

47
00:02:48,200 --> 00:02:51,200
trying to tackle the more
industrial problems.

48
00:02:51,440 --> 00:02:55,440
And that ultimately led to him
moving over and becoming the

49
00:02:55,440 --> 00:02:59,520
director of physics informed AI
at Autodesk Research.

50
00:03:00,160 --> 00:03:03,880
And I found this conversation
fascinating because we, we sort

51
00:03:03,880 --> 00:03:06,520
of went back and forth a little
bit on some of the history and

52
00:03:06,520 --> 00:03:10,000
the differences between reduced
order modelling and, and machine

53
00:03:10,000 --> 00:03:15,080
learning and then pivoted a lot
to, you know, this, this, this

54
00:03:15,080 --> 00:03:18,160
question of is, can we just
spend billions of dollars on

55
00:03:18,160 --> 00:03:20,960
data and, and will it all be
good and we have some foundation

56
00:03:20,960 --> 00:03:24,840
model or do we really need to,
to have some physics in there to

57
00:03:24,840 --> 00:03:28,560
make this more affordable and
achievable and interpretable?

58
00:03:29,480 --> 00:03:33,280
We we also pivoted and taught
quite a lot on the, the agents

59
00:03:33,280 --> 00:03:37,880
front and how we see those as
being key to distilling

60
00:03:37,880 --> 00:03:42,320
knowledge from, from what is
very expert driven processes.

61
00:03:42,720 --> 00:03:45,240
You know, we were saying that
it's not good enough just to

62
00:03:45,240 --> 00:03:48,720
have the access to a, to a
highly efficient code to

63
00:03:48,720 --> 00:03:52,800
generate data or a, a great
machine learning architecture.

64
00:03:52,800 --> 00:03:57,080
You need the human knowledge,
the processes that often these

65
00:03:57,080 --> 00:03:59,200
companies had.
And we said that distilling this

66
00:03:59,200 --> 00:04:02,840
into agents for those companies
could really help them to move

67
00:04:02,840 --> 00:04:04,920
faster.
And and we finished off looking

68
00:04:04,920 --> 00:04:08,680
more broadly at where the
world's going with this sort of

69
00:04:08,680 --> 00:04:10,880
technology.
One of the stand out things that

70
00:04:10,880 --> 00:04:15,360
he said, and I fully agree with
him, is in this age, the need to

71
00:04:15,360 --> 00:04:17,920
think big is more important than
ever.

72
00:04:17,920 --> 00:04:21,279
Things are moving so fast that
actually having that mindset of

73
00:04:21,279 --> 00:04:24,480
really challenging your own
thinking is important.

74
00:04:25,240 --> 00:04:28,800
And then the final topic was
really more career advice to to

75
00:04:28,800 --> 00:04:32,160
new people in this space and how
they should deal with this

76
00:04:32,160 --> 00:04:35,800
changing world of AI and what
they should be studying going

77
00:04:35,840 --> 00:04:38,600
forward.
As with all these episodes, you

78
00:04:38,600 --> 00:04:41,640
know, I think we could have
carried on for many hours.

79
00:04:41,640 --> 00:04:44,400
And I hope I have an opportunity
to do that again with Nathan and

80
00:04:44,400 --> 00:04:46,680
speak back to him in a few years
and see if some of the

81
00:04:46,680 --> 00:04:50,040
predictions were were correct.
But I certainly learned a lot

82
00:04:50,040 --> 00:04:52,960
from this conversation, and I
hope that you do too.

83
00:04:52,960 --> 00:04:55,960
So sit back and listen to this
conversation with Professor

84
00:04:55,960 --> 00:04:58,920
Nathan Cuts.
Well, thank you, Nathan, for for

85
00:04:58,920 --> 00:05:02,080
coming on this.
This has been on my sort of wish

86
00:05:02,080 --> 00:05:06,480
list to, to speak to you every
time I talk about machine

87
00:05:06,480 --> 00:05:09,880
learning related to engineering,
to CFD, to any of these

88
00:05:09,880 --> 00:05:12,040
problems.
You know, your name is very high

89
00:05:12,040 --> 00:05:18,040
up the list, some Seminole
pieces of work and arguably you,

90
00:05:19,040 --> 00:05:21,440
you've been working on this way
before.

91
00:05:21,480 --> 00:05:25,640
This was a sort of buzz word
topic, you know, a sort of sexy

92
00:05:25,640 --> 00:05:30,200
area to get into.
You know, where, where did you

93
00:05:30,200 --> 00:05:32,680
sort of start to get into this?
What, what was your career

94
00:05:32,680 --> 00:05:35,360
trajectory?
And when did you see this as

95
00:05:35,360 --> 00:05:39,880
like the missing piece to start
using some of these machine

96
00:05:39,880 --> 00:05:41,520
learning or reduced order
techniques?

97
00:05:42,120 --> 00:05:47,120
Yeah, so I thanks for having me
first of all, Neil to be here.

98
00:05:47,960 --> 00:05:49,640
I'm I'm excited to talk to you
as well.

99
00:05:49,640 --> 00:05:54,440
And, and, and I would say that
I, I, I suppose I got into it

100
00:05:54,440 --> 00:05:58,040
just out of pure interest in
sort of a very accidental way.

101
00:05:58,120 --> 00:06:04,320
I was working when I started as
a faculty late 90s, a while ago

102
00:06:04,320 --> 00:06:07,080
now.
And that was the beginning of my

103
00:06:07,080 --> 00:06:09,080
career.
I was working a lot in atomic

104
00:06:09,080 --> 00:06:12,000
and optical physics.
So I was doing theory for mode

105
00:06:12,000 --> 00:06:14,960
block lasers, lots of
computation, lots of modeling,

106
00:06:14,960 --> 00:06:17,480
integration of those two
together, trying to work with

107
00:06:17,480 --> 00:06:21,760
experimentalists.
And I really enjoyed that.

108
00:06:22,280 --> 00:06:29,120
But around the mid 2000s, what I
found was I, I couldn't get

109
00:06:29,120 --> 00:06:31,480
students interested in this
stuff.

110
00:06:31,800 --> 00:06:34,920
Like a lot of the students just
did not want to work on these,

111
00:06:35,800 --> 00:06:38,640
let's call them harder physics
problems, right, where you had

112
00:06:38,640 --> 00:06:43,040
to know some quantum and E&M.
They were very interested at the

113
00:06:43,040 --> 00:06:47,040
time in neuroscience, like that
was the hot applied math fields

114
00:06:47,040 --> 00:06:54,480
that everybody wanted to do.
And so around 2007, six and

115
00:06:54,480 --> 00:06:57,760
seven, I decided, you know,
look, having a hard time getting

116
00:06:57,760 --> 00:06:59,640
students.
I really like the work, but

117
00:06:59,640 --> 00:07:02,600
there's so many new interesting
areas.

118
00:07:02,600 --> 00:07:07,400
And I also had a little bit of,
I guess I'll just say it's an

119
00:07:07,400 --> 00:07:09,240
academic crisis.
I think everybody goes through

120
00:07:09,240 --> 00:07:11,280
it a little bit where you kind
of get a little bored with what

121
00:07:11,280 --> 00:07:14,240
you're doing.
You see that, yes, I could

122
00:07:14,240 --> 00:07:20,320
continue this for a long time in
my career or I take the risk and

123
00:07:20,320 --> 00:07:22,400
do something really new.
And of course, this is a little

124
00:07:22,400 --> 00:07:25,240
scary when you're more of a
senior person because you know,

125
00:07:25,360 --> 00:07:27,240
like you're you're like a
beginner again.

126
00:07:27,240 --> 00:07:32,360
I remember going to this
neuroscience workshop where

127
00:07:32,360 --> 00:07:35,520
clearly every grad student in
the room knew more than me about

128
00:07:35,520 --> 00:07:38,120
neuroscience.
But they out this odd doc, which

129
00:07:38,120 --> 00:07:42,080
like, well, he seems to know
something like he's not just an

130
00:07:42,080 --> 00:07:44,640
illiterate science person, but
on the other hand, doesn't seem

131
00:07:44,640 --> 00:07:46,960
to know much neuroscience.
And then, you know, to put

132
00:07:46,960 --> 00:07:50,440
yourself in that, I would say a
very vulnerable position was,

133
00:07:51,200 --> 00:07:54,960
was kind of what I did.
And I started actually in about

134
00:07:54,960 --> 00:07:59,720
2006 seven to do neuroscience.
And then also at that time

135
00:07:59,720 --> 00:08:02,920
seeing some of the data, I
thought, hey, there's these data

136
00:08:02,920 --> 00:08:06,080
analysis methods.
What if I started integrating

137
00:08:06,080 --> 00:08:09,160
that into sort of building
models?

138
00:08:09,960 --> 00:08:15,080
And so that was really the, the
initial part of that.

139
00:08:15,080 --> 00:08:18,960
I would say I got into sort of,
let's call it broadly, maybe I

140
00:08:18,960 --> 00:08:21,120
wouldn't, maybe you wouldn't
call it that now.

141
00:08:21,120 --> 00:08:23,720
It's actually the, the language
around what machine learning is

142
00:08:23,720 --> 00:08:26,920
and deep, you know, it's, it's
almost all deep learning now.

143
00:08:27,680 --> 00:08:31,800
Let's call, if you go back 2007
and eight, it wasn't neural

144
00:08:31,800 --> 00:08:34,960
networks yet, but I started
using a lot of these

145
00:08:34,960 --> 00:08:39,240
methodologies that came out of
the data-driven piece.

146
00:08:39,559 --> 00:08:43,240
And really part of the incentive
there was I saw that there was

147
00:08:43,240 --> 00:08:45,240
this great opportunity in
neuroscience where we didn't

148
00:08:45,240 --> 00:08:47,600
have first principle models,
right?

149
00:08:47,600 --> 00:08:51,520
Where you didn't have F equals
MA, you didn't have a Maxwell's

150
00:08:51,520 --> 00:08:53,880
equations.
You had data and neurons and

151
00:08:53,880 --> 00:08:57,240
people were making up models.
And but you're also looking at

152
00:08:57,520 --> 00:09:01,120
trying to build models that, you
know, populate population levels

153
00:09:01,120 --> 00:09:03,400
of neurons.
And we just started building

154
00:09:03,400 --> 00:09:08,320
these data-driven models back in
that time frame, 2000, 2008, 78.

155
00:09:08,800 --> 00:09:12,000
And I thought it was awesome.
I just really enjoyed the new

156
00:09:12,000 --> 00:09:14,080
direction.
And then I kind of just pivoted

157
00:09:14,080 --> 00:09:17,840
over to there and started
working in data-driven modeling.

158
00:09:19,720 --> 00:09:22,920
I started to teach a class on it
at the University of Washington.

159
00:09:22,920 --> 00:09:26,000
It was 2008.
So it was very early on.

160
00:09:27,920 --> 00:09:30,080
And mostly I just did it because
I, I liked it.

161
00:09:30,240 --> 00:09:34,440
I I didn't, I didn't see the
what was coming in 10 years,

162
00:09:34,440 --> 00:09:36,600
right.
I mean, it's, it's money.

163
00:09:36,600 --> 00:09:37,600
You know, you can always
bracken.

164
00:09:37,600 --> 00:09:41,680
Oh, I saw it.
But I mean, really I just did it

165
00:09:41,680 --> 00:09:44,440
because I was interested and I
thought it was very powerful.

166
00:09:45,320 --> 00:09:49,680
And I remember going to some of
these optics conferences because

167
00:09:49,680 --> 00:09:51,400
I started doing this and I'm
like, wait a minute, we can

168
00:09:51,400 --> 00:09:54,320
actually start doing some of
this to model lasers.

169
00:09:55,800 --> 00:09:57,680
And I'd show up these optics
conferences and they're like,

170
00:09:57,680 --> 00:10:02,200
what is this stuff you're doing?
Like there was like when you're

171
00:10:02,200 --> 00:10:04,920
talking about machine learning
integration into the modeling

172
00:10:04,920 --> 00:10:09,320
pipeline at that point, it was
it was so foreign to people.

173
00:10:09,320 --> 00:10:12,120
It was like this, you know, this
guy's lost his marbles.

174
00:10:12,120 --> 00:10:17,040
He does really falling off the
wagon in terms of he's way out

175
00:10:17,040 --> 00:10:19,280
there.
And then, you know, you advance

176
00:10:19,280 --> 00:10:23,000
a decade and then everybody is
using it, right.

177
00:10:24,120 --> 00:10:25,680
Yeah.
I'd love to say I saw that, but

178
00:10:25,680 --> 00:10:29,360
I, I just did it because I liked
it and turned out to be

179
00:10:29,360 --> 00:10:31,440
something that everybody started
doing.

180
00:10:31,440 --> 00:10:33,480
So that was actually my journey
into it.

181
00:10:34,240 --> 00:10:38,320
And then I built from there.
And now I'm just in fear of

182
00:10:38,320 --> 00:10:42,320
getting left behind because like
these things, they, what they

183
00:10:42,320 --> 00:10:45,080
could do with, you know, is, is
unbelievable.

184
00:10:45,080 --> 00:10:48,120
Like they work so much faster
than I ever did in my whole

185
00:10:48,160 --> 00:10:52,680
career.
That's funny.

186
00:10:52,720 --> 00:10:58,320
So how did you, I mean,
obviously the IT looked like

187
00:10:58,320 --> 00:11:03,560
2016 seventeen was like quite
big time.

188
00:11:03,560 --> 00:11:05,640
Obviously with with Steve as
well.

189
00:11:05,640 --> 00:11:08,000
Like what?
How did, how did the years come

190
00:11:08,000 --> 00:11:11,000
up to that sort of Seminole
piece of work?

191
00:11:11,400 --> 00:11:18,480
Yeah.
So I the academic year starting

192
00:11:18,480 --> 00:11:23,200
2012 I had brought on Steve is
my postdoc.

193
00:11:23,560 --> 00:11:27,960
He was at Princeton finishing up
and I was very fortunate to get

194
00:11:27,960 --> 00:11:32,040
somebody of that quality.
Partly it was there was a 2 body

195
00:11:32,040 --> 00:11:35,200
problem.
So I, I jointly hired Bing, his

196
00:11:35,200 --> 00:11:39,160
wife with Tom Daniel over in
biology and myself as we are.

197
00:11:39,960 --> 00:11:42,880
I didn't have enough money for
two postdocs out of money for 1

198
00:11:42,880 --> 00:11:45,240
1/2.
So I put half for Bing and got

199
00:11:45,240 --> 00:11:47,800
Tom to do another half and then
hired Steve on.

200
00:11:47,800 --> 00:11:50,840
And at that time I was already
doing some of the data-driven

201
00:11:50,840 --> 00:11:53,440
money, especially things like
dynamic mode decomposition.

202
00:11:53,440 --> 00:11:57,960
I had started working on already
in 20/12/2013.

203
00:11:58,880 --> 00:12:02,080
And so already was this
regression framework towards

204
00:12:02,080 --> 00:12:05,800
thinking about dynamical systems
and, and, and then it just

205
00:12:05,800 --> 00:12:10,840
turned out that it was just the
perfect timing with Steve and

206
00:12:10,840 --> 00:12:16,120
with an amazing group of
graduate students and post docs

207
00:12:16,680 --> 00:12:19,080
and also a little bit more open
space.

208
00:12:19,080 --> 00:12:22,200
You know, back then it wasn't
sort of nowadays it's almost

209
00:12:22,200 --> 00:12:26,280
like there's just so many people
working in this area that it's

210
00:12:26,880 --> 00:12:29,760
hard to gain any room.
But back then it was still

211
00:12:29,760 --> 00:12:33,040
pretty early on in the sciences.
You know, we have Imagenet,

212
00:12:33,040 --> 00:12:37,560
right 2014 that hadn't quite
made its transition over to the

213
00:12:37,560 --> 00:12:39,800
sciences.
I think that would more like

214
00:12:40,280 --> 00:12:43,600
20181920 like where they were
really getting into it.

215
00:12:44,040 --> 00:12:48,400
So we were early on had some
free space there and also had

216
00:12:48,400 --> 00:12:53,840
less rules I guess.
And we just started doing some

217
00:12:54,120 --> 00:12:56,520
stuff there.
And I think obviously Cindy came

218
00:12:56,520 --> 00:13:00,680
out of that time period and some
PD finders the dynamic mode

219
00:13:00,680 --> 00:13:05,640
decomposition for and then tying
all that together with what was

220
00:13:05,640 --> 00:13:07,480
happening also in reduced order
modeling.

221
00:13:07,480 --> 00:13:10,240
So it just was a really
fantastic time.

222
00:13:10,240 --> 00:13:13,400
And then Steve then took a
faculty position there in

223
00:13:13,400 --> 00:13:16,800
mechanical engineering and then
we continue to have a really

224
00:13:16,800 --> 00:13:21,600
great partnership with our
students and postdocs.

225
00:13:21,600 --> 00:13:25,280
And it was incredibly productive
in terms of starting to build,

226
00:13:25,360 --> 00:13:27,920
you know, these data-driven
models and trying to bring in

227
00:13:28,320 --> 00:13:32,960
machine learning overall into
the standard dynamical systems

228
00:13:32,960 --> 00:13:35,480
framework.
So maybe for people who are not

229
00:13:35,480 --> 00:13:39,760
so familiar, how would you, how
would you describe Cindy and the

230
00:13:39,760 --> 00:13:42,440
sort of POD methods?
You know what, what was the what

231
00:13:43,320 --> 00:13:46,400
was the core aim or thing that
you were trying to solve?

232
00:13:47,680 --> 00:13:49,960
Yeah.
So I think, I think there's

233
00:13:50,080 --> 00:13:52,320
there's two different aspects
here.

234
00:13:52,320 --> 00:13:58,600
So, so I always think about
physics as sort of, if you go

235
00:13:58,600 --> 00:14:02,760
back to like every engineer,
they got sort of a classic

236
00:14:02,960 --> 00:14:05,160
except for computer scientists.
If you want to consider them

237
00:14:05,160 --> 00:14:07,840
engineers, let's they're a
little different brand of

238
00:14:07,920 --> 00:14:11,360
engineer, but almost every other
engineer had to take statics and

239
00:14:11,360 --> 00:14:14,360
dynamics.
And then statics and dynamics.

240
00:14:14,360 --> 00:14:17,160
There was only one thing you
really did, which is you drew a

241
00:14:17,160 --> 00:14:21,200
free body diagram.
And then once you drew the right

242
00:14:21,200 --> 00:14:25,520
free body diagram, that was
really almost every homework set

243
00:14:25,520 --> 00:14:27,880
was just draw the right free
body diagram.

244
00:14:27,880 --> 00:14:31,240
Because once you draw the right
free body diagram, you could

245
00:14:31,240 --> 00:14:34,760
actually either sum of forces
equals 0, that's statics sum of

246
00:14:34,760 --> 00:14:39,400
forces equals not 0.
But if you didn't draw the right

247
00:14:39,400 --> 00:14:42,760
free body diagram, it was going
to be a really hard solve.

248
00:14:44,640 --> 00:14:48,240
And the way I think about that
in context of what we are trying

249
00:14:48,240 --> 00:14:52,160
to do, like even with POD or
these, you know, even what we do

250
00:14:52,160 --> 00:14:55,960
now with encoders is can I find
the right representation or the

251
00:14:55,960 --> 00:14:59,480
right coordinate system in which
to express my dynamics?

252
00:15:01,040 --> 00:15:06,160
And so, so that was already sort
of a philosophical idea of like,

253
00:15:06,160 --> 00:15:09,680
hey, we have all this data, you
know, it's high dimensional, it

254
00:15:09,680 --> 00:15:11,480
looks crazy.
But at the end of the day that

255
00:15:11,480 --> 00:15:15,040
we can find these some kind of
embedding where it's actually

256
00:15:15,040 --> 00:15:18,160
looks kind of simple maybe,
right.

257
00:15:18,560 --> 00:15:21,480
But then what's happening in
that space is now you have to

258
00:15:21,480 --> 00:15:23,840
have some kind of dynamical
system relationships.

259
00:15:23,840 --> 00:15:28,760
And so the Cindy part really
comes from this idea that if we

260
00:15:28,760 --> 00:15:33,120
look at our physics models, all
the physics models that we have

261
00:15:33,120 --> 00:15:38,960
are really essentially
expressions of relationships

262
00:15:38,960 --> 00:15:42,400
among derivatives, time and
space, right?

263
00:15:42,400 --> 00:15:49,160
And it's in most remarkable
compression of knowledge that

264
00:15:49,160 --> 00:15:52,960
we, I think we have ever
developed as humans, right?

265
00:15:52,960 --> 00:15:54,720
You know, you write down
Maxwell's equations.

266
00:15:54,720 --> 00:15:57,120
You have this, you fit it on the
T-shirt, right?

267
00:15:57,120 --> 00:16:00,160
You can order online on Amazon
this afternoon, you get it

268
00:16:00,160 --> 00:16:01,760
right.
And you have those Maxwell's

269
00:16:01,760 --> 00:16:04,920
equations on your T-shirt.
And what's amazing about it is

270
00:16:05,200 --> 00:16:10,240
it models, you know, all of our
Wi-Fi signals like coming, you

271
00:16:10,240 --> 00:16:15,560
know, radar, you know, radio
waves, the diversity of what

272
00:16:15,560 --> 00:16:20,160
you're modeling in this compact
representation, which is, is

273
00:16:20,160 --> 00:16:22,440
astounding.
Same thing with how we are able

274
00:16:22,440 --> 00:16:25,040
to get, you know, when think
about quantum mechanics, it's

275
00:16:25,040 --> 00:16:28,560
like, you know, 3 terms, right?
You have your dispersion term,

276
00:16:28,560 --> 00:16:31,080
your potential and, and your
time derivative.

277
00:16:31,080 --> 00:16:35,840
And it's, so I think the
motivation for Cindy was this

278
00:16:35,840 --> 00:16:42,440
idea that we've had tremendous
success in history with these

279
00:16:43,640 --> 00:16:47,560
compact representations of
relationships among derivatives.

280
00:16:48,240 --> 00:16:52,240
And so the Cindy algorithm goes
directly after that says, well,

281
00:16:52,600 --> 00:16:56,280
there's lots of derivatives and
derivatives relationships.

282
00:16:57,480 --> 00:17:00,960
When you look at this data, how
few of these derivative

283
00:17:01,080 --> 00:17:04,480
relationships can you use that
actually fit the data?

284
00:17:05,119 --> 00:17:08,599
And that's where this idea of
having a library of potential

285
00:17:08,599 --> 00:17:13,319
candidates and, and then
sparsely regressed to just a few

286
00:17:13,319 --> 00:17:19,160
terms came out of.
And it's kind of a philosophical

287
00:17:19,160 --> 00:17:20,920
thing.
So I think like physicists like

288
00:17:20,920 --> 00:17:22,640
this idea.
It's not clear that computer

289
00:17:22,640 --> 00:17:25,400
scientists like this idea.
Like computer scientists might,

290
00:17:25,800 --> 00:17:28,560
I've heard this directly.
They feel like, well, that's

291
00:17:28,560 --> 00:17:32,200
your, that's your 20th century
crutch of physics.

292
00:17:32,200 --> 00:17:34,600
Like, that's because that's
what, you know what, let's let

293
00:17:34,600 --> 00:17:37,080
the computer figure out.
Right.

294
00:17:37,600 --> 00:17:43,760
So it's still a tension point,
but as we talk further about

295
00:17:43,760 --> 00:17:50,080
some of the where physics AI is
going, I think one of the things

296
00:17:50,080 --> 00:17:54,400
that's really incredibly
valuable to consider is that

297
00:17:55,160 --> 00:17:58,960
governing equations so far, I,
I, I think this is true.

298
00:17:58,960 --> 00:18:00,960
I mean, it's hard to, you know,
these are statements and of

299
00:18:00,960 --> 00:18:04,680
opinions, I guess.
But, but really historically, we

300
00:18:04,680 --> 00:18:07,440
can see that they've been the
most incredible engines of

301
00:18:07,440 --> 00:18:12,320
extrapolation in predictions
where I've never had data about

302
00:18:12,320 --> 00:18:15,120
something that might happen.
You know, if I go over there in

303
00:18:15,120 --> 00:18:18,960
parameter space, my model
predicts like this bifurcation,

304
00:18:19,320 --> 00:18:23,280
like the change of the system.
And then you can start doing an

305
00:18:23,280 --> 00:18:26,640
experiment over there and
validate that, right?

306
00:18:26,640 --> 00:18:30,480
Whereas machine learning, it's
very difficult to predict

307
00:18:30,480 --> 00:18:34,240
outside of if you don't have
data there, like you're just not

308
00:18:34,240 --> 00:18:37,040
going to get it right.
So governing equations still, at

309
00:18:37,040 --> 00:18:41,440
least in my heart, hold massive
value because of this

310
00:18:41,440 --> 00:18:45,040
extrapolation capability that
we, I think are going to still

311
00:18:45,040 --> 00:18:48,600
need in the future.
Yeah, that that's a very good

312
00:18:49,640 --> 00:18:53,640
transition point, I guess.
Well, first of all, I'd be

313
00:18:53,640 --> 00:18:56,560
interested to how you would
describe before we get on to the

314
00:18:57,120 --> 00:19:01,480
maybe today's debate on machine
learning methods and surrogates

315
00:19:01,480 --> 00:19:06,680
and, you know, the influence of
physics versus data-driven, some

316
00:19:06,680 --> 00:19:11,920
more skeptics, you know, we'll,
we'll sort of say, ah, this

317
00:19:11,920 --> 00:19:15,240
whole fuzz now about machine.
It's just basically reduced

318
00:19:15,240 --> 00:19:17,160
order modelling.
We've been doing it, you know,

319
00:19:17,480 --> 00:19:21,800
for for ages.
How would you define the

320
00:19:21,800 --> 00:19:26,440
differences and the similarities
between sort of those earlier

321
00:19:26,440 --> 00:19:30,680
reduced order modelling sort of
techniques and today with the

322
00:19:30,680 --> 00:19:35,120
more sort of transformer based
neural networks, what we would

323
00:19:35,120 --> 00:19:38,680
call sort of machine learning in
a very broad term?

324
00:19:39,880 --> 00:19:45,840
Yeah, so early days of reduced
order modeling, I think there

325
00:19:45,840 --> 00:19:49,240
was still this idea that these
POD modes, you know you do the

326
00:19:49,240 --> 00:19:51,720
SVD, you have these structures
that come out.

327
00:19:52,080 --> 00:19:57,840
I think it's still allowed for
some flexibility in the idea of

328
00:19:57,840 --> 00:20:02,040
an of interpretation, right.
So you you would have things

329
00:20:02,040 --> 00:20:06,160
that were coming out that you
would argue made sense, like so

330
00:20:06,160 --> 00:20:09,160
for instance even.
POD modes of flow around the

331
00:20:09,160 --> 00:20:13,320
cylinder, you look at it and go
like I can explain what that is

332
00:20:13,520 --> 00:20:17,960
and and there was something very
comforting about that like to

333
00:20:17,960 --> 00:20:21,120
have sort of in some sense a
basis.

334
00:20:21,120 --> 00:20:26,040
And so I think for us, if you
look at the 20th century applied

335
00:20:26,040 --> 00:20:30,480
math literature and what the
impact was, we did a lot of

336
00:20:30,480 --> 00:20:35,760
expansions and bases or a
transform was a workhorse.

337
00:20:35,760 --> 00:20:38,120
We could say look in signal
processing, right?

338
00:20:38,120 --> 00:20:42,920
Like just do a Fourier transform
process everything in the signal

339
00:20:42,920 --> 00:20:44,880
domain come back.
And we had this idea of

340
00:20:44,880 --> 00:20:47,600
interpretability built in
because we kind of knew what

341
00:20:47,600 --> 00:20:51,120
cosines and Sonic were.
We followed that up even with

342
00:20:51,120 --> 00:20:54,880
wavelets.
So we and even special functions

343
00:20:54,880 --> 00:20:57,160
which were sort of like in
mathematical physics, the

344
00:20:57,160 --> 00:21:01,200
foundation of the 1950s and 60s
for doing analysis of systems,

345
00:21:02,040 --> 00:21:05,080
it was this idea that there is a
basis functions.

346
00:21:05,440 --> 00:21:08,320
In other words, again, I would
call this a nice coordinate

347
00:21:08,320 --> 00:21:10,360
system that's sort of
interpretable to you.

348
00:21:10,360 --> 00:21:13,080
So you can project everything
into this, do your work in

349
00:21:13,080 --> 00:21:18,040
there, you feel comfortable with
it because it's you interpret it

350
00:21:18,280 --> 00:21:20,760
and then you can come back and
do your and your analysis.

351
00:21:20,760 --> 00:21:22,720
And by the way, we're very
successful with this, right?

352
00:21:23,840 --> 00:21:26,760
And largely we constrained
ourselves in the Fifties, 60s

353
00:21:26,760 --> 00:21:32,720
and 70s, even up to the 80s to
linear problems, because then we

354
00:21:32,720 --> 00:21:36,320
could use superposition, we
could have solutions which were

355
00:21:36,320 --> 00:21:38,120
linear combinations of these
things.

356
00:21:39,120 --> 00:21:42,280
And then I think at the end of
the 80s, of course, in the early

357
00:21:42,280 --> 00:21:45,120
90s, we had the first
computational revolution, which

358
00:21:45,120 --> 00:21:50,320
was, oh, I could just throw that
PD on the computer and it's good

359
00:21:50,320 --> 00:21:55,520
ties and start simulating.
And so, so I feel like when I

360
00:21:55,520 --> 00:21:59,560
went to grad school, I was in
grad school in 1990 to 94 and I

361
00:21:59,560 --> 00:22:02,440
feel like I was the first
generation where there was an

362
00:22:02,440 --> 00:22:06,280
expectation you're going to
simulate PDS and ODS in your

363
00:22:06,280 --> 00:22:12,240
thesis somewhere.
At that point we had now, you

364
00:22:12,240 --> 00:22:14,520
know, desktop computers in the
lab.

365
00:22:14,840 --> 00:22:19,320
It wasn't like there's some
central computer, a deck machine

366
00:22:19,320 --> 00:22:22,400
somewhere in a court, you know,
that you run jobs on.

367
00:22:22,400 --> 00:22:24,760
This is now like, oh, it's just
sitting here in the office and

368
00:22:24,760 --> 00:22:26,920
you're interacting and
programming with it.

369
00:22:27,720 --> 00:22:32,080
And so for the first time, we
are actually starting able to

370
00:22:32,080 --> 00:22:36,160
solve non linear PD ES because
we could just numerically

371
00:22:36,840 --> 00:22:39,000
simulate them.
And of course, it's very

372
00:22:39,000 --> 00:22:41,400
interesting because I think the
faculty back then were like, you

373
00:22:41,400 --> 00:22:43,560
know, these kids, they don't
know what they're doing,

374
00:22:44,360 --> 00:22:46,760
understand physics yet they're
doing, you know, they're

375
00:22:46,760 --> 00:22:50,160
claiming all this stuff.
And, and of course, it's sort of

376
00:22:50,160 --> 00:22:52,880
true, right, because they were
used to solving things, really

377
00:22:52,880 --> 00:22:55,800
understanding the problem at
some fundamental level and

378
00:22:56,200 --> 00:23:00,120
especially doing lots of
asymptotic reductions in two

379
00:23:00,120 --> 00:23:03,400
corners of parameter space where
you could linear eyes and say

380
00:23:03,400 --> 00:23:05,920
quite a bit, right?
That's, that's almost the whole

381
00:23:06,160 --> 00:23:09,800
fluid mechanics mantra.
It's like, look over here, we

382
00:23:09,800 --> 00:23:13,480
have JFM to tell us what's going
over there on, right?

383
00:23:15,000 --> 00:23:19,400
But then we had that simulation
capability and then and then we

384
00:23:19,400 --> 00:23:23,320
start saying, OK, but so we
still had some interpretation

385
00:23:23,320 --> 00:23:26,600
like I'm expanding in 40 basis
or I'm doing finite differences,

386
00:23:27,080 --> 00:23:29,480
my elements.
And then, and then from there,

387
00:23:29,480 --> 00:23:32,280
from those simulations, then we
started a new basis, which was

388
00:23:32,280 --> 00:23:34,320
the POD, right?
Which is like, oh, let's just

389
00:23:34,560 --> 00:23:36,560
build basis directly from the
SVD.

390
00:23:36,840 --> 00:23:40,320
So we still felt it was very
morally equivalent to what we

391
00:23:40,320 --> 00:23:44,640
were kind of doing for.
And then the, the transformation

392
00:23:44,640 --> 00:23:48,520
now is sometimes it's just
really hard to get your head

393
00:23:48,520 --> 00:23:50,400
around what happened in this
thing.

394
00:23:50,800 --> 00:23:53,800
Like I transformer and then I
have an attention layer and I

395
00:23:53,800 --> 00:23:58,120
got right, I got, I got this
decoder that's then being

396
00:23:58,120 --> 00:23:59,640
modulate.
I mean, right.

397
00:23:59,680 --> 00:24:02,360
We, we could see some of the
sophistication and some of the,

398
00:24:02,360 --> 00:24:05,360
some of the structures are kind
of simple, but more and more

399
00:24:05,360 --> 00:24:11,320
they're becoming fairly complex
and it's really hard for us to

400
00:24:11,320 --> 00:24:14,400
decipher what's happening there.
But you can't argue often with

401
00:24:14,840 --> 00:24:17,680
how well it works, right?
You get the results and you're

402
00:24:17,680 --> 00:24:19,720
going like, yeah, but I can't
beat this.

403
00:24:21,200 --> 00:24:24,280
So I have to, you know, I want
to use it.

404
00:24:24,400 --> 00:24:27,960
And by the way, I started to
appreciate computer scientists a

405
00:24:27,960 --> 00:24:31,440
little bit, quite a bit.
Because at first, you know, I

406
00:24:31,440 --> 00:24:34,280
was like that old grumpy
professor, which is like these

407
00:24:34,280 --> 00:24:36,560
computer science kids, they
don't know any physics.

408
00:24:36,560 --> 00:24:40,480
They just go do stuff and they
get a, they get a score and they

409
00:24:40,480 --> 00:24:43,560
got a cross validation score and
they move on with their lives,

410
00:24:43,560 --> 00:24:48,520
right.
The flip side now is I have

411
00:24:48,520 --> 00:24:55,720
grown to absolutely love the
fearlessness of a, you're a

412
00:24:55,720 --> 00:24:59,480
science grad student.
They just like try stuff not

413
00:24:59,480 --> 00:25:03,040
encumbered by all this physics
knowledge.

414
00:25:03,040 --> 00:25:07,440
Like, you know, like if I look
at this typical physics student

415
00:25:07,440 --> 00:25:10,920
or engineering or applied math,
they kind of are, they're

416
00:25:10,920 --> 00:25:13,520
prejudiced by their what they
learned.

417
00:25:13,520 --> 00:25:17,600
So they're like, they would try
to make smart things structures,

418
00:25:17,600 --> 00:25:20,160
whereas the computer scientists
are just like, just try stuff.

419
00:25:20,160 --> 00:25:22,560
What if we did this?
What if we did this?

420
00:25:22,560 --> 00:25:28,080
And and that fearlessness has
LED them to unbelievable

421
00:25:28,080 --> 00:25:31,360
results.
And I think it's going to be

422
00:25:31,360 --> 00:25:34,280
left to the applied
mathematicians and engineers and

423
00:25:34,280 --> 00:25:38,680
theory people to kind of pick up
the pieces to try to understand.

424
00:25:38,680 --> 00:25:41,320
OK, but what did you what OK in
here?

425
00:25:41,320 --> 00:25:44,360
I we can now got to dissect this
to see what you actually learned

426
00:25:44,360 --> 00:25:45,960
here, because somehow it is
working.

427
00:25:45,960 --> 00:25:47,280
I'm going to give you credit for
that.

428
00:25:47,280 --> 00:25:50,560
But I'd really love to
understand what this thing's

429
00:25:50,840 --> 00:25:52,880
actually learning about the
physics, right?

430
00:25:52,880 --> 00:25:55,160
And that and I think that's kind
of what big next steps that's

431
00:25:55,160 --> 00:25:56,760
going to happen in the
community.

432
00:25:56,760 --> 00:26:00,080
So you'd say almost the and I've
I've heard similar things.

433
00:26:00,080 --> 00:26:06,040
The interpretability is the key
bit like how does this work?

434
00:26:06,040 --> 00:26:11,080
Why does it work?
What is it that is the knob that

435
00:26:11,080 --> 00:26:13,720
has changed that you know, how
do you do?

436
00:26:13,800 --> 00:26:19,240
And I guess like you say, it's
all good when it works, but when

437
00:26:19,240 --> 00:26:23,120
it doesn't work, it's like why
didn't it work is the question.

438
00:26:23,160 --> 00:26:25,680
Yeah.
And at least in our standard

439
00:26:25,680 --> 00:26:28,880
physics modelling, right, we
always had some recourse to like

440
00:26:29,320 --> 00:26:32,760
there was just a few parameters
we could sort of, we could

441
00:26:32,760 --> 00:26:36,200
actually diagnose why it broke
or, or, or went through a

442
00:26:36,200 --> 00:26:38,360
bifurcation or there was a
transition.

443
00:26:38,640 --> 00:26:44,160
Like we had a way understanding
what would might happen where in

444
00:26:44,160 --> 00:26:47,840
these things.
It's a little bit like, I don't

445
00:26:47,840 --> 00:26:50,440
know what kind of car you drive,
but like when something breaks

446
00:26:50,440 --> 00:26:55,280
there, you just automatically
have to go to a mechanic because

447
00:26:55,280 --> 00:26:58,280
he's caught.
The cars are so complex now that

448
00:26:58,280 --> 00:27:01,920
you have to have specialist fix
it where, you know, I think, you

449
00:27:01,920 --> 00:27:05,280
know, when I was a kid, I think
there's a lot of people that fix

450
00:27:05,280 --> 00:27:09,040
their own cars, right?
It was a little bit more

451
00:27:09,040 --> 00:27:11,920
interpretable what was happening
in that engine.

452
00:27:12,280 --> 00:27:15,640
But now it's like, no, no, no,
that everything's run off a

453
00:27:15,640 --> 00:27:18,600
bunch of electronic components
with all these different pieces.

454
00:27:20,240 --> 00:27:21,680
And I think of it like that,
like that.

455
00:27:21,680 --> 00:27:25,360
And when they break, you just
have no idea and started fixing

456
00:27:25,360 --> 00:27:27,080
it.
Maybe you just train another

457
00:27:27,080 --> 00:27:30,080
neural network or the different
structure that doesn't break

458
00:27:30,080 --> 00:27:31,880
there.
Yeah, Yeah, Yeah.

459
00:27:33,120 --> 00:27:35,640
I mean, how, how do you see then
today because that that just

460
00:27:35,640 --> 00:27:39,280
seemed to be the challenge of
our time, doesn't it that you

461
00:27:39,280 --> 00:27:43,040
are, I mean, I guess out of
distribution versus in

462
00:27:43,040 --> 00:27:46,160
distribution the train on a
bunch of cars.

463
00:27:47,080 --> 00:27:51,960
Is it, is it just the case that
that is a essentially not

464
00:27:52,280 --> 00:27:56,360
unsolvable, but a problem that
can only really be addressed

465
00:27:56,360 --> 00:27:59,680
just by having more data that
makes that out of distribution

466
00:27:59,680 --> 00:28:02,240
in distribution.
And so the solution is just keep

467
00:28:02,240 --> 00:28:05,520
building bigger and bigger
models with more and more data?

468
00:28:07,600 --> 00:28:11,800
Or do you feel that that is
essentially inefficient and and

469
00:28:12,120 --> 00:28:15,600
not a scalable method and
therefore you need some ability

470
00:28:15,600 --> 00:28:21,640
to, as you say you know, use
some modeling knowledge or

471
00:28:21,640 --> 00:28:25,800
physics knowledge to be able to
model past the data?

472
00:28:26,200 --> 00:28:34,440
Yeah, I OK, so this question is
hard to answer for, for for one

473
00:28:34,440 --> 00:28:39,040
reason, which is not scientific
at all, which is just money.

474
00:28:40,360 --> 00:28:45,040
When you look at this product,
you know, Project Prometheus for

475
00:28:45,040 --> 00:28:49,680
instance, that Bezos is involved
with and you look at their what

476
00:28:49,680 --> 00:28:55,680
6.2 billion.
Now at some point you ask

477
00:28:55,680 --> 00:28:59,320
questions like it's OK, even now
being at Autodesk.

478
00:28:59,320 --> 00:29:01,400
So first of all, let's if I, if
I go back to when I was a

479
00:29:01,400 --> 00:29:07,480
faculty, the ability to get data
like really say we need to

480
00:29:07,480 --> 00:29:10,040
collect this data.
It's like, oh, wait, you're

481
00:29:10,120 --> 00:29:11,880
going to have to write some
massive grants.

482
00:29:11,880 --> 00:29:13,960
There's probably almost no way
you're going to get the kind of

483
00:29:13,960 --> 00:29:16,160
grant levels you need to collect
the data you would need.

484
00:29:16,160 --> 00:29:19,680
So you have, you're already
constrained saying I've got to

485
00:29:19,680 --> 00:29:22,360
outsmart this and got to figure
out how to do this.

486
00:29:23,240 --> 00:29:27,520
Now you move to a company and
now I'm at Autodesk.

487
00:29:27,520 --> 00:29:31,760
But even then you're like, yeah,
but you it's, it's not clear.

488
00:29:33,080 --> 00:29:35,760
There's a few companies, there's
a there's the trillion dollar

489
00:29:35,760 --> 00:29:40,920
companies, right?
That potentially, if they desire

490
00:29:40,920 --> 00:29:44,840
to do so, could potentially set
up the architecture to just

491
00:29:44,960 --> 00:29:50,120
collect the data, spend the
billions to do things like this

492
00:29:50,120 --> 00:29:53,680
and and do that.
Now, that's a brute force

493
00:29:53,680 --> 00:29:59,400
approach, but I think the win is
so big that they might just say,

494
00:29:59,400 --> 00:30:02,520
yeah, we could spend $10 billion
because it's going to make us a

495
00:30:02,520 --> 00:30:07,240
trillion.
Like very few places that can do

496
00:30:07,240 --> 00:30:08,320
it.
Like, but even Project

497
00:30:08,320 --> 00:30:11,720
Prometheus might have the kind
of cash resources because my

498
00:30:11,720 --> 00:30:17,280
normal answer would be like, it
is completely, I think,

499
00:30:19,920 --> 00:30:23,400
intractable to get the kind of
data we need to do something

500
00:30:23,680 --> 00:30:25,240
with the science problems we
need.

501
00:30:26,400 --> 00:30:30,720
But when you have 6.2 billion,
right, and maybe more, I don't

502
00:30:30,720 --> 00:30:34,200
know, like maybe you go like,
you know, we can do it.

503
00:30:34,560 --> 00:30:38,240
Like we're going to spend $3
billion to collect the data,

504
00:30:38,400 --> 00:30:40,640
right?
That's higher investment in the

505
00:30:40,760 --> 00:30:43,200
in the data.
It isn't is that money is to get

506
00:30:43,200 --> 00:30:46,120
the data and then now we can
build whatever the foundation

507
00:30:46,120 --> 00:30:50,960
model we need or so.
So that one, I think the jury's

508
00:30:50,960 --> 00:30:53,000
still out.
I so, you know, it goes right

509
00:30:53,000 --> 00:30:56,440
back to, you know, Richard
Sutton's right, the the bitter

510
00:30:56,440 --> 00:31:01,120
lesson issue, which is so
Sutton, I think is right.

511
00:31:01,120 --> 00:31:04,520
If you have enough data, right,
you really, if you can collect

512
00:31:04,520 --> 00:31:09,600
enough data in an area, it's so
far what we've seen, it's almost

513
00:31:10,160 --> 00:31:12,480
these machine learning, deep
learning algorithms and

514
00:31:12,480 --> 00:31:15,480
structures are, are basically
unbeatable.

515
00:31:17,040 --> 00:31:19,920
But the question is, can you
really do this?

516
00:31:19,920 --> 00:31:22,960
I, I love also Maxwelling wrote
a response to this, which was

517
00:31:22,960 --> 00:31:25,240
well, yeah.
But a lot of times what we're

518
00:31:25,240 --> 00:31:28,480
trying to do scientifically is
extrapolate.

519
00:31:28,880 --> 00:31:31,240
We're trying to build a new
technology.

520
00:31:31,760 --> 00:31:36,240
So even if I collect all this
data, the the goal is to give me

521
00:31:36,240 --> 00:31:38,760
a great direction of like
actually, I think the future

522
00:31:38,760 --> 00:31:41,000
technologies over there, I'd
have no data there Now.

523
00:31:41,000 --> 00:31:44,840
Maybe I could set up a pipeline,
have enough money to to build

524
00:31:44,840 --> 00:31:47,760
the scaffolding for the data to
collect on the way out there.

525
00:31:49,080 --> 00:31:52,200
But, you know, this is what I
learned in my time at McLaren

526
00:31:52,200 --> 00:31:57,800
too, is the answer.
The fast car is an extrapolation

527
00:31:57,840 --> 00:31:59,600
that's far away from where
you're starting.

528
00:31:59,840 --> 00:32:02,680
Actually, far away is a little
bit of exaggeration.

529
00:32:03,480 --> 00:32:07,160
Tenths of second.
Yeah, well, it feels far away

530
00:32:07,160 --> 00:32:09,560
when you're trying to like, you
know, the difference between

531
00:32:09,560 --> 00:32:12,120
winning the world title.
Like, you know, which McLaren

532
00:32:12,120 --> 00:32:14,560
did while I was the time I was
there, they were like had them

533
00:32:14,560 --> 00:32:19,520
MCL 3839, which were openers.
You know, you're, you're working

534
00:32:19,520 --> 00:32:23,320
at the margins of these updates,
they're trying to get you 2/10

535
00:32:23,320 --> 00:32:25,480
of a second, a 10th of a second,
3/10 of a second.

536
00:32:26,400 --> 00:32:29,040
But you're also going into a
regime that there is no data.

537
00:32:29,880 --> 00:32:33,560
I have to actually get this data
from simulations, do some wind

538
00:32:33,560 --> 00:32:35,920
tunnel to validate that it's
there.

539
00:32:37,040 --> 00:32:44,040
So that, so whether that's
viable, you know, they have $150

540
00:32:44,040 --> 00:32:47,480
million a year budget.
I don't know, whatever the app

541
00:32:47,480 --> 00:32:48,840
is, it's a little bit higher
than that.

542
00:32:48,840 --> 00:32:52,600
Now on that budget, there's no
way you could do this, right?

543
00:32:52,600 --> 00:32:56,280
Just collect just money there to
collect the data you'd need.

544
00:32:57,720 --> 00:33:00,320
So, so it's, it's, it's a really
interesting question.

545
00:33:00,320 --> 00:33:03,960
I, I, there's a group of people
that fundamentally feel I will

546
00:33:03,960 --> 00:33:06,800
just spend so much money to
collect the data so that I can

547
00:33:06,800 --> 00:33:12,200
make it just like language.
But physics isn't I physics

548
00:33:12,200 --> 00:33:14,920
isn't language either.
So if it, so there's still a bit

549
00:33:14,920 --> 00:33:20,000
here that we have to figure out.
So my own personal belief is, or

550
00:33:20,000 --> 00:33:25,240
at least what I'm going to do is
embed physics type knowledge

551
00:33:25,480 --> 00:33:27,760
into these neural networks so we
can do this much more

552
00:33:27,760 --> 00:33:34,760
efficiently anyway.
Yeah, no, I I've had this, you

553
00:33:34,760 --> 00:33:37,480
know, debate many time and you
know, that was part of the

554
00:33:37,480 --> 00:33:40,880
reason for doing that paper with
Johannes and Sid was to try and

555
00:33:40,880 --> 00:33:45,280
like try and project out some of
this economies of things.

556
00:33:45,640 --> 00:33:49,560
But we still, I think even
writing that paper, we, we still

557
00:33:49,560 --> 00:33:52,200
got to the end.
And you know, I can't, I can't

558
00:33:52,200 --> 00:33:54,320
speak to the others, but at
least for myself, we still

559
00:33:54,320 --> 00:33:57,280
didn't really resolve the
fundamental challenge, which is

560
00:33:58,280 --> 00:34:02,240
even with all the predictions we
made, they weren't reaching the

561
00:34:02,240 --> 00:34:08,280
accuracy that was necessarily
required.

562
00:34:08,320 --> 00:34:14,679
And the some of that accuracy is
very deep within the domain

563
00:34:14,679 --> 00:34:19,400
knowledge of those companies.
And so a Formula One team would

564
00:34:19,400 --> 00:34:23,880
argue that they are the only
ones who deeply know how to, you

565
00:34:23,880 --> 00:34:27,120
know, set up a a case and get it
to give the right accuracy.

566
00:34:27,120 --> 00:34:31,120
If you just brute force even
take an LES model and try and

567
00:34:31,120 --> 00:34:37,000
run it even with something like
a billion cells, you could quite

568
00:34:37,000 --> 00:34:39,520
easily get it very wrong, very
wrong.

569
00:34:39,920 --> 00:34:43,880
And even if you gave it 10
billion doesn't actually, but

570
00:34:43,880 --> 00:34:46,560
you know you'd you almost then
have to extrapolate it all the

571
00:34:46,560 --> 00:34:50,800
way to the end, which is to do a
complete DNS or something.

572
00:34:50,800 --> 00:34:53,080
But then what if you don't know
the porosity values of the

573
00:34:53,080 --> 00:35:01,640
radiator that they're using?
So this I kind of feel that in

574
00:35:01,640 --> 00:35:04,560
one sense I agree with you that
this does feel a little bit like

575
00:35:04,560 --> 00:35:08,440
just a data issue and then, you
know, a bit like ChatGPT and

576
00:35:08,440 --> 00:35:11,240
others to show that, you know,
just put a lot of money behind

577
00:35:11,240 --> 00:35:17,640
something you can solve it.
But that I guess the difference

578
00:35:17,640 --> 00:35:20,400
I would say, which is why I
think I would agree more with

579
00:35:20,400 --> 00:35:25,440
your approach is it makes sense
when the opportunity of the

580
00:35:25,440 --> 00:35:28,320
market is trillions of dollars,
which is basically enterprise

581
00:35:28,320 --> 00:35:34,480
AI, but CFDI don't think it's a
trillion dollar market.

582
00:35:34,480 --> 00:35:39,000
It's probably more than the 10s
of billions maybe.

583
00:35:39,000 --> 00:35:44,120
So if you spend 10s of billions
that makes you assume that you

584
00:35:44,120 --> 00:35:48,000
can take the entire market,
which you know so from a like

585
00:35:48,000 --> 00:35:55,280
return on investment that to me
is a harder like balance.

586
00:35:55,520 --> 00:35:59,920
Well, and, and it's also
interesting for like for some of

587
00:35:59,920 --> 00:36:01,280
these companies, we need to do
science.

588
00:36:01,280 --> 00:36:04,080
And I think you hit on such a
critical issue, which is this

589
00:36:04,080 --> 00:36:08,080
idea of tolerances.
There's you're always going to

590
00:36:08,080 --> 00:36:12,440
have to have, I think, recourse
to a physical manifestation of

591
00:36:12,440 --> 00:36:16,440
what you're doing.
Because if you're if OK.

592
00:36:16,720 --> 00:36:21,000
So think about world models a
little bit, right?

593
00:36:21,680 --> 00:36:23,960
They're just fakes.
They look cool.

594
00:36:25,760 --> 00:36:29,760
But you know, people who build
world models, no lives are

595
00:36:29,760 --> 00:36:32,160
dependent upon it.
It's not like by going here I

596
00:36:32,160 --> 00:36:36,920
could I could die, But if you
have these people develop in the

597
00:36:36,920 --> 00:36:39,280
world model, I want you to
design an airplane for me.

598
00:36:39,360 --> 00:36:41,360
Like there's no way I'm getting
on that airplane.

599
00:36:43,320 --> 00:36:46,000
So there's there's there's sort
of these zero failure

600
00:36:46,000 --> 00:36:50,440
environments where the site
tolerance is absolutely king.

601
00:36:50,560 --> 00:36:52,920
And even I think Bezos is quite
interested in this.

602
00:36:53,080 --> 00:36:55,640
You know, again, I'm just
throwing out this project

603
00:36:55,640 --> 00:36:58,120
Prometheus, because I'm just
the, the staggering amount of

604
00:36:58,120 --> 00:37:01,480
money that's been invested in it
in the space.

605
00:37:02,920 --> 00:37:05,840
But you know, he's interested in
space travel and rockets and

606
00:37:05,840 --> 00:37:11,040
you're like, OK, so those are 0
failure environments.

607
00:37:11,920 --> 00:37:19,320
And I, you know, when, when a
prompt goes bad like unchat GPT

608
00:37:19,320 --> 00:37:23,000
or Gemini or quad, nobody gets
hurt, right?

609
00:37:23,000 --> 00:37:24,240
Really.
I mean, at the end of the day,

610
00:37:24,240 --> 00:37:27,720
it's like, Oh, it didn't quite
get the I didn't die cause of

611
00:37:27,720 --> 00:37:30,640
it.
But like when you're going to do

612
00:37:30,640 --> 00:37:33,560
something like a rocket, right,
where people's lives are at

613
00:37:33,560 --> 00:37:36,520
stake and you have to have the
tolerances or else it's going to

614
00:37:38,040 --> 00:37:42,440
that's where I think there's
just such a an amazing pressure,

615
00:37:42,760 --> 00:37:48,720
right, to get this right.
And and it's not clear to me how

616
00:37:48,720 --> 00:37:52,320
AI does that, right.
It could be that agentic systems

617
00:37:52,320 --> 00:37:54,600
will learn how to do like
that's.

618
00:37:54,680 --> 00:37:56,960
I think we can certainly
program.

619
00:37:56,960 --> 00:38:00,160
I think agentic systems to say
now that I got most the answer

620
00:38:00,160 --> 00:38:06,840
right, my agents are all about
doing the tolerance checking and

621
00:38:06,840 --> 00:38:09,240
figuring out what experiments
have to be run.

622
00:38:10,360 --> 00:38:12,800
Because in my view, I think
really what happens is I think

623
00:38:12,800 --> 00:38:15,720
the agenic system should come
back to you and say, hey, Neil,

624
00:38:16,400 --> 00:38:18,720
we need to do some experiments
because right now I'm starting

625
00:38:18,720 --> 00:38:21,840
to hallucinate or at least I'm
uncertain my models.

626
00:38:23,000 --> 00:38:25,200
And of course it they're not
going to actually talk to you.

627
00:38:25,200 --> 00:38:27,720
They're going to actually talk
to their robot friends and say,

628
00:38:27,720 --> 00:38:31,600
can you do this experiment?
Yeah, The kind of result, you

629
00:38:31,600 --> 00:38:34,160
know, I need you to do it under
these kind of load conditions.

630
00:38:34,160 --> 00:38:36,960
So they can, once I get that
back, I can proceed along with

631
00:38:36,960 --> 00:38:41,680
my iteration.
So I think, I think that's going

632
00:38:41,680 --> 00:38:45,520
to be key for us somewhere along
the way, right.

633
00:38:45,520 --> 00:38:49,080
But this tolerance idea is we
still don't have a proof of, and

634
00:38:49,080 --> 00:38:51,040
it's not good enough.
Just have a world model that

635
00:38:51,040 --> 00:38:57,320
looks cool and looks right.
Yeah, that's, yeah.

636
00:38:57,320 --> 00:39:01,520
I think that's fundamentally the
also I guess the practical

637
00:39:01,520 --> 00:39:06,000
challenge of this where I feel
probably individual companies

638
00:39:06,000 --> 00:39:10,200
will, will still play more of a
role in this.

639
00:39:11,000 --> 00:39:14,400
You know, if you're a large
aircraft manufacturer, I would

640
00:39:14,400 --> 00:39:18,680
imagine that at least in the,
you know, short to medium term,

641
00:39:18,680 --> 00:39:22,720
you're going to be the one that
is most likely using some of

642
00:39:22,720 --> 00:39:27,040
your data to be fine-tuned this
or, or you'll be still do it.

643
00:39:27,040 --> 00:39:30,160
You know, it's, it's, I don't
think you completely outsource

644
00:39:30,160 --> 00:39:39,080
this to, to A, to a separate,
you know, entity that that's

645
00:39:39,200 --> 00:39:41,360
feels, although you raise
interesting point of

646
00:39:41,360 --> 00:39:43,840
experimental data because I
think that's probably often

647
00:39:43,840 --> 00:39:48,640
overlooked that the, I guess the
analogy, all the simulation is

648
00:39:48,640 --> 00:39:52,200
just synthetic data and it's not
actually the ground truth, which

649
00:39:52,200 --> 00:39:56,280
is why I always struggle with
initiatives to, you know, if I

650
00:39:56,280 --> 00:39:58,880
train a machine learning model,
we've had this debate, I'd be

651
00:39:58,880 --> 00:40:01,600
interested to know your fault.
So if we take like a standard

652
00:40:01,600 --> 00:40:04,680
data set and the goal of the
exercise is that your surrogate

653
00:40:04,680 --> 00:40:07,960
model should be able to predict
the ground truth and whoever

654
00:40:07,960 --> 00:40:10,440
gets it the closest gets the
highest score and wins.

655
00:40:11,240 --> 00:40:16,040
But that.
Has a certain ground truth baked

656
00:40:16,040 --> 00:40:19,040
into it because it was done with
a random model and LES model.

657
00:40:19,040 --> 00:40:22,800
So if someone comes along who's
trained this big foundation and

658
00:40:22,800 --> 00:40:26,640
then they try and predict your
case, they could actually have a

659
00:40:26,640 --> 00:40:29,600
worse score because their method
was trained on the different

660
00:40:29,600 --> 00:40:32,440
underlying simulation data.
But then which one's right?

661
00:40:32,440 --> 00:40:35,640
Because you've never actually
said what's the ground truth?

662
00:40:35,880 --> 00:40:37,560
The like the real world truth,
you know?

663
00:40:38,160 --> 00:40:40,520
Yeah.
Well, actually I I think this is

664
00:40:40,520 --> 00:40:46,080
also one of the very difficult
and fundamental challenges in

665
00:40:46,080 --> 00:40:49,600
science and engineering We are
right now using the ground truth

666
00:40:49,600 --> 00:40:52,880
is my simulation often like when
we test our methods, you know,

667
00:40:52,880 --> 00:40:55,120
including us, you know, it's
like, how will this method work?

668
00:40:55,120 --> 00:40:57,480
Well, I'll pretend the truth is
my simulator.

669
00:40:59,200 --> 00:41:03,520
The problem I see is that we
say, OK, well, what if you had

670
00:41:03,520 --> 00:41:07,800
the experiment?
Well, in experiments we actually

671
00:41:07,800 --> 00:41:10,480
can never also have access to
the ground truth because in an

672
00:41:10,480 --> 00:41:15,560
experiment you have sensors or
either point sensors or you're

673
00:41:15,840 --> 00:41:21,640
measuring thumb observables, but
you never have the full state

674
00:41:22,320 --> 00:41:24,360
information like you do in the
simulation.

675
00:41:25,000 --> 00:41:28,920
So this idea of a ground truth
is really interesting because

676
00:41:30,120 --> 00:41:32,160
you never have it in real
systems.

677
00:41:33,080 --> 00:41:36,320
You have you have some
manifestations of of

678
00:41:36,320 --> 00:41:41,400
measurements of that ground
truth through your sensors, your

679
00:41:41,400 --> 00:41:44,400
full state knowledge is only
through the in a simulation

680
00:41:44,400 --> 00:41:47,000
world.
And how this connection of how

681
00:41:47,000 --> 00:41:50,040
do you assimilate those two
together becomes now very

682
00:41:50,040 --> 00:41:53,600
important.
And so somehow this, you know,

683
00:41:53,600 --> 00:41:56,720
the whole data simulation effort
and we've seen the success of it

684
00:41:56,720 --> 00:41:59,640
and whether right.
So a lot of companies have

685
00:41:59,640 --> 00:42:04,320
gotten into it because I think
they, they benefited from

686
00:42:04,320 --> 00:42:07,040
basically the idea that like
we've been collecting weather

687
00:42:07,040 --> 00:42:10,640
model weather data for like
decades upon decades, very well

688
00:42:10,720 --> 00:42:13,760
resolved.
And so they, they can train

689
00:42:13,760 --> 00:42:17,120
these big models and a lot of
the people working that space

690
00:42:17,120 --> 00:42:18,520
like look what we can do with
machine learning.

691
00:42:18,520 --> 00:42:23,440
It's like you understand you're
in a very not representative

692
00:42:23,440 --> 00:42:28,240
scientific field, but for 40
years, 50 years, we've been

693
00:42:28,240 --> 00:42:31,960
collecting data ad nauseam in
this space and that's why your

694
00:42:31,960 --> 00:42:36,000
models work whereas and most
other scientific fields, it's

695
00:42:36,000 --> 00:42:39,920
just not there.
Or as you've pointed out, a lot

696
00:42:39,920 --> 00:42:43,640
of a lot of people guard their
data closely.

697
00:42:43,640 --> 00:42:46,280
Like if you're for Formula One
team, you don't share your

698
00:42:46,280 --> 00:42:50,040
simulation data.
If you're a car company, you

699
00:42:50,920 --> 00:42:52,960
typically don't share your
simulation data.

700
00:42:53,360 --> 00:42:56,680
People don't share their supply
chain data like there.

701
00:42:56,680 --> 00:43:01,280
There's just so much data that
is there, but nobody shares it.

702
00:43:01,280 --> 00:43:04,560
So we have that.
It's kind of off limits a little

703
00:43:04,560 --> 00:43:07,440
bit.
Which is quite I've often joked

704
00:43:07,440 --> 00:43:10,360
that probably the best company
to make a foundation model for

705
00:43:10,360 --> 00:43:12,120
automotive is an automotive
company.

706
00:43:12,440 --> 00:43:16,760
You know, frankly, if I was a
car company, I'd be half tempted

707
00:43:16,760 --> 00:43:19,400
if I, you know, if I wasn't
worried more about the car

708
00:43:19,400 --> 00:43:21,160
situation at the moment.
I guess that's not their

709
00:43:21,160 --> 00:43:23,280
priority to build a foundation
model.

710
00:43:23,280 --> 00:43:25,880
But like, you know, that's the
ironic thing.

711
00:43:25,880 --> 00:43:29,040
They've got all the expertise,
all the data, all the knowledge,

712
00:43:29,040 --> 00:43:33,680
all the facilities and they
literally have it sat there,

713
00:43:33,680 --> 00:43:35,480
wind tunnel probably running
24/7.

714
00:43:35,480 --> 00:43:38,560
Every all the, the stuff that,
you know, a startup would dream

715
00:43:38,560 --> 00:43:41,880
of having, they actually have
there.

716
00:43:41,880 --> 00:43:46,800
So it's yeah on that.
What about your time at McLaren

717
00:43:46,800 --> 00:43:47,120
then?
What?

718
00:43:47,120 --> 00:43:52,240
What I guess you must be into
cars to, to also, you know, want

719
00:43:52,240 --> 00:43:55,040
to do that.
But what did it teach you going

720
00:43:55,440 --> 00:43:59,040
from, you know, being a faculty
member to taking that time at

721
00:43:59,040 --> 00:44:01,080
McLaren?
How did it shape your thinking?

722
00:44:02,120 --> 00:44:07,000
Yeah, so, so first I was, it was
such a privilege to be there.

723
00:44:07,000 --> 00:44:10,840
I I've been a Formula One fan
since I was a little kid, so I'm

724
00:44:10,840 --> 00:44:13,520
half Brazilian.
So in the 1970s, growing up a

725
00:44:13,520 --> 00:44:16,880
kid in Brazil at Emerson,
Filipaldi was a Formula One

726
00:44:16,880 --> 00:44:19,320
champion.
So every Brazilian loved

727
00:44:19,320 --> 00:44:22,040
Filipaldi.
But then I started watching in

728
00:44:22,040 --> 00:44:26,920
high school, in the first
Formula One race I watched was a

729
00:44:26,920 --> 00:44:31,040
Tonsena, and I was just some
Sunday morning early.

730
00:44:31,040 --> 00:44:34,880
I started watching this race.
It was on ESPN and there was

731
00:44:34,880 --> 00:44:39,480
this young Brazilian kid.
I told Senna won that race and I

732
00:44:39,480 --> 00:44:41,480
just became a super fan of
Senna.

733
00:44:41,480 --> 00:44:44,240
And, and you know, of course
then I'd loved McLaren because

734
00:44:44,240 --> 00:44:46,200
he was a three time champion
with McLaren.

735
00:44:46,200 --> 00:44:48,800
And so I, I, I was a Formula One
fan from the early days.

736
00:44:48,800 --> 00:44:54,080
And so I was on a sabbatical
here in London and, and I

737
00:44:54,080 --> 00:44:56,080
thought, you know, what, if I
can connect up there.

738
00:44:56,320 --> 00:44:59,720
And the awesome thing is a
Spencer Sherwin who is at

739
00:44:59,880 --> 00:45:03,360
Imperial College in aerospace
engineering, one of his former

740
00:45:03,360 --> 00:45:07,880
PhD students is head of CFT,
Julian Hosler.

741
00:45:08,160 --> 00:45:13,480
So it's just he connected me and
then I was was out there and it

742
00:45:13,480 --> 00:45:18,800
taught me two things. 22 really
big take home messages that

743
00:45:18,800 --> 00:45:22,200
happened for me at McLaren that
actually shaped part of why I

744
00:45:22,200 --> 00:45:25,360
even came to London to come to
auto desk number one.

745
00:45:26,280 --> 00:45:29,560
And this was really memorable.
And it's, it's almost sad to say

746
00:45:29,560 --> 00:45:34,160
a little bit, but they were a
team, they worked together.

747
00:45:34,240 --> 00:45:37,040
They had a common objective.
You could talk to anybody from

748
00:45:37,040 --> 00:45:40,680
the person making coffee to the
cafeteria people to head of

749
00:45:40,680 --> 00:45:45,560
engineering to the the brand new
engineer, everybody knew and had

750
00:45:45,560 --> 00:45:51,640
a shared common goal.
And we're really working

751
00:45:51,640 --> 00:45:55,520
together to that goal.
And it's interesting coming from

752
00:45:55,520 --> 00:46:00,760
the academic environment, which
is, yes, we're collaborative,

753
00:46:00,800 --> 00:46:07,080
but really we are ultra selfish
in our, in our structure, right?

754
00:46:08,160 --> 00:46:15,680
And I, I love that I loved being
around these people who wanted

755
00:46:15,680 --> 00:46:19,560
to do great work, who had a
shared vision.

756
00:46:19,960 --> 00:46:22,800
That doesn't mean people didn't
have, I'm not saying that

757
00:46:22,800 --> 00:46:26,720
everything was, you know,
perfect, but I'm but that was

758
00:46:26,720 --> 00:46:30,680
felt to me so much healthier
than what I've been in for

759
00:46:30,680 --> 00:46:35,600
decades.
So that was the so that that

760
00:46:35,600 --> 00:46:38,680
that really stuck with me.
I have to say it was a really,

761
00:46:38,760 --> 00:46:42,480
it was a really and, and just to
for my own self reflection about

762
00:46:42,480 --> 00:46:47,480
like, Oh, I didn't even realize
how much of A world I live in of

763
00:46:47,880 --> 00:46:51,480
self centeredness.
So.

764
00:46:51,480 --> 00:46:53,600
That is the negative part of
academia, isn't it?

765
00:46:53,600 --> 00:46:58,040
I mean, you know that sort of I
always, I guess the best

766
00:46:58,040 --> 00:47:00,800
description I give, it feels
like every professor is running

767
00:47:00,800 --> 00:47:04,400
their own startup or their own,
you know, because it's very like

768
00:47:04,400 --> 00:47:06,960
people often call it something
lab or some, you know, it's,

769
00:47:07,480 --> 00:47:12,520
it's and I guess at worse, you
know, good luck being a head of

770
00:47:12,520 --> 00:47:16,760
department or a faculty.
It's like all rivals going after

771
00:47:16,760 --> 00:47:18,320
each other, challenging for
funding.

772
00:47:18,720 --> 00:47:21,600
I guess in the best scenario,
it's an amazing cross

773
00:47:21,600 --> 00:47:25,440
collaborative environment of of
of of people.

774
00:47:26,600 --> 00:47:28,240
But yeah, I, I see what you
mean.

775
00:47:28,240 --> 00:47:31,320
There's a certain.
I think I was very collaborative

776
00:47:31,320 --> 00:47:34,160
and I was very friendly across
and worked with lots of people.

777
00:47:34,160 --> 00:47:38,000
But it's still was this very
much like I still have to get my

778
00:47:38,000 --> 00:47:39,160
grant.
Yeah.

779
00:47:39,360 --> 00:47:42,680
And I still have to take care of
my students and, and this was

780
00:47:42,680 --> 00:47:46,080
just kind of refreshing, but
there was this piece there.

781
00:47:46,080 --> 00:47:48,640
And then the thing that I came
there, I was like, oh, you know,

782
00:47:48,720 --> 00:47:50,520
I've been doing reduced order
models.

783
00:47:50,520 --> 00:47:52,880
I'm super excited about this to
do things.

784
00:47:54,000 --> 00:48:00,240
But really the big goal there
was to update shape of the car.

785
00:48:00,240 --> 00:48:02,200
And then you just ask a very
simple question.

786
00:48:02,200 --> 00:48:06,680
Here's a, a mesh, which is this
car, which is, let's call it a

787
00:48:06,680 --> 00:48:10,440
billion parameter mesh.
How do you how do you do grading

788
00:48:10,440 --> 00:48:12,760
descent updates on a
three-dimensional mesh?

789
00:48:13,200 --> 00:48:16,840
And then I just saw that what
you had there was a group of

790
00:48:16,840 --> 00:48:18,680
experts.
And so it was human gradient

791
00:48:18,680 --> 00:48:24,760
descent, like true knowledge
base of people who had long time

792
00:48:24,760 --> 00:48:28,320
experience understanding the
flow physics off the front wing,

793
00:48:28,320 --> 00:48:32,040
the back wing, the underside,
the side pods who were making

794
00:48:32,040 --> 00:48:37,080
collectively a decision about
like what's the next upgrade of

795
00:48:37,080 --> 00:48:42,080
the shape that we try and even
understanding when they felt

796
00:48:42,080 --> 00:48:44,680
like we have to go to the wind
tunnel with this design, the

797
00:48:44,680 --> 00:48:47,320
wind tunnel is very expensive.
We get very limited time.

798
00:48:48,280 --> 00:48:52,840
But this recourse to reality
that would sort of pin them

799
00:48:53,440 --> 00:48:56,480
their designs down or actually
invalidate them, right?

800
00:48:57,080 --> 00:49:01,560
It was really interesting to see
this and I felt honestly, I

801
00:49:01,560 --> 00:49:05,840
guess I felt super like, I feel
like I come in, I got this, you

802
00:49:05,840 --> 00:49:09,520
know, tool set and I felt
powerless a little bit, which is

803
00:49:09,800 --> 00:49:15,720
here is this inverse design
problem can't help with and and

804
00:49:15,720 --> 00:49:17,680
it really stucked with me
afterwards.

805
00:49:17,680 --> 00:49:21,360
Like, how do I get into this
geometry grain and understand

806
00:49:21,360 --> 00:49:23,840
how to handle geometry in a much
better way?

807
00:49:23,840 --> 00:49:26,200
And that, you know, how do I get
a latent representation of

808
00:49:26,200 --> 00:49:30,720
geometry to do updates?
And so, you know, when Autodesk

809
00:49:30,720 --> 00:49:33,600
came along, it was like, wait a
minute, this, this is the most,

810
00:49:33,760 --> 00:49:37,320
this is probably one of the best
geometry companies in the world

811
00:49:37,840 --> 00:49:41,720
and do this design.
And so it felt a very natural

812
00:49:41,720 --> 00:49:45,360
thing to come over because it
was intellectually for me was

813
00:49:45,360 --> 00:49:48,120
one of the main things I wanted
to go after.

814
00:49:48,880 --> 00:49:51,880
And I felt like if I try to do
it in the academic environment,

815
00:49:51,880 --> 00:49:55,400
this pivot towards trying to
generate all the skill set

816
00:49:55,400 --> 00:49:59,400
around geometry would take me a
decade, right, to get like

817
00:49:59,720 --> 00:50:03,040
really there.
Whereas coming to Autodesk was

818
00:50:03,040 --> 00:50:06,120
like, I immediately came into a
group of colleagues who were

819
00:50:06,640 --> 00:50:09,160
they, you know, live and breathe
this stuff.

820
00:50:09,160 --> 00:50:12,640
So it's like, amazing.
So those are the two things that

821
00:50:12,640 --> 00:50:19,640
McLaren that really stuck out to
me and and also maybe the third

822
00:50:19,640 --> 00:50:23,120
was just to walk into this
McLaren tech center where.

823
00:50:23,840 --> 00:50:26,720
Quite nice.
Very nice and also just to see

824
00:50:26,720 --> 00:50:29,960
like, you know, we rarely as an
academics also see that, you

825
00:50:29,960 --> 00:50:32,080
know, by the time we write a
paper, it's a year till it gets

826
00:50:32,080 --> 00:50:36,120
public.
Like there's there's no here on

827
00:50:36,120 --> 00:50:39,320
this factory floor.
What every, all these engineers

828
00:50:39,320 --> 00:50:43,520
could see was I've helped build
this thing and it's right there.

829
00:50:43,520 --> 00:50:45,880
It's like this physical
manifestation of beauty.

830
00:50:45,880 --> 00:50:48,720
And it's like a gallery, an art
gallery, right?

831
00:50:48,720 --> 00:50:50,160
You're like, Oh my gosh, this is
the.

832
00:50:51,080 --> 00:50:55,600
And even to see Santa's car, his
MP4 four from when he, I think

833
00:50:55,600 --> 00:51:00,080
his first championship, it's
like it's, it's so motivating,

834
00:51:00,080 --> 00:51:03,600
right, To have that creative
inspiration that's there.

835
00:51:03,720 --> 00:51:04,920
Yeah.
Yeah, I know.

836
00:51:04,920 --> 00:51:10,160
It's I, I find that was my
experience coming into Formula

837
00:51:10,160 --> 00:51:12,960
One and I've seen almost exactly
the same thing.

838
00:51:12,960 --> 00:51:17,160
And I, I think almost every
start up that comes in trying to

839
00:51:17,160 --> 00:51:20,480
sell into an F1 team faces the
same reality that they all

840
00:51:20,480 --> 00:51:25,560
think, oh, surely, you know,
you, you just need to do this

841
00:51:25,560 --> 00:51:28,800
or, or you know, like you need
to use this better turbans model

842
00:51:28,800 --> 00:51:31,600
or you need to do this surrogate
model.

843
00:51:31,600 --> 00:51:35,120
And yeah, you realize that This
is why I always say about

844
00:51:35,120 --> 00:51:39,840
foundation models or something
that the idea that you come in

845
00:51:39,840 --> 00:51:43,200
and you would use a method to
come up with a brand new design

846
00:51:43,200 --> 00:51:46,680
from scratch.
In reality, they already know

847
00:51:46,680 --> 00:51:50,840
what needs to be done and it's
tweaking the most fine things

848
00:51:51,440 --> 00:51:55,520
that is often very is never
usually tested.

849
00:51:55,520 --> 00:51:58,200
It's usually like a, a big
thing, isn't it?

850
00:51:58,200 --> 00:52:00,520
Like the delta between this and
this and you're looking for the

851
00:52:00,520 --> 00:52:04,240
difference of, you know, 2 drag
counts or something.

852
00:52:05,160 --> 00:52:09,000
It's, yeah, I've often found it
quite humbling when you go in

853
00:52:09,000 --> 00:52:11,160
there thinking you can make a
difference and then you're like,

854
00:52:11,160 --> 00:52:13,280
oh, this is already pretty well
optimized.

855
00:52:14,320 --> 00:52:18,080
And also to see these people who
have put in the time and effort

856
00:52:18,080 --> 00:52:21,560
and have the passion and they
have that special skill set,

857
00:52:21,560 --> 00:52:25,800
like I know how to squeeze out a
delta here in these

858
00:52:26,240 --> 00:52:30,200
manipulations.
That sort of are the the the way

859
00:52:30,200 --> 00:52:33,960
I almost think about it is if we
think about what AI is largely

860
00:52:33,960 --> 00:52:37,800
trained on, RMSERMSE is sort of
a blunt instrument.

861
00:52:37,800 --> 00:52:41,800
It's sort of this, it's sort of
I got all the big scale stuff

862
00:52:41,800 --> 00:52:44,480
for you, right.
But the winning race car is

863
00:52:44,480 --> 00:52:49,680
about all the fine details and
promoting these in the loss

864
00:52:49,680 --> 00:52:52,280
function.
Somehow, you know, this thinks

865
00:52:52,280 --> 00:52:56,040
I, you know, you can see that
neural networks act act as

866
00:52:56,040 --> 00:52:59,160
bandpass filters essentially.
Like I got the big stuff right.

867
00:52:59,160 --> 00:53:02,080
All the small stuff doesn't
really contribute to that RMSC

868
00:53:02,080 --> 00:53:04,800
score.
So like, whatever, we'll throw

869
00:53:04,800 --> 00:53:06,640
it out.
Now, of course we can make

870
00:53:06,640 --> 00:53:09,240
efforts to, you know, put a
diffusion model down there or

871
00:53:09,240 --> 00:53:12,000
something like this to try to
like fill in in so it makes it

872
00:53:12,000 --> 00:53:17,960
look right.
But the this attention to small

873
00:53:17,960 --> 00:53:23,920
detail which a human can make
the focus, maybe we teach our

874
00:53:23,920 --> 00:53:28,520
neural networks to do this, but
like in their general deployment

875
00:53:28,520 --> 00:53:32,760
that's is not really, you know,
these little micro adjustments

876
00:53:32,760 --> 00:53:36,640
are so fine detail, they don't
score in a training score

877
00:53:36,720 --> 00:53:39,440
hardly.
Now that is where they'll like

878
00:53:39,440 --> 00:53:44,920
do think the agents I am far
more bullish on because I feel

879
00:53:44,920 --> 00:53:49,720
like with the greatest respect
to the Formula One engineers,

880
00:53:49,760 --> 00:53:53,760
what they're doing is, is still
something that is just you can

881
00:53:53,760 --> 00:53:57,800
describe what they do and and
actually, I think if you look

882
00:53:57,800 --> 00:54:03,480
across the grid, my hypothesis
of why is one team, you know,

883
00:54:03,480 --> 00:54:06,680
one year amazing and then
another year it's not, you know,

884
00:54:06,680 --> 00:54:09,840
like if they really understood
everything that's needed to be

885
00:54:09,840 --> 00:54:13,160
done, surely they would just
each year build too great.

886
00:54:13,400 --> 00:54:16,680
The fact that they take wrong
turns, they go down the wrong

887
00:54:16,680 --> 00:54:19,680
direction.
They have a a very large

888
00:54:19,680 --> 00:54:22,760
optimization space that once you
start going down a route, you

889
00:54:22,760 --> 00:54:25,880
can't really turn back and you
sort of have to go.

890
00:54:25,880 --> 00:54:31,400
And I do feel that it is still a
very large optimization problem

891
00:54:31,400 --> 00:54:35,000
that a machine could ultimately
do better than a human in in

892
00:54:35,400 --> 00:54:38,400
theory with ultimate resources.
But.

893
00:54:39,560 --> 00:54:43,160
Yeah, yeah.
So if if I was, if I was in

894
00:54:43,160 --> 00:54:46,320
somehow in charge of some
version of some formula that one

895
00:54:46,320 --> 00:54:49,880
team and sort of as a let's say
call it iOS, I'm the AI director

896
00:54:49,880 --> 00:54:55,080
of of some form on team.
What I think I would be doing,

897
00:54:55,080 --> 00:54:59,000
and I'm also bullish on the
agent piece of this.

898
00:54:59,000 --> 00:55:03,400
I would be setting up extensive
interviews with all of my

899
00:55:03,400 --> 00:55:06,880
experienced engineers.
Like how are they making their

900
00:55:06,880 --> 00:55:08,800
decision?
So you made this decision.

901
00:55:08,800 --> 00:55:13,800
Why I want you to verbalize it.
Why did you make decision?

902
00:55:14,080 --> 00:55:17,040
What do you see in the data?
Because a lot of the

903
00:55:17,040 --> 00:55:22,240
experiential learning that
these, you know, master

904
00:55:23,040 --> 00:55:28,120
engineers have, they have a
pipeline.

905
00:55:28,440 --> 00:55:30,840
It's just they haven't maybe
verbalized it.

906
00:55:30,840 --> 00:55:34,480
And if you could pull that out
and you say my agent's going to

907
00:55:34,480 --> 00:55:38,040
try to take on your persona the
way you think, but it needs to

908
00:55:38,040 --> 00:55:40,440
be trained in the way you think,
the way you're making your

909
00:55:40,440 --> 00:55:44,480
decision points, I actually
think that is a viable way

910
00:55:44,480 --> 00:55:47,520
forward.
And that could accelerate things

911
00:55:47,880 --> 00:55:53,120
significantly.
And so the downloading of

912
00:55:53,120 --> 00:55:58,920
experience of people, somehow
some next step that needs to

913
00:55:58,920 --> 00:56:01,240
happen in it.
And it could be that these

914
00:56:01,240 --> 00:56:05,000
people have never thought about
like deeply about what was my

915
00:56:05,000 --> 00:56:08,320
algorithm if I had to, if I have
to write down what I do and when

916
00:56:08,320 --> 00:56:10,720
I'm looking at the car.
And because it's actually

917
00:56:10,720 --> 00:56:12,600
interesting, when you look at
McLaren, you can walk up and

918
00:56:12,600 --> 00:56:15,080
down this pit of engineers.
And then when you're down on the

919
00:56:15,600 --> 00:56:18,480
design arrow side, you see
basically people have the double

920
00:56:18,480 --> 00:56:21,760
screen like they do a very tough
company and what they're got

921
00:56:21,760 --> 00:56:23,360
pictures on there is flow
physics.

922
00:56:23,360 --> 00:56:27,560
And these guys are just studying
this all day long to make their

923
00:56:27,560 --> 00:56:32,560
decisions.
But if you could understand, if

924
00:56:32,560 --> 00:56:36,400
they could verbalize what
they're doing in that study, you

925
00:56:36,400 --> 00:56:38,960
know what machine could do it
Like, and just like, OK, I did

926
00:56:38,960 --> 00:56:42,160
in five seconds.
What you all morning looking at

927
00:56:42,400 --> 00:56:45,400
here at 5 seconds later.
Here's here's here's the

928
00:56:45,400 --> 00:56:47,120
assessment.
Yes.

929
00:56:47,120 --> 00:56:49,560
If you don't know what they're
looking at, it's really hard for

930
00:56:49,560 --> 00:56:52,280
the agent to to know that,
right.

931
00:56:52,720 --> 00:56:56,400
The only thing on that and then,
and I want to go too down a

932
00:56:56,400 --> 00:57:00,360
rabbit hole on this one, but is
if I was an experienced F1

933
00:57:00,360 --> 00:57:03,360
engineer who's done it for 20
years, I'd almost want to, you

934
00:57:03,360 --> 00:57:08,280
know, copyright my skills dot MD
or something or, or monetize it.

935
00:57:08,280 --> 00:57:11,320
Because, you know, as you could
imagine, like once that person's

936
00:57:11,320 --> 00:57:15,120
described F4 process and you put
it into some skills file or

937
00:57:15,120 --> 00:57:17,440
whatever, and then the team
goes, right, well, thank you

938
00:57:17,440 --> 00:57:19,440
very much.
The agent actually now runs much

939
00:57:19,440 --> 00:57:22,200
faster than you can.
So, you know, see you later.

940
00:57:22,880 --> 00:57:26,080
I do often want, because I
completely agree with you, this

941
00:57:26,120 --> 00:57:30,320
human knowledge can be
extracted, you know, because it

942
00:57:30,320 --> 00:57:33,360
is often quite repeatable.
But I, I think it's an

943
00:57:33,360 --> 00:57:35,760
interesting, you know, if
someone's creating a startup of

944
00:57:35,760 --> 00:57:40,200
like, you know, I, I'm going to
monetize my knowledge and yeah,

945
00:57:40,200 --> 00:57:45,520
use me as an agent, but I need
some money out of it because.

946
00:57:46,400 --> 00:57:47,840
Yeah.
So this is actually, I think

947
00:57:47,840 --> 00:57:51,960
this, this argument is one of
the most amazing, I think things

948
00:57:51,960 --> 00:57:55,640
we're going to have to deal with
overall in the tech community.

949
00:57:55,640 --> 00:57:57,800
Even, you know, here at
Autodesk, they're in America,

950
00:57:57,800 --> 00:58:00,880
which is OK.
So if these agents are so

951
00:58:00,880 --> 00:58:05,400
effective and we can program
them to be effective, if you're

952
00:58:05,720 --> 00:58:09,280
a Formula One team, do you say
like, if I can do this, I could

953
00:58:09,280 --> 00:58:13,880
shrink my workforce or do you go
the other ways with my

954
00:58:13,880 --> 00:58:18,120
workforce, I can 10X the number
of designs I'm doing.

955
00:58:19,000 --> 00:58:24,360
Going through like this is a
really interesting point for us.

956
00:58:24,360 --> 00:58:28,640
Like a company like here we say
like, well, hey, you know,

957
00:58:28,640 --> 00:58:32,240
essentially one person can do
now 10X the work they used to be

958
00:58:32,240 --> 00:58:35,040
able to.
Well, guess what, we could take

959
00:58:35,040 --> 00:58:38,080
10X customers.
Yeah, yeah.

960
00:58:38,080 --> 00:58:40,760
Or we could cut the company down
by 10.

961
00:58:41,440 --> 00:58:42,240
Yeah.
So.

962
00:58:43,080 --> 00:58:47,480
I, I suspect it will probably be
more that you'll just be able to

963
00:58:47,480 --> 00:58:51,400
do more, you know, with the,
with the, the staff that you

964
00:58:51,400 --> 00:58:54,600
have.
I I, I suspect so that's my gut

965
00:58:54,600 --> 00:59:00,840
feeling because whenever we've
had more, you know, your HPC

966
00:59:00,840 --> 00:59:04,760
facility can now run your cases
twice as fast or the code can

967
00:59:04,760 --> 00:59:07,320
run.
But normally it's been that, OK,

968
00:59:07,320 --> 00:59:08,960
great.
Now you just have to do double

969
00:59:08,960 --> 00:59:12,640
the amount of work or you do
twice as many simulations or you

970
00:59:12,640 --> 00:59:14,880
do whatever.
But I, yeah, I definitely think

971
00:59:14,880 --> 00:59:19,120
that is a, an interesting one.
The question for you on

972
00:59:19,120 --> 00:59:21,360
Autodesk, do you have any
regrets?

973
00:59:21,360 --> 00:59:25,400
You didn't go into industry
earlier on, like now that you're

974
00:59:25,400 --> 00:59:28,120
there, is there any?
Yeah.

975
00:59:28,600 --> 00:59:34,760
Yeah, actually not so much.
I mean, I, I, I actually, I

976
00:59:34,960 --> 00:59:39,280
think this is just a perfect
time in life for me to come here

977
00:59:39,440 --> 00:59:48,280
at this point.
But I, I, I think that, I mean,

978
00:59:48,280 --> 00:59:53,680
I guess in some sense I haven't
been so money motivated overall,

979
00:59:53,680 --> 00:59:57,040
career wise.
I think obviously could have

980
00:59:57,040 --> 01:00:00,680
gone into a tech company in
Seattle much earlier.

981
01:00:00,800 --> 01:00:02,320
Like you pick one, they're all
there.

982
01:00:04,240 --> 01:00:07,520
There wasn't I, I, I was more
interested in doing really

983
01:00:07,520 --> 01:00:13,080
interesting work if I could.
Autodesk kind of like I said,

984
01:00:13,080 --> 01:00:17,320
was a confluence of the right
time intellectually.

985
01:00:17,480 --> 01:00:20,480
Like I said, I wanted to really
think about how can I handle

986
01:00:20,480 --> 01:00:23,920
geometry and physics jointly?
How does geometry induce

987
01:00:23,920 --> 01:00:27,400
physics, right?
I was just so fascinated after

988
01:00:27,400 --> 01:00:31,720
that McLaren year that.
So it was, it was the right time

989
01:00:32,640 --> 01:00:40,360
intellectually for me to come to
to Autodesk and I'm very happy

990
01:00:40,360 --> 01:00:44,160
with it now.
But I don't regret my time at

991
01:00:44,160 --> 01:00:45,560
UW.
I think that was found, you

992
01:00:45,560 --> 01:00:49,040
know, what I learned there and
the students I had, I loved my

993
01:00:49,040 --> 01:00:51,600
students, my postdocs, my
collaborators.

994
01:00:52,400 --> 01:00:54,320
It's just, it's just a good time
in life.

995
01:00:54,320 --> 01:00:57,160
And also my kids graduated from
high school, so I didn't have to

996
01:00:58,040 --> 01:01:00,480
try to live in a good school
district in London.

997
01:01:01,400 --> 01:01:04,040
Me and my wife could just say,
where do we want to live?

998
01:01:04,040 --> 01:01:07,680
I went off to worry about like
all this kids stuff, right?

999
01:01:07,680 --> 01:01:10,080
Which you know, which is, you
know, I went through that

1000
01:01:10,080 --> 01:01:10,840
already.
I'm done.

1001
01:01:13,400 --> 01:01:16,360
So.
So where's the future lie then?

1002
01:01:16,360 --> 01:01:18,040
Where?
Where do you see if we could put

1003
01:01:18,200 --> 01:01:21,680
a looking glass?
And we skipped forward in five

1004
01:01:21,680 --> 01:01:24,640
years and you and I talk again.
Hopefully it's somewhere like

1005
01:01:24,640 --> 01:01:27,200
Barcelona or some nice, you
know, some nice location.

1006
01:01:28,240 --> 01:01:30,040
Well, where?
Where do you think we'll be?

1007
01:01:32,040 --> 01:01:35,000
Yeah.
So the first thing that I think

1008
01:01:35,000 --> 01:01:39,760
is going to happen sooner than
later is people aren't going to

1009
01:01:39,760 --> 01:01:41,920
put out GitHub code.
They're going to put out GitHub

1010
01:01:41,920 --> 01:01:44,880
agents.
I'm just using that language.

1011
01:01:44,880 --> 01:01:47,560
Like, you know, right now you
say like, hey, I wrote this code

1012
01:01:47,560 --> 01:01:49,680
you can download on GitHub.
It's like, no, no, you're just

1013
01:01:49,680 --> 01:01:54,400
going to give me access to your
agent that manages all of that,

1014
01:01:54,920 --> 01:01:56,480
right?
So it feels to me like this, a

1015
01:01:56,480 --> 01:01:58,800
genic push is quite a real
thing.

1016
01:01:58,960 --> 01:02:01,840
Like even programming on a
higher level, like with a

1017
01:02:02,160 --> 01:02:04,400
something like Kiln, just in
terms of the workflow.

1018
01:02:04,400 --> 01:02:08,800
Instead of me starting to just
run your code, I just, I'm

1019
01:02:08,800 --> 01:02:10,520
interacting with your agent,
with your code.

1020
01:02:11,840 --> 01:02:14,720
I think that's one thing that is
bound to happen.

1021
01:02:14,720 --> 01:02:17,960
I, I, maybe I'm wrong, but I, I
kind of feel like that's right.

1022
01:02:17,960 --> 01:02:22,080
That's just the such a clean
pathway for people to share code

1023
01:02:22,080 --> 01:02:25,080
as you're sharing the agent.
So it's not like it's just like

1024
01:02:25,080 --> 01:02:27,960
you're saying, here's my code
and the grad student that wrote

1025
01:02:27,960 --> 01:02:29,560
it on.
So if you have any questions,

1026
01:02:29,560 --> 01:02:31,840
they can answer all of it
because they built this code.

1027
01:02:32,240 --> 01:02:36,120
But it's like genic part, the
genic part of that.

1028
01:02:36,120 --> 01:02:43,040
No Second, I, I just feel like
we're going to have like this

1029
01:02:43,160 --> 01:02:48,880
ability to deploy so many of
these agents and partnerships

1030
01:02:48,880 --> 01:02:50,640
that we're going to be able to
come up with.

1031
01:02:51,120 --> 01:02:55,520
I, I think we're going to have
much bigger thoughts than we've

1032
01:02:55,520 --> 01:02:59,640
had in the past.
Because in a lot of our future

1033
01:02:59,640 --> 01:03:03,240
thinking, we don't just think,
what if we did this?

1034
01:03:03,480 --> 01:03:06,280
We also have to balance it.
Like, yeah, but what could I

1035
01:03:06,280 --> 01:03:11,960
actually maybe do if I stretch?
But now what you could do with

1036
01:03:11,960 --> 01:03:15,480
you stretch is it's it's so much
bigger.

1037
01:03:16,280 --> 01:03:22,840
So I think our I think about our
goal setting capabilities now in

1038
01:03:22,840 --> 01:03:27,120
terms of where do.
And I think I'm still trying to

1039
01:03:27,120 --> 01:03:30,480
get my head around that now
because I'm trying to also train

1040
01:03:30,480 --> 01:03:34,200
myself to think much bigger
thoughts about what we could

1041
01:03:34,200 --> 01:03:38,440
achieve given the tools that
have just just even in one year

1042
01:03:39,080 --> 01:03:41,800
have emerged.
Like one year alone has all of a

1043
01:03:41,800 --> 01:03:45,360
sudden give you gives you this
transformational ability.

1044
01:03:47,800 --> 01:03:50,720
And the hard part about making
these projections into the

1045
01:03:50,720 --> 01:03:57,000
future is that right now, you
don't know, like if someone's

1046
01:03:57,000 --> 01:03:59,880
going to all of a sudden pop
something out and like one

1047
01:03:59,880 --> 01:04:02,680
month, by the way, here's this
new tool, check it out.

1048
01:04:02,680 --> 01:04:05,880
And it's like, Oh my gosh, which
is everything, right?

1049
01:04:06,080 --> 01:04:09,160
Like this is starting to happen
even in design space.

1050
01:04:09,160 --> 01:04:13,360
So I'll promote a paper that
these these guys wrote from sort

1051
01:04:13,360 --> 01:04:19,000
of a largely a Oxford Cambridge
collaboration called Art of

1052
01:04:19,000 --> 01:04:20,440
Craft.
Maybe you saw this one.

1053
01:04:20,440 --> 01:04:24,800
This is just this generative
design with articulated

1054
01:04:24,960 --> 01:04:29,200
engineering products and it's
just like this fascinating thing

1055
01:04:29,200 --> 01:04:33,000
that they were able to build out
and you're like, OK, I didn't

1056
01:04:33,000 --> 01:04:36,560
think this was this was a cry.
Like incredible that they

1057
01:04:36,560 --> 01:04:40,600
achieved it, right.
And it just feels like, So what

1058
01:04:40,600 --> 01:04:47,120
people are able to achieve
sometimes is beyond what I like,

1059
01:04:47,840 --> 01:04:49,680
like I'm not imagining big
enough.

1060
01:04:49,680 --> 01:04:52,600
Frankly, it is.
I think that's my the the take

1061
01:04:52,600 --> 01:04:55,720
home message for myself in this
last year is like I've got to be

1062
01:04:55,720 --> 01:05:00,200
much more grand scoped in my
imagination about what could be.

1063
01:05:02,480 --> 01:05:05,240
And it's also part of what I'm
stealing with our team is like,

1064
01:05:05,280 --> 01:05:10,600
we need to think a lot bigger.
Yes, some things I I would I

1065
01:05:10,600 --> 01:05:14,040
mean that's been probably one of
the things that I've enjoyed

1066
01:05:14,040 --> 01:05:17,480
being NVIDIA is it's a company
that, you know, obviously thanks

1067
01:05:17,480 --> 01:05:20,720
to the guy the top Jensen, you
know, tends to think quite big

1068
01:05:20,720 --> 01:05:25,080
and it is quite infectious.
You know, you do start to now,

1069
01:05:25,080 --> 01:05:27,280
of course, that's going to be
based on delivery, you know,

1070
01:05:27,320 --> 01:05:29,760
that you can deliver it and it
has to have some reality.

1071
01:05:30,040 --> 01:05:35,440
But I think, yeah, thinking big
is probably something that in

1072
01:05:35,440 --> 01:05:37,240
some ways academia is good at
doing.

1073
01:05:37,240 --> 01:05:41,440
But I feel like to your earlier
point is also one where we're

1074
01:05:41,440 --> 01:05:44,680
quite quick to shut things down,
you know, especially the review

1075
01:05:44,680 --> 01:05:49,360
process and the sort of there is
a little bit of a skepticism or

1076
01:05:49,360 --> 01:05:55,520
so, you know, whereas probably
the tech world is more willing

1077
01:05:55,520 --> 01:05:58,920
to like, which is I guess why
all startups come about.

1078
01:05:58,920 --> 01:06:03,600
And, and yeah, so I I would, I
would tend to agree with you

1079
01:06:03,600 --> 01:06:07,360
that thinking big is actually a
requirement at the moment given

1080
01:06:07,360 --> 01:06:09,880
how fast things are moving.
Yeah.

1081
01:06:10,040 --> 01:06:15,000
And and also I, I, I think that
academics used to have some of

1082
01:06:15,000 --> 01:06:18,160
the bigger thought life, I guess
I would say.

1083
01:06:19,000 --> 01:06:23,400
And there's still some truth to
that, but at least in our

1084
01:06:23,400 --> 01:06:26,080
fields, it's not clear they have
the resources now.

1085
01:06:26,480 --> 01:06:31,120
Well, that's.
To, to go after, and this is

1086
01:06:31,120 --> 01:06:34,760
partly the success, what I've
seen of the computer science

1087
01:06:34,760 --> 01:06:39,800
crowd is if you really look at
some of the big pushes and some

1088
01:06:39,800 --> 01:06:42,600
of the academics involved, it's
because they've been in

1089
01:06:42,600 --> 01:06:47,240
partnerships with resource rich
companies like Google, like

1090
01:06:47,240 --> 01:06:51,480
Meta, like NVIDIA, where it's
like, like if they were just

1091
01:06:51,480 --> 01:06:54,640
sitting at Stanford, like I have
my friends there, Ali Labs, one

1092
01:06:54,640 --> 01:06:57,400
of my grad students is going to
go there for a postdoc and I'm

1093
01:06:57,400 --> 01:06:59,880
super excited.
I was like, you know, it's like,

1094
01:06:59,880 --> 01:07:02,000
if you want to do something
interesting there, it's pretty

1095
01:07:02,000 --> 01:07:07,400
easy to reach out and I think
have these people partnering

1096
01:07:07,400 --> 01:07:12,720
with you to do big scale work.
Like if you were just there like

1097
01:07:12,720 --> 01:07:15,880
me and you have to write a grant
to get a little machine that

1098
01:07:15,880 --> 01:07:20,440
like can only do a fraction of
it, like, but now they can they

1099
01:07:20,440 --> 01:07:24,280
can really work with in this
environment to do great things.

1100
01:07:26,200 --> 01:07:32,040
So yeah, I, I think right now
industry is favored in terms of

1101
01:07:32,040 --> 01:07:36,280
transformational parts.
And because the resources are

1102
01:07:36,280 --> 01:07:44,240
there to do things, I think the
pressure on industry is to get

1103
01:07:44,240 --> 01:07:49,120
the right partnerships with
academic people, right and vice

1104
01:07:49,120 --> 01:07:50,840
versa.
Like so if you I think that

1105
01:07:50,840 --> 01:07:56,160
stills really great strategy is
that, you know, even here at

1106
01:07:56,160 --> 01:07:59,440
Autodesk, I have definitely am
reaching out to people that I

1107
01:07:59,440 --> 01:08:02,600
think are really valuable and
making connections.

1108
01:08:04,160 --> 01:08:08,240
So I'm valuable to them now
sitting on the industry side,

1109
01:08:08,880 --> 01:08:11,280
they're valuable to me sitting
on sort of the intellectual

1110
01:08:11,280 --> 01:08:14,640
thought life side.
But the partnership is fantastic

1111
01:08:14,640 --> 01:08:17,399
because everybody really wins in
it, yes.

1112
01:08:17,800 --> 01:08:21,319
And so that's, that's I think a
a really important place to go

1113
01:08:21,319 --> 01:08:24,160
forward to.
Yeah, No, I I agree with you

1114
01:08:24,160 --> 01:08:26,560
that the I should probably
should rephrase what I said

1115
01:08:26,560 --> 01:08:28,319
before.
It's true academia can have big

1116
01:08:28,359 --> 01:08:31,240
thoughts, but because they know
they don't have the resources,

1117
01:08:31,240 --> 01:08:33,240
it's almost like, well, that's a
nice forward, but I'm never

1118
01:08:33,240 --> 01:08:36,200
going to be able to do it.
So what can I actually achieve

1119
01:08:36,279 --> 01:08:39,479
and get a paper out and, you
know, get this funding in?

1120
01:08:39,479 --> 01:08:42,760
So they then have to think
smaller, whereas you're right,

1121
01:08:42,760 --> 01:08:45,800
if you're a, you know, a big
tech company, you can be a bit

1122
01:08:45,800 --> 01:08:48,560
bolder.
So the when the two come

1123
01:08:48,560 --> 01:08:51,479
together, you get the best, I
guess.

1124
01:08:51,680 --> 01:08:55,240
So maybe as a final question for
you, and we touched upon it a

1125
01:08:55,240 --> 01:09:00,160
little bit, if you're, we have,
let's say, some students

1126
01:09:00,160 --> 01:09:04,920
listening now, whether they're
undergraduates or PhDs, it's a

1127
01:09:04,920 --> 01:09:09,840
tricky time, right?
You know, what would you

1128
01:09:10,279 --> 01:09:14,880
recommend that they would focus,
let's say, APHD on or what

1129
01:09:14,880 --> 01:09:18,840
should they study to be relevant
in the next sort of five years

1130
01:09:18,840 --> 01:09:24,120
of this wave of transformation?
Yeah, so my first thing I tell

1131
01:09:24,120 --> 01:09:31,200
them is they're living, I think
in one of the most exciting

1132
01:09:31,840 --> 01:09:38,720
times in human history because I
think this this time, the future

1133
01:09:38,720 --> 01:09:42,080
generations, they'll pinpoint
this period of time is like the

1134
01:09:42,080 --> 01:09:45,399
world changed.
And this is massively

1135
01:09:45,399 --> 01:09:49,680
influential about what what it
means for us as humanity right

1136
01:09:49,680 --> 01:09:51,720
now.
This is and to be part of it.

1137
01:09:51,920 --> 01:09:53,560
I mean, it doesn't mean it has a
good ending.

1138
01:09:53,640 --> 01:09:55,440
Whatever.
I'm I'm just saying that you but

1139
01:09:55,440 --> 01:09:58,040
they are part of the they are
part of the puzzle piece in

1140
01:09:58,040 --> 01:10:01,480
there.
And not only so it's, it's

1141
01:10:01,480 --> 01:10:03,080
fascinating, It's a little
scary.

1142
01:10:03,080 --> 01:10:05,000
They got a strap on their seat
belt and go.

1143
01:10:05,040 --> 01:10:09,480
And just like this is going and
there's no stopping this thing.

1144
01:10:09,480 --> 01:10:12,280
I mean, as much as people want
to step back and say, let's

1145
01:10:12,280 --> 01:10:16,000
wait, let's talk about it, it's
like it's too much inertia.

1146
01:10:16,000 --> 01:10:17,960
Is it?
We're we're going and we just

1147
01:10:17,960 --> 01:10:22,800
have to do this.
But there's this famous quote of

1148
01:10:23,000 --> 01:10:26,520
from Picasso that I always like
to share with people.

1149
01:10:27,880 --> 01:10:31,800
It's what was from 1968, the
year I was born, and Picasso

1150
01:10:31,800 --> 01:10:33,960
1968.
He made a comment about

1151
01:10:33,960 --> 01:10:37,640
computers and he his his quote,
computers are worthless.

1152
01:10:38,280 --> 01:10:44,040
They can only answer questions.
And I think that statement is

1153
01:10:44,040 --> 01:10:50,920
amazing today, which is the real
value, I think is us as humans

1154
01:10:50,920 --> 01:10:53,160
are still there asking the what
ifs.

1155
01:10:54,240 --> 01:10:56,320
You know, a lot of these
startups come from people like,

1156
01:10:56,320 --> 01:10:59,440
what if we could do this?
What if we could do that?

1157
01:10:59,440 --> 01:11:01,560
What if I want to go to the
moon?

1158
01:11:01,560 --> 01:11:04,120
I want to build a airplane, I
want to build a Formula One car?

1159
01:11:04,840 --> 01:11:10,280
How would I So the largely it
feels like even with this AI

1160
01:11:10,280 --> 01:11:13,200
kick, these are amazing new
tools.

1161
01:11:13,840 --> 01:11:18,560
But really you're still in
charge of really directing where

1162
01:11:18,560 --> 01:11:21,840
a lot of this goes.
You're not a passenger, you're

1163
01:11:21,840 --> 01:11:24,400
the driver.
It's just that you now have a

1164
01:11:24,400 --> 01:11:28,800
Lamborghini or a Ferrari, right?
And you didn't get taught how to

1165
01:11:28,800 --> 01:11:30,720
drive this thing.
So that's what makes it a little

1166
01:11:30,720 --> 01:11:35,120
scared.
So, so that's one thing the,

1167
01:11:35,200 --> 01:11:38,120
the, the other is that I still
think you need foundational

1168
01:11:38,120 --> 01:11:39,720
knowledge.
You still need to be as well

1169
01:11:39,720 --> 01:11:42,680
educated as you can and, you
know, just really foundational

1170
01:11:42,680 --> 01:11:45,440
thinking, whether that's it's
for the mathematics that

1171
01:11:45,440 --> 01:11:48,800
underlies, whether it's
computational math, you know,

1172
01:11:48,840 --> 01:11:52,840
deep linear algebra knowledge or
deep physics knowledge.

1173
01:11:52,840 --> 01:11:56,280
These things still matter a
great deal because at some point

1174
01:11:56,280 --> 01:11:59,320
they're going to play
fundamental roles in largely

1175
01:11:59,320 --> 01:12:03,640
developing your thought
processes and your critical

1176
01:12:03,640 --> 01:12:05,720
thinking ability.
But also maybe you bring you

1177
01:12:05,720 --> 01:12:11,000
back to foundational thinking it
when you're developing these

1178
01:12:11,080 --> 01:12:13,800
these models.
And of course for them, they

1179
01:12:13,800 --> 01:12:16,960
just have to adopt these tools
like Clot.

1180
01:12:16,960 --> 01:12:19,680
If they're not using Clot code
or something like Kilner or

1181
01:12:19,680 --> 01:12:24,240
Cursor, it's like every day
you're getting further behind

1182
01:12:24,240 --> 01:12:26,880
from where people are working,
right?

1183
01:12:28,560 --> 01:12:31,160
And it feels a little
uncomfortable, right?

1184
01:12:31,320 --> 01:12:33,200
Ultimately for some of them,
right?

1185
01:12:33,200 --> 01:12:35,480
It's like, I mean, I'm getting
this code.

1186
01:12:36,040 --> 01:12:37,960
I don't should I read through
it?

1187
01:12:37,960 --> 01:12:41,880
Should I read me this thousand
lines of code and it works, but

1188
01:12:41,880 --> 01:12:44,920
I don't I kind of just only have
a vague idea of what these

1189
01:12:44,920 --> 01:12:49,200
pieces doing that feels very
uncomfortable because like I'm

1190
01:12:49,200 --> 01:12:51,520
sure when you were in school and
when I was in school, it's like

1191
01:12:52,080 --> 01:12:53,800
you had to know every single
line in your code.

1192
01:12:53,880 --> 01:12:58,200
You wrote every single your code
as those such a transformation

1193
01:12:58,200 --> 01:13:01,600
of like, yes, you have to kind
of know that you have to know

1194
01:13:01,600 --> 01:13:03,720
what's going on, but on the
other hand, you just like to

1195
01:13:03,720 --> 01:13:08,400
figure out how to move at the
speed of what's happening.

1196
01:13:08,400 --> 01:13:11,920
So like and it's such a 2
opposite poles that have to be

1197
01:13:11,920 --> 01:13:13,880
there.
I did.

1198
01:13:15,040 --> 01:13:18,160
I was having a conversation with
one professor who mentioned

1199
01:13:18,160 --> 01:13:19,920
something interesting.
I won't say his name just in

1200
01:13:19,920 --> 01:13:22,640
case he didn't want it to be
shared, but although it's

1201
01:13:22,640 --> 01:13:27,440
nothing controversial, it was
just a point of how would you,

1202
01:13:28,400 --> 01:13:32,840
if you're an undergraduate, say
you should use AI as a tutor,

1203
01:13:33,440 --> 01:13:36,560
not to do it for you.
So you should still do

1204
01:13:36,560 --> 01:13:39,600
everything yourself, but
essentially use AI to guide you.

1205
01:13:39,600 --> 01:13:43,360
It's like they are unbelievably
good at explaining things in any

1206
01:13:43,360 --> 01:13:47,160
tone you want in any adapter,
you know, like, but you've got

1207
01:13:47,160 --> 01:13:49,680
to do it yourself because if
not, you won't learn it.

1208
01:13:50,040 --> 01:13:54,600
And then as you transition into,
let's say, APHD, it starts to

1209
01:13:54,600 --> 01:14:00,040
become more of a, an assistant,
you know, where maybe you have

1210
01:14:00,040 --> 01:14:03,960
learnt some of the physics.
So now you can rely on it more

1211
01:14:04,520 --> 01:14:05,880
and you're just sort of checking
it.

1212
01:14:05,880 --> 01:14:08,880
And then as you get even more
late in your career, you're

1213
01:14:08,880 --> 01:14:12,000
almost using it as a PhD or as a
junior engineer.

1214
01:14:12,720 --> 01:14:15,640
And so I feel like the risk is
if you use AI at the

1215
01:14:15,640 --> 01:14:19,480
undergraduate level like
somebody does, maybe in our

1216
01:14:19,480 --> 01:14:23,680
situation that's the danger
because then you've not really

1217
01:14:23,680 --> 01:14:27,920
learnt it.
But of course the temptation is

1218
01:14:27,920 --> 01:14:30,040
to use it when you're an
undergraduate in that way.

1219
01:14:30,440 --> 01:14:32,240
And that's probably the risk
factor, isn't it?

1220
01:14:32,240 --> 01:14:35,920
Like, do you ever really
understand things when you have

1221
01:14:35,920 --> 01:14:38,600
this cheat code?
Since you start playing computer

1222
01:14:38,600 --> 01:14:42,320
game with the cheat code, you
know you it's hard not to use it

1223
01:14:42,320 --> 01:14:46,160
sometimes.
Well, and, and I think the I

1224
01:14:46,160 --> 01:14:47,960
think that's a fair assessment
of things.

1225
01:14:47,960 --> 01:14:53,560
And I think ultimately if, if,
if you know, if you, if I were

1226
01:14:53,560 --> 01:14:58,280
still an academic academia and
so forth, I think where I think

1227
01:14:58,280 --> 01:15:02,400
the, the thought really needs to
be spent on how to use this tool

1228
01:15:02,400 --> 01:15:05,800
is the high end students that
you're super smart students, you

1229
01:15:05,800 --> 01:15:08,440
don't have to worry about them.
They'll just figure it all out.

1230
01:15:08,800 --> 01:15:10,400
They didn't need you in the 1st
place.

1231
01:15:10,400 --> 01:15:12,840
They can just your stuff out.
They'll learn that on their own.

1232
01:15:12,840 --> 01:15:16,080
They'll have deep knowledge of
stuff and they'll, they're not a

1233
01:15:16,080 --> 01:15:18,600
concern.
The low end students have always

1234
01:15:18,600 --> 01:15:21,320
been problematic because they
never quite get it no matter how

1235
01:15:21,320 --> 01:15:24,040
much you try to, you know,
they're maybe not spending the

1236
01:15:24,040 --> 01:15:26,640
time they need.
It's that middle group who are

1237
01:15:26,640 --> 01:15:29,200
going to be your day-to-day
engineers and so many companies.

1238
01:15:30,480 --> 01:15:35,000
How do you educate that group to
be good stewards of the software

1239
01:15:35,000 --> 01:15:39,440
of the practices, right?
Because I think that group needs

1240
01:15:39,600 --> 01:15:44,680
probably the most guidance of
how to be an intelligent

1241
01:15:44,680 --> 01:15:51,280
engineer in a world where it is
so tempting to just pawn it off

1242
01:15:51,280 --> 01:15:54,520
here, right?
It's easy and I get the right

1243
01:15:54,520 --> 01:15:58,560
hand, you know, whatever, but it
but how, how do you teach that

1244
01:15:58,560 --> 01:16:03,880
group to use these things
responsibly so that they're so

1245
01:16:03,880 --> 01:16:05,760
they are, you know, because
look, these are the people who

1246
01:16:05,760 --> 01:16:10,760
build our airplanes and build
our cars like it's we need, we

1247
01:16:10,760 --> 01:16:14,000
need them to be proficient and
be responsible, right, because

1248
01:16:14,000 --> 01:16:16,360
we are going to be in their
product space.

1249
01:16:16,360 --> 01:16:19,480
Yeah, yeah.
And so, and I don't know what

1250
01:16:19,480 --> 01:16:22,440
the quite the right answer there
is, except that we need them to

1251
01:16:22,440 --> 01:16:24,720
know these tools.
We need them to also balance it

1252
01:16:24,720 --> 01:16:28,720
with like, yeah, but you check
these tools and have maturity

1253
01:16:28,720 --> 01:16:33,040
about using them because people
depend upon you.

1254
01:16:33,920 --> 01:16:36,920
Check because I'm going to get
in the car you built me.

1255
01:16:36,920 --> 01:16:38,800
And yeah.
And I don't want that thing

1256
01:16:38,800 --> 01:16:41,160
falling apart when I'm, you
know, going down the freeway.

1257
01:16:41,320 --> 01:16:42,760
Right.
Yeah, exactly.

1258
01:16:44,320 --> 01:16:45,520
Great.
Well, thank you so much for

1259
01:16:45,520 --> 01:16:47,800
taking the time to speak.
I I'm sure we could have carried

1260
01:16:47,800 --> 01:16:50,960
on for many hours and I hope we
can do that.

1261
01:16:50,960 --> 01:16:53,800
But maybe over, you know, a
drink sometime.

1262
01:16:53,880 --> 01:16:55,440
Yeah.
That sounds good to me.

1263
01:16:56,840 --> 01:16:58,960
But yeah, all the best.
It's Alteredesk.

1264
01:16:58,960 --> 01:17:01,920
I'm very excited to see what
your group's going to create and

1265
01:17:02,280 --> 01:17:05,200
I hope, I'm sure we'll, we'll
hear about it over the coming

1266
01:17:05,200 --> 01:17:06,440
years.
Awesome.

1267
01:17:06,760 --> 01:17:07,920
Thank you again.
Great.

1268
01:17:07,920 --> 01:17:09,160
You got it.
Thanks, Neil.