Latest / Financial Survival Network / The Rise of Artificial Intelligence in Investment Finance - David Marra #6047

Transcript

The Rise of Artificial Intelligence in Investment Finance - David Marra #6047 · Financial Survival Network

0:00
If you were to summarize this generative AI, what's different about all the machine
0:06
learning algorithms that have come before it and other artificial intelligence other algorithm come before
0:10
it was the learning that if we just threw massive amounts of data at it,
0:16
we could get a step level increase in what we could get out of
0:21
it, what it could do for us. That's incredible. You are listening
0:25
to Carrie Leutz's Financial Survival Network, where you get valuable information. You just
0:30
can't find anywhere else to thrive in today's trying times. You need the Financial
0:36
Survival Network now more than ever. Go to Financial Survivalnetwork dot com and get
0:42
your free newsletter and gift. Financial Survival Network now more than ever. And
0:52
welcome. You are listening to and watching the Financial Survival Network. I'm your
0:56
host, Carrie Lutz. Well, we're midway through April. Things very murky
1:02
looking out there, and maybe you need a different approach towards the market.
1:07
But with AI and all the technology going on now, how is that affecting
1:11
the way you invest, the way you look at investing, the way you
1:15
screen potential investments. Well, David Merritt is here with us now, David,
1:23
you are kind of a groundbreaker disruptor here in the financial sector, one
1:30
of the original quants using quantum analysis to allocate capital, and you kind of
1:40
become kind of a legend in this area, and we're thrilled to have you
1:45
on. We don't really usually get somebody who's really into the technical weeds and
1:51
able to handle huge volumes of data and actually get draw conclusions from it,
1:59
So we're thrilled to have you on. And my first question for you is
2:02
AI got to be making an impact in your field now, not just in
2:08
the amount of data you can you can handle and analyze, but in the
2:15
conclusions that you draw from that analysis. And you briefly tell us what AI
2:21
is, what kind of havoc it's reeking in your field? Yeah, sure,
2:24
Kerrie, and I'm delighted to be with you today. So sure AI
2:29
is as having a big impact. You know for quantitative investors like us,
2:37
You know, we're used to dealing with lots of numbers, but I think
2:43
even even for us, AI is really a huge step level advance. So
2:50
I'll let me explain that a little bit, so you know, what we
2:54
try to do as quantitative investors is really accomplished. Two things. One is
3:00
to have a high market awareness. Right. We want to consume as much
3:04
data as possible, as much relevant data as possible. Tease out what is
3:07
relevant and isn't relevant, sort of from the noise is one part of it.
3:14
So have broad market awareness, not look at a small subset of numbers
3:17
and kind of be biased by that. The second thing we try to do
3:21
is we try to invest in a very systematic way, right, have a
3:23
very repeatable investment process, so that no surprises or deviations come up in the
3:30
way we make decisions. Right. And then when we think about how AI
3:35
is affecting that, I think of the affecting in two ways, in terms
3:40
of the actual way we invest. All of a sudden, now we have
3:46
this thing called generative AI, which means we can actually not just predict,
3:51
which we've been able to do for a long period of time. I started
3:53
my first artificial intelligence company back in nineteen ninety nine and in the endurnet search
4:00
engine space. But that's essentially machine learning algorithms and doing prediction. But what
4:05
generative AI allows us to do is to ask questions and have the machine go
4:12
off into that all of that deep data and leverage the learning it can extract
4:16
from us and come back with answers, right, And that's the new thing
4:20
here is how easy it is to get not perfect as you mentioned, but
4:26
high quality answers from from even larger amounts of data than we could have imagined
4:32
a year ago. So that allows us to, for example, say,
4:38
you know, what do you think of the most recent conference call for a
4:42
general electric you know or what have you? These kinds of questions, and
4:47
then the other way it's affecting our business and we can These are two different
4:51
branches, of course, but as you know, obviously we're investing in the
4:55
ai revolution as well. Okay, so have you been surprised by the answers
5:00
you're getting back and how quickly this ai AT put in quotes revolution, but
5:05
it truly is a revolution, how quick it's going, how how profound the
5:11
changes are. So Uh. I think we were really you know when Chad
5:15
GPT came out. Uh, you know, we were all shocked at how
5:20
how quickly we could get good answers to all kinds of general questions. That
5:26
just shocked everybody into the world. But the interesting story I think on that
5:30
is, and I've just come back from the Nvidia GtC conference a few weeks
5:34
ago. But and you know, listening to to tech industry people kind of
5:40
talk about their own surprise was. You know, it's long been thought that
5:46
if you put less information into a model, but it was higher quality information,
5:50
you'd get a better answer. And a few years ago some data scientists
5:55
got thinking, well, what if we just did a crazy thing and us
6:00
through the whole kitten kaboodle at it. Let's just throw all every piece of
6:03
information that's ever been on the internet, you know, kind of any air
6:06
quotes here but pretty close to it, you know, at it and see
6:11
if the model gets better. Now, in order to even test that,
6:14
you've got to have billions of dollars of chips, right, which you know,
6:16
so it took some time, but anyway, they did it, and
6:19
what they found was, much to their surprise, not only did the models
6:25
get better, they got way better. Wow. Uh. And so that's
6:30
the core of the revolution, if you will, If you were to summarize
6:34
this generative AI, what's different about all the machine learning algorithms that have come
6:41
before it and other artificial intelligence, other algorithm can before it was the learning
6:46
that if we just threw massive amounts of data at it. We could get
6:50
a step level increase in what we could get out of it, what it
6:55
could do for us. That's incredible. And yeah, things that So that
7:00
got to change your view on things a bit, because things that you think
7:04
are trivial, I think are trivial not important. The AI looks at them,
7:11
at them in a different way. And you know, it's like the
7:14
butterfly effect, right, or you throw you throw a pebble in a stream,
7:18
and in a thousand years that pebble will have diverted the stream as water,
7:24
you know, flows to the point of least resistance. It's kind of
7:28
like that kind of awakening here, isn't it. Yeah, it is,
7:31
it is. It's exactly that. And it's a good analogy. And but
7:35
but to compare human beings, of course, like let's compare it to human
7:40
beings for a second. Human beings, of course, we don't. I
7:44
don't. I never read the entire Internet, right I got If I spend
7:47
my entire life twenty four hours a day, never sleeping, never eating,
7:51
just reading, I would never get through the internet. Right But yet,
7:56
you know, and the same for everyone. But yeah, we can be
7:59
really small about a lot of things. Right, we can learn history,
8:01
we can learn mathematics, we can learn physics, we can have complex do
8:05
complex reasoning. Right, we can do all these things and we don't even
8:07
need to read all of the Internet. So just to put I put that
8:11
out there just to say, the machines are still pretty stupid in this sense
8:15
that they need to read the entire Internet just to be able to you know,
8:18
write English, you know, so to speak. So they're not as
8:22
smart as we are in that sense. Right, our ability to reason off
8:28
of small take take small amounts of information and use that very efficiently to do
8:33
amazing things. That's still the domain of human beings. You know, obviously,
8:39
you know, technology advances quickly. You know, we should expect AI
8:43
very roughly as an automation tool, that its productivity should roughly double every two
8:50
years, because it's on Moore's law, right, so roughly duffele every two
8:54
years would be the kind of long term. And so if you think it
8:58
gets it gets doubly smart every two years, it can. You know that
9:03
that that that can move along quite quickly. So we'll see where the future
9:05
is. But today, yeah, it consumes huge amounts of information. It
9:11
needs to consume humans amounts of information, but it can make use of it.
9:16
Okay, so what about the efficiency of it? You know, basically,
9:22
I think in another five six years, most of our are a large
9:26
portion of our electrical infrastructure is going to be going to running Nvidia processors and
9:33
other AI processors. Is that a sustainable model? So yeah, of course
9:39
the answer to that is no, because when you look at the numbers,
9:43
you know, it's I think it's I think data centers today and don't quote
9:46
me on this, but I think data centers today are two or two percent
9:50
of US electricity consumption, and over I think the next decade it's supposed to
9:54
data spenders are supposed to go to four percent. I think something like that.
9:58
It's another word doubling in a very short period of time. Uh,
10:03
These these chips require huge, huge, huge amounts of energy. In fact,
10:07
some of the bigger data centers that are being planned are being planned thinking
10:13
you're going to have to build a small nuclear reactor next to them, right,
10:16
so you're not just building a data sending anymore. You're building the electricity
10:20
to power it as well at the same time. Right, So all many
10:22
nuclear reactors or you know, huge amounts of wind farms or solar solar cells,
10:31
so it's probably not sustainable. No, I mean again, you know
10:35
we have to if if AI is going to progress a long More's law,
10:39
it's going to do so in part by the way the chip industry has done
10:43
today, which is, you know, every iteration of chips lowers the wattage
10:48
consumption per unit of work. So just for example, the Hopper chip,
10:52
which is the nvidious current chip, which is currently powering the AI revolution,
10:56
is being replaced by the Blackwell chip. That's what Jensen introduced in his keynote,
11:03
and the black World chip. When you put it all together and you
11:07
just finalize all the numbers, it's about twenty five twenty thirty percent, you
11:11
know, more efficient than the previous generation. So those those gains will have
11:18
to continue to happen because otherwise the energy numbers just get will constrain what AI
11:24
can do in the future. Okay, all right, so you know as
11:28
far as your profits, your funds profits go, and you've kind of been
11:33
doing AI before it was called AI, right, but it's reaching the point
11:37
now where it's in everybody's awareness on the planet. You know, are your
11:43
competitors going to be able to catch up. Is it going to rationalize your
11:48
business to the point where everybody's got the same tools and they're able to basically
11:54
replicate your outcomes? So let me draw on in now. The answer of
11:58
the short answer is no. I think. I think it becomes more difficult
12:07
for discretionary managers overall, and it becomes easier for systematic managers, right,
12:13
some sort of quant data based investment type of managing versus description. I think
12:18
it's the bigger trend. But what I think about is I think about when
12:24
data that we have the data revolution, Right, the last twenty years have
12:28
been about you mentioned it earlier, big data, more availability of data,
12:33
cheaper data sets, more variety of data sets. All this stuff has been
12:37
going on now for a while. And I remember the arguments early in the
12:41
day, Well, now data is a commodity, so you can't make any
12:45
money off of data anymore, and it's clearly not true. There are clearly
12:48
people who can find things in data and people can't. In fact, it
12:52
lent itself, the big data revolution didn't commoditize data at all. Maybe for
13:00
a purchasing point of view, right, the data set's got cheaper, but
13:03
it didn't commoditize it in terms of the use cases and what we can get
13:05
out of it, because it turns out the more data you have, the
13:11
more smartly you have to you have to look for the needle in the haystack,
13:15
because the hate stack got bigger, right, So I don't I don't
13:18
think so it's always possible. You know what you laid out. But I
13:24
think finance is such a complex business. It's always increasing in complexity. There
13:30
are more and more ways to trade, there are more vehicles to trade through.
13:33
Information gets processed quicker than it used to before. I think it's it's
13:37
more about staying on top of this these changes, first the data revolution,
13:43
now machine learning, than machine learning, now artificial intelligence. It's about staying
13:48
on top of these so you can stay at the top of the game.
13:50
So there's a question a little a field, but with your knowledge, I
13:54
think you'll know the answer. Or high frequency trading outfits Citadel is AI going
14:03
to give them yet another leg up on the general market. So that's interesting.
14:09
I'm not a high frequency trader, so I'm not gonna answer it.
14:13
Be able to answer that directly. Maybe somebody who is ivrogency trader, but
14:16
I didn't note that. Last week, you know, Citadel announced that they
14:24
were partnering I believe it was with Microsoft, and they were moving a bunch
14:30
of their in house data processing onto the I believe it is the Azure cloud.
14:35
It was either Google's cloud or Zero Club anyway, forget which cloud it
14:39
is. They were moving on to somebody's third party cloud. That's a huge
14:43
thing. I mean today out of d out of out of security considerations,
14:46
and also out of just proprietary technology considerations. People in the finance business do
14:52
all of this stuff internally for the most part. You don't just move your
14:56
entire data operations to a third party cloud. That's probably a signal of something
15:03
happening there that's different than has been happening in the industry for the last twenty
15:07
years, in that they may be saying, you know, there's some uncertainty
15:15
here about you know, our investment in buying our own stuff, and maybe
15:18
we It could be that what I sort of perceive it is, the technology
15:24
that's going to be powering AI could change very very rapidly. Right today it's
15:28
all Invidia chips, but maybe tomorrow it's somebody else's chips that it's going to
15:31
could change very rapidly, and therefore investing all of your money and you need
15:35
to hold those chips for at least you know, three, five, seven
15:37
years, depending on what type of chip you're invested in. You know that
15:43
there's uncertainty there, and so it makes it better to go with a cloud
15:46
provider and offload that risk. That there's a revolution and who's providing the best
15:50
chips, something like that could be going on. So there may be some
15:54
interesting things to exploit there. I mean, they spent the billions of dollars
15:58
left their own proprietary fiber. If you read Michael Lewis's book The Flash Boys,
16:04
they spent billions of dollars running it to get that tenth of the second
16:08
advantage. Maybe AI will give additional advantages, but take away that the timing
16:17
advantage. You have to hold your money to open to those possibilities right now,
16:21
because you know you're in a nascent industry generative AI. AI. Itsels
16:25
on a nascent but jenerative AI is and I think what we'll call it,
16:30
looking back in time, we'll call the AI revolution is nason Right now,
16:33
and when you're in a Nason industry, a lot of there's always a lot
16:37
of flux at the beginning. A lot of things can change quickly. Yeah,
16:41
that's for certain. All right. Well, hey, we really appreciate
16:42
your coming on, David, sharing your experience and your wisdom with us here.
16:49
People want to find out more about what you're doing, connect with you
16:52
on the web. How do you do that? Oh? Sure, please,
16:56
you know, visit our website Reguarie blog on the website and talk about
17:03
these exactly these kinds of issues in a more at lengthy sense as well as
17:10
there's lots of things to see about what we do and what we offer in
17:12
terms of investment products. And you can visit us at marketinfunds dot com.
17:18
That's m A r K I n f U n DS dot com. And
17:22
once again I'm David marav Marke and Asset Management. All right, great,
17:27
And the link is in the show notes this interview on Financial Survival Network dot
17:32
com. Make sure you go there, sign up for your free newsletter,
17:36
then click that link you get right to David's site. David, appreciate you
17:38
coming on. We'll definitely have you back on as these things develop and unfold
17:42
and the impact of the revolution that's taking place now really becomes apparent to everybody,
17:49
and maybe you can give us a little leg up on it. We
17:52
appreciate you coming on. Thanks so much, Carrie for having me, Thanks
17:56
for listening to carry Letz's Financial Survival Network, your solution to today's trying times.
18:03
For the latest, go to Financial Survivalnetwork dot com. Financial Survival Network
18:08
now more than ever,