00:00:00
On Tuesday, the New York Times reported
00:00:01
that Meta was looking at downsizing its
00:00:04
AI division as part of a larger
00:00:06
restructuring. Wall Street Journal then
00:00:08
reported Meta has a hiring freeze across
00:00:10
the AI divisions, which is just crazy
00:00:13
because just eight weeks ago, Zuck went
00:00:15
crazy. He was trying to buy Ilia's uh
00:00:19
super intelligence startup. You remember
00:00:21
Ilia declined. So, Zuck poached Daniel
00:00:23
Gross. Meta Aqua hired the scale AI team
00:00:27
and then they agreed to invest 14
00:00:29
billion in that. Sam Waltman claimed
00:00:31
Zuck was making hundred million dollar
00:00:33
offers regularly for open AI talent. I'm
00:00:36
curious, you know, what you think of
00:00:37
that, Sachs? Is the talent war seem to
00:00:40
go insane for a couple of months and now
00:00:43
is pausing. What's your take on all this
00:00:47
I don't know uncertainty boom bust cycle
00:00:49
in such a compressed period of time or
00:00:51
is this just
00:00:52
I mean you you are seeing founders
00:00:54
turning down multi-billion dollar
00:00:56
acquisition offers for startups that
00:00:58
hadn't even released a product yet as if
00:01:00
those types of offers grow on trees and
00:01:04
they don't I mean these types of hundred
00:01:06
million or billion dollar job offers
00:01:08
they don't come along very often I mean
00:01:10
you have to be at kind of the sweet spot
00:01:13
of a boom cycle and you need a huge
00:01:15
company with tons of money that feels
00:01:17
like it's strategically vulnerable and
00:01:19
is at risk of being left behind. And if
00:01:22
you get a confluence of those factors,
00:01:24
then you can get kind of crazy offers
00:01:26
like that, but they don't come along
00:01:27
very often. I think that what Meta is
00:01:30
doing is probably digesting a little
00:01:32
bit. They've now made a bunch of talent
00:01:34
acquisitions. They've done some aqua
00:01:36
hires at, you know, very expensive aqua
00:01:38
hires and they're probably just
00:01:40
consolidating a little bit. Like I said,
00:01:42
I don't think this is the bust part of
00:01:44
the cycle. I don't think that a bubble
00:01:46
has popped or anything like that. I
00:01:48
actually think that we're still probably
00:01:50
early to the middle of this investment
00:01:52
super cycle. And it's just a healthy
00:01:55
correction in sentiment here that people
00:01:57
are realizing it's going to be a little
00:01:59
bit harder and take more work than just,
00:02:02
oh, the AI is going to figure out how to
00:02:03
improve itself and we get to super
00:02:05
intelligence. That was always a little
00:02:06
bit of a fantasy. uh Mera should have
00:02:09
taken the billion dollar offer from Zach
00:02:10
or Ilia should have taken the 30 billion
00:02:12
Zachs. I mean, these things never
00:02:13
happen. And as you're saying,
00:02:15
I mean, I guess it depends how much they
00:02:17
have in the bank account. You know, if
00:02:19
if they're already like billionaires
00:02:20
from whatever they did before, then then
00:02:23
maybe it doesn't matter. But if you were
00:02:25
just starting out, for example, and
00:02:26
didn't have any money in the bank,
00:02:28
that's a pretty hard thing to turn down.
00:02:30
I mean, realistically,
00:02:32
yeah,
00:02:33
they probably they probably sold a bunch
00:02:34
of OpenAI equity at 300 and 500 billion.
00:02:37
They're both OpenAI co-founders, so
00:02:39
they seem like they're free rolling, so
00:02:41
they have no incentive to sell.
00:02:43
Yeah. Which is fine. It's just that
00:02:44
there's a lot of people who've never
00:02:46
been through a bus cycle before. And if
00:02:49
they think that this is the normal state
00:02:51
of the world, they're going to be sorely
00:02:53
mistaken.
00:02:54
By the way, you're right, Sax. It's hard
00:02:56
to build as you know because you did it
00:02:58
a billion dollar company that then exits
00:03:00
for more than a billion. It is hard
00:03:02
right where you justify that valuation
00:03:04
based on fundamentals. So right now
00:03:06
we're in a part of the cycle where you
00:03:08
can justify that valuation based on it
00:03:11
strategic value to a multi-t trillion
00:03:14
dollar market cap company. But that only
00:03:16
lasts while those companies are in the
00:03:18
market for strategic acceleration.
00:03:21
Yeah. if they're behind, if they're
00:03:22
stuck down,
00:03:24
then you have to make your company work
00:03:26
on its own as an actual business
00:03:28
and to get to a $30 billion valuation
00:03:31
based on fundamentals, that is
00:03:33
extraordinarily difficult. I mean, that
00:03:35
implies, you know, multiple billions of
00:03:37
revenue, actual revenue.
00:03:38
Does somebody want to underwrite 500
00:03:40
billion for OpenAI common shares?
00:03:42
Anybody think that that is a good trade?
00:03:44
I'll make the I'll make the bull case.
00:03:45
I'll make the bull case.
00:03:46
Good idea, please. Yeah. I think the
00:03:48
simplest way to make the bull case is if
00:03:50
you look at the kager on their mouse and
00:03:53
the conversion from ma to Dows,
00:03:56
you essentially take a small minor
00:03:58
percentage of the Facebook or Google
00:04:00
terminal arpoo and apply it to some
00:04:03
number of ma
00:04:05
3 four years from now. So if I had to
00:04:07
guess, if you take 4 500 million MADOW
00:04:10
growing at I'm guessing, let's just be
00:04:13
conservative 50%
00:04:14
500 million weekly active users right
00:04:17
now.
00:04:17
750. Okay. So then, so then I would
00:04:19
probably put that at like 500 DAO to be
00:04:23
conservative. It's probably doubling
00:04:25
every two years. So 500 goes to a
00:04:27
billion, billion goes to two in four
00:04:30
years. And at 2 billion DAO, they
00:04:33
generate a tenth of Facebook's revenue.
00:04:36
just to be very conservative and you
00:04:38
probably get to a trillion five
00:04:39
valuation then
00:04:41
right so it's you could you could triple
00:04:42
up on that bet I mean look open AI has
00:04:44
actual revenue I mean they they have
00:04:46
actual revenue because they have
00:04:48
subscriptions and they appear to have
00:04:50
the dominant position in the consumer
00:04:52
space and it is a replacement for a lot
00:04:55
of people for search which is the most
00:04:57
lucrative franchise on the internet and
00:05:00
this is one of their applications so I
00:05:02
actually think that you can make that
00:05:03
case
00:05:05
you pretty comfortably. But can we could
00:05:08
we go back to a point you were making
00:05:09
before JCAL about in the survey that the
00:05:12
attempts to just apply some sort of
00:05:14
generalized AI model didn't work very
00:05:16
well like 95% of the time it didn't
00:05:18
succeed in these large enterprises but
00:05:21
if they used a more specific vertical
00:05:22
application or vertical model or an SLM
00:05:26
approach which is more a smaller
00:05:27
specialized model then it showed much
00:05:30
greater success. I mean that makes a lot
00:05:32
of sense to me is that in order to drive
00:05:34
business value there's a lot of what I
00:05:36
would call last mile problems right like
00:05:39
LLMs need context and so you have to
00:05:41
first of all connect to all of your
00:05:43
enterprise data sources and you have to
00:05:46
prompt them in a very detailed way in
00:05:48
order to get to a good answer and then
00:05:51
you have to validate that answer to make
00:05:52
sure it's not a hallucination and then
00:05:54
you need to iterate on it and so this
00:05:56
idea that you're just going to have one
00:05:58
super intelligence that just figures all
00:06:00
this stuff out it's just not the way
00:06:01
it's playing out in the real world.
00:06:03
You're seeing again a lot of very
00:06:05
specific business problems that have to
00:06:07
be solved. But I I think that this is a
00:06:09
great thing ultimately for the ecosystem
00:06:13
because it implies that you're going to
00:06:14
get lots of vertical applications and
00:06:17
lots of specialized models that capture
00:06:22
value in lots of different markets. And
00:06:24
that that's actually the way that we're
00:06:26
going to drive this throughout the
00:06:27
economy as opposed to it just being one
00:06:29
foundation model eating all the value.
00:06:31
So I I think this is a very healthy
00:06:32
thing for the ecosystem.
00:06:34
Yeah, it makes total sense that the
00:06:35
vertical systems would feel more
00:06:38
deterministic sacks because they're
00:06:40
giving they have a tighter problem set,
00:06:43
a tighter data set to actually come to
00:06:45
an answer whereas the you know
00:06:48
probabilistic you know just asking an
00:06:50
LLM to come up with a business plan for
00:06:52
you. It it just feels like it could be
00:06:54
80% correct, 90% correct as opposed to
00:06:57
99% correct, which the vertical ones are
00:07:00
actually getting very good at getting
00:07:02
the correct answer. Okay, let's talk a
00:07:04
little bit
00:07:04
that last 10% is fundamental and that's
00:07:06
where you get all the the last mile
00:07:08
problems and you have to understand the
00:07:10
the industry in order to solve the the
00:07:12
problems. How you kind of go from let's
00:07:14
say 90% accuracy or effectiveness to 99
00:07:18
which is where the business value
