00:00:00
a lot of people complaining about me uh
00:00:01
watching Fox News videos they say oh you
00:00:04
keep watching Fox News Fox News is fake
00:00:07
well we're watching CNN now so stop
00:00:08
crying stock market's open just a few
00:00:11
minutes ago and it's not pretty down big
00:00:15
especially the tech heavy NASDAQ c as
00:00:17
Matt Egan is here to explain what we're
00:00:20
seeing and why Matt yeah John we've got
00:00:22
a bit of a tech sell off this morning
00:00:24
and it's being caused by Earth
00:00:26
shattering developments in the AI space
00:00:29
let's take a look at this you see down
00:00:30
down more than 140 points right now the
00:00:33
S&P solidly lower but the real action is
00:00:35
over here in the NASDAQ 600 points lower
00:00:38
nearly 3% on track for one of well again
00:00:40
it's a Vibe right I mean I think that a
00:00:42
lot of stock market stuff is like
00:00:44
straight up just Vibes and like if the
00:00:47
vibe is that China is beating us the
00:00:48
[ __ ] out and it's like basically they
00:00:51
showed this they I mean they pretty much
00:00:53
like stealth launched this it's not like
00:00:55
people like everybody knew this was
00:00:58
coming but
00:01:00
it's one thing to know it's coming and
00:01:02
it's another for it to be here and so
00:01:04
it's like if I'm a if I'm a Gambling Man
00:01:08
and somebody plays this card and this
00:01:10
card is such a strong play and this
00:01:13
person hasn't played any other cards
00:01:14
before then really I'm going to think to
00:01:17
myself I bet they have more
00:01:20
cards and that's probably why I think
00:01:22
it's going down from those vibes and I
00:01:25
think it's more complex than that
00:01:27
obviously but I think that that's that's
00:01:30
at least a little bit of the essence of
00:01:33
it worst days in two years and here's
00:01:37
because they're directly correlated
00:01:39
Chinese startup that few people had ever
00:01:41
heard of until the past few days and it
00:01:43
has emerged as a real player in the AI
00:01:47
arms race it's called Deep seek and
00:01:50
investors and I would imagine officials
00:01:52
in Washington are stunned to learn that
00:01:54
deep seeks AI model has developed
00:01:57
technology that can actually be
00:01:58
competitive with open Ai and Google and
00:02:01
xai and all these more and another thing
00:02:03
about why uh it's coming in and why it's
00:02:06
such a big disruptor is that it's as far
00:02:08
as I know it's open source and free so a
00:02:11
lot of these other services you need to
00:02:13
pay money in order to activate them and
00:02:15
use them but deep seek is free so it's
00:02:19
like now it's like imagine somebody's P
00:02:21
you know it's like imagine like oh I'm
00:02:23
investing in McDonald's well McDonald's
00:02:25
is great and uh but what happens when
00:02:28
somebody is able what happens when a
00:02:30
wizard comes in and makes a shop next to
00:02:33
McDonald's and he can just immediately
00:02:35
[ __ ] summon free hamburgers and just
00:02:37
give them to everybody well less people
00:02:40
are going to go to [ __ ]
00:02:44
McDonald's abish players it's only been
00:02:46
around for a bit but deep has already
00:02:49
vaulted to the top of
00:02:51
the apple as the most downloaded you're
00:02:55
you're you're making the assumption that
00:02:56
because it's Chinese the reason why
00:02:57
they're adding it in is a SC in order to
00:03:00
have Americans give them more
00:03:01
information so the Chinese government
00:03:03
can use it against us I see what your
00:03:04
point is you can use deep seek without
00:03:06
an internet connection that's my
00:03:08
understanding
00:03:09
again passing chat PT which is pretty
00:03:12
shocking think destroys that argument
00:03:14
veteran uh investor billionaire Mark
00:03:17
Andre legendary tech investor he said
00:03:20
that deep seek is quote one of the most
00:03:23
amazing and impressive breakthroughs
00:03:26
that he has ever seen now the most
00:03:29
stunning thing here is it necessarily
00:03:31
that China has developed a pretty good
00:03:33
AI app it's how cheap it is deep seek
00:03:38
says that their AI model only cost
00:03:42
$5.6 million now we don't know that but
00:03:45
if that's true that is pretty stunning
00:03:48
anthropic one of the leading AI
00:03:50
companies has said that it cost about a
00:03:52
100 million to one billion dollars to
00:03:55
develop an AI model we know that Mark
00:03:57
Zuckerberg The Meta CEO says that his
00:03:59
company plans to spend 65 billion
00:04:02
dollars on AI so John look this is all
00:04:05
really questioning sort of the
00:04:07
foundation here of the AI boom which is
00:04:09
that it requires a lot of spending which
00:04:11
is that the US is running away with the
00:04:13
AI arms race and by the way I think it
00:04:15
makes sense that China is able to do it
00:04:17
cheaper uh they could have gotten the
00:04:18
information from other places they maybe
00:04:20
didn't have to follow as many
00:04:21
regulations there's a lower standard of
00:04:23
living which requires which which
00:04:25
results in a lower cost of of of product
00:04:28
of production and so like there are a
00:04:30
lot of things that that are a factor now
00:04:33
I'm not saying that of course meta and
00:04:35
open AI might have morons that they're
00:04:37
spending too much money on on the
00:04:38
product but I'm saying that there are
00:04:40
also like larger you know just like
00:04:42
overarching factors that would cause
00:04:45
this to happen on its own I think black
00:04:46
myth Wukong is a good parallel to this
00:04:48
black myth Wukong was developed for I
00:04:50
think uh less than like uh $70 million
00:04:54
and it was like a massive game but like
00:04:56
how much would that game cost to make in
00:04:57
America considering like cost of living
00:04:59
a adjustments probably more than like
00:05:02
200 also questioning some of the big
00:05:05
positives that have pushed markets to
00:05:07
record high it's a guess a lot of people
00:05:09
put a lot of money into Ai and now
00:05:11
they're wondering if that money is
00:05:13
needed the way that some of these
00:05:15
American companies have said it is exact
00:05:17
we just don't know if the claims being
00:05:18
made by these Chinese companies this is
00:05:20
moving so fast it's stunning all right
00:05:21
we're going to stay on top of this as we
00:05:23
said big drops especially in the NASDAQ
00:05:26
speaking a lot more about deep seek it's
00:05:28
one of these big stocks um that has
00:05:31
really changed the game it seems in just
00:05:33
a couple of days uh without warning uh
00:05:37
big Known Unknown Chinese firm that's
00:05:39
another reason why it [ __ ] up the stock
00:05:41
market people hate uncertainty people
00:05:44
are terrified of
00:05:45
uncertainty rattling the US market Shel
00:05:49
Palmer is CEO of the Palmer group and a
00:05:51
professor of advanced media he joins me
00:05:53
now from New York great to have you with
00:05:56
us great to be here it seems like
00:05:58
there's a bit of a coup happening in the
00:05:59
AI space tell me about deep seek what
00:06:01
you know about it I mean it's a startup
00:06:03
it's oneyear old company um a lot of
00:06:05
people are downloading it and the
00:06:07
question is is it taking on the big guys
00:06:10
in the United States so deep seek is um
00:06:14
a company out of China it I think that I
00:06:17
think that the players in the AI space
00:06:19
knew this was going to happen and I
00:06:21
think that's the reason why they were
00:06:22
working with Trump ahead of time to
00:06:24
develop the Stargate initiative and I
00:06:26
think that's the reason why Trump signed
00:06:28
away a lot of the uh AI limitations and
00:06:31
regulations that Biden put in I think
00:06:34
they knew this was
00:06:37
coming and I I think that in the context
00:06:40
of them knowing that it was coming and
00:06:41
investing all this money into it it
00:06:43
means that also they have a confident
00:06:47
perspective that they can still come out
00:06:49
ahead and I feel like I think VC said
00:06:51
this on Twitter I think it was a really
00:06:53
good idea a really good point is that
00:06:55
this is kind of like a Sputnik moment
00:06:57
for America is because like you know for
00:07:00
context Sputnik like so we had like the
00:07:02
space race in like the' 60s right and
00:07:04
Sputnik was the first thing that got
00:07:06
sent up into space and it was done by
00:07:08
the Russians which made us here in
00:07:10
America just go [ __ ] like massive
00:07:13
Whole Hog big dick on getting people to
00:07:16
the Moon right the Space Race exactly
00:07:18
and so I I think that this actually
00:07:20
could be a very good motivator for uh a
00:07:22
lot of the companies here to like really
00:07:23
push because I do think that ASI uh like
00:07:27
artificial super intelligence
00:07:30
I think that it could be
00:07:32
more
00:07:35
um uh what's the word for it more
00:07:37
effective and more powerful effectively
00:07:40
than a nuclear
00:07:43
bomb is a startup as you said it's about
00:07:45
a year old they have a model called Deep
00:07:48
seek V3 and in the last little bit they
00:07:51
released R1 which is a reasoning engine
00:07:54
and it's open source on an MIT license
00:07:58
you can download it magic here is that
00:08:01
instead of taking months to train and
00:08:03
hundreds of millions of dollars it took
00:08:05
under two months to train and cost under
00:08:07
according to the company under $6
00:08:09
million to train when you download it
00:08:12
you can post-train it and pre-trained
00:08:14
Transformer models need to be postt
00:08:16
trained in order to be valuable and the
00:08:18
compute cost is a fraction we're talking
00:08:21
a few percentage points fraction of what
00:08:23
it costs uh to use the large language
00:08:26
models from the big hyperscalers and
00:08:28
foundational mod Builders like open AI
00:08:31
or anthropic or or or any of the other
00:08:34
um large foundational mod Builders so
00:08:37
this cost savings along with the the
00:08:41
basically lightweight of the model
00:08:43
meaning you can download it and run it
00:08:44
on a laptop has sent kind of Shivers
00:08:47
through the SP that's also been like a
00:08:49
huge level of criticism that a lot of
00:08:52
people have had against AI is that AI is
00:08:55
so energy intensive and now that that
00:08:58
energy intensive is being uh you know
00:09:01
optimized I think that's going to make
00:09:03
it harder to criticize AI also keep in
00:09:06
mind uh some people are saying that this
00:09:08
might be different than the Space Race
00:09:10
because it's not Nations I would
00:09:11
disagree with you I think that Trump
00:09:15
absolutely put America in the ring for
00:09:19
AI by doing Stargate I think they
00:09:22
absolutely he absolutely did and then
00:09:25
also every company in China is directly
00:09:28
linked they have to have big companies
00:09:30
have to have direct relation to the CCP
00:09:33
the Chinese Communist party so I think
00:09:35
that yes you're right that it's
00:09:37
different but it's basically like a
00:09:39
proxy war this time but be very aware
00:09:43
that I think this is absolutely two
00:09:45
companies that are working with or sorry
00:09:47
two countries that are working with
00:09:49
companies to overtake each other that's
00:09:52
what I think is
00:09:53
happening that that's my take on
00:09:56
it because what this really means is
00:09:59
there's a pro possible future where
00:10:02
algorithmic efficiency beats Brute Force
00:10:05
computation and the entire AI industry
00:10:08
in the United States and around the
00:10:09
world has been thinking about highlevel
00:10:12
computation and Big Data Centers and a
00:10:14
lot of energy use and chips and now
00:10:16
they're thinking maybe
00:10:19
not well there my understanding is that
00:10:22
there is a hypothesis that the chips
00:10:24
that they're using are not ones that are
00:10:26
available yet that were produced by
00:10:28
Nvidia they were the h100 chips that's
00:10:31
what Elon Musk said on Twitter so I I
00:10:33
just I don't I don't have enough
00:10:34
information about that to like really
00:10:36
know whether it's true or not but I
00:10:38
think that maybe that's what's happening
00:10:40
there's a good chance there's no proof
00:10:43
yeah so it's it's a really good one uh
00:10:46
to look at in terms of the numbers
00:10:47
because you're saying and and deep seek
00:10:49
saying that they spent just $5.6 million
00:10:52
on this new AI model meta last week said
00:10:54
it would spend upward of $65 billion how
00:10:57
come deep seek was able to crack this
00:10:59
code and do it so much more efficiently
00:11:02
I wonder and the big guys in the US
00:11:06
unable to do so first of all it's hard
00:11:08
to speculate but let's just say the
00:11:10
following there is a cliche that I
00:11:11
believe is a cliche for a reason there
00:11:13
are more honors students in China than
00:11:16
there are students in the United States
00:11:17
you're talking about a very deep pool of
00:11:19
very smart people that's thing one two
00:11:22
they've been restricted from getting all
00:11:24
of the tools they need from the West
00:11:26
including specifically the kind of chips
00:11:28
that would do the computation they need
00:11:30
to do the advanced computation this is a
00:11:33
a mathematical solution to a hardware
00:11:36
problem it's called algorithmic
00:11:37
efficiency they've actually written code
00:11:39
that's more efficient than the code
00:11:41
that's being used generally in the space
00:11:44
throughout the world so how did they do
00:11:46
it it's open source you can see their
00:11:48
research they've been very good and
00:11:50
transparent about publishing their
00:11:52
research they haven't exactly shown the
00:11:53
secret sauce but it is something it's
00:11:55
like yeah it's like video games right
00:11:57
it's like when video games aren't
00:11:58
optimized and it's it's like on unreal 5
00:12:00
and like you look at a tree and like
00:12:01
your frame rate goes to seven it's
00:12:04
like yeah that's what's
00:12:06
happening that we'll be able to
00:12:08
replicate and by the way the markets are
00:12:10
reacting in a knee-jerk and most people
00:12:12
don't understand any of what I'm about
00:12:14
to say to you exactly no one in the
00:12:17
United States no American no company
00:12:20
anywhere on Earth to be fair is going to
00:12:22
lay down and die because a couple of
00:12:24
Engineers figured out a better algorithm
00:12:27
math is math and engineering is
00:12:29
engineering and these algorithmic
00:12:31
efficiency Co um approaches will be
00:12:34
figured out worldwide now does that mean
00:12:36
for invidious chip prices what does that
00:12:38
mean for nuclear power plants for data
00:12:40
centers what does that mean for future
00:12:42
data centers the jury's still out by the
00:12:44
way no one's replicated this for all we
00:12:47
know this is uh the Sputnik moment or
00:12:50
this is the Cyber Pearl Harbor where
00:12:52
where they faked this whole thing and it
00:12:54
really wasn't done for five and a half
00:12:56
million dollars in two months we don't
00:12:58
know my guess is it was done for five
00:13:01
million $6 million and was done in two
00:13:03
months that's my guess and we're all
00:13:05
going to have to rise to the challenge
00:13:06
and by the way I am confident that
00:13:08
everybody
00:13:10
will I think this is also another really
00:13:12
big component is that like when you
00:13:14
accept mediocrity and you accept people
00:13:16
that are shitty and like bad performance
00:13:21
you basically set yourself up to lose to
00:13:23
people who
00:13:25
don't because that's it like it's like
00:13:28
oh well whenever it's oh well things are
00:13:29
fine enough the way that they are other
00:13:31
people will come in and they will
00:13:33
outperform
00:13:35
you yeah this is the thing we were
00:13:37
talking about the the race uh the AI
00:13:39
race and I think this is pretty much
00:13:41
indicative of of the race that has truly
00:13:44
just begun as you say you can crack this
00:13:46
code and Nvidia Engineers I don't think
00:13:49
that's what's being I don't think that's
00:13:50
what's being said no nobody thinks that
00:13:54
it's a mathematical uh solution then
00:13:56
you've got people that could work on it
00:13:58
but you touched something in terms of
00:13:59
the hardware and the gpus and Nvidia and
00:14:02
so forth and even what it means for
00:14:04
electricity consumption for data centers
00:14:07
if what we're hearing from Deep seek is
00:14:09
in fact all factual what does it
00:14:11
actually mean does this change the
00:14:13
calculation completely for AI so at the
00:14:16
moment it is much much less expensive to
00:14:19
run deep seek R1 locally than it is to
00:14:23
or even in in the cloud that word keep
00:14:25
in mind locally that means that you can
00:14:27
run this without an internet
00:14:30
connection that's [ __ ]
00:14:33
huge that's
00:14:35
insane how than it is to run any of the
00:14:38
other models so the question is not is
00:14:41
this better or worse the question is
00:14:43
going forward is algorithmic efficiency
00:14:46
this ability to run smaller models more
00:14:50
efficiently going to be a better path
00:14:52
forward than the Brute Force massive
00:14:55
compute models that we've used so far at
00:14:57
open Ai and anthr ropic and at Google
00:15:00
Gemini and at llama for metas llama we
00:15:03
don't know the answer to that right now
00:15:05
it points towards two Futures one future
00:15:08
is algorithmic efficiency the other is
00:15:10
Brute Force compute there's no one in
00:15:12
the world who can answer this question
00:15:13
right now it's all speculation well I I
00:15:16
I I don't know I mean how can you say
00:15:18
that like the obvious answer is both
00:15:20
like whoever creates the best
00:15:22
equilibrium between the two things
00:15:24
they're the ones who are going to win
00:15:26
like these aren't mutually exclusive and
00:15:29
they've probably already done that I'm
00:15:31
will tell and as you've just said it's
00:15:33
very early days what is like the
00:15:36
beginning of the race we're nowhere near
00:15:38
the end of this race um we've never seen
00:15:40
anything come this quickly we've never
00:15:42
seen exponential Improvement that has
00:15:44
really hit us at this speed so it's
00:15:47
accelerating and and we're going to all
00:15:49
learn together that's just the way it's
00:15:51
going to go and the markets are going to
00:15:52
follow along because markets do that I
00:15:55
think he's pretty much right I mean like
00:15:57
I I I think that it's very hard to tell
00:15:59
super computer is the Finish Line well I
00:16:01
think that a lot of people are banking
00:16:03
on things like Quantum Computing and
00:16:05
[ __ ] like that to like really kind of be
00:16:07
at that edge but I I have no idea right
00:16:10
I mean I don't know enough about it like
00:16:13
I know a little bit about it but I don't
00:16:15
know enough about it Forth video is good
00:16:17
yeah yeah no I'll look at this probably
00:16:19
well this is the one that they made a
00:16:21
little bit ago but um it does seem like
00:16:25
obviously that this this thing's going
00:16:27
to get huge I mean it will
00:16:29
and I think also you're probably going
00:16:30
to see more people developing their own
00:16:32
models as well I think this is like
00:16:34
Edison versus Tesla well it's very hard
00:16:36
really for me to say about that I I
00:16:38
don't know but either way I I think this
00:16:41
is really
00:16:42
interesting and that's only just beg
00:16:45
yeah did you see Nvidia stock no I
00:16:46
didn't let me look at it uh give me a
00:16:49
second I'll pull it up uh Nvidia
00:16:54
stock not so good wow went down a lot oh
00:16:58
my God oh my God it didn't go down a lot
00:17:01
it went down by
00:17:03
20% holy
00:17:10
[ __ ] oh my
00:17:12
god did Pelosi pull out yeah I don't
00:17:16
know yet uh we'll have to see what
00:17:17
happens it crashed I mean I wouldn't say
00:17:20
this is a crash I I I I can't say it's a
00:17:23
crash yet like if you look at 6 months
00:17:27
it's still above where it was at for 6
00:17:29
months and I don't know about you
00:17:33
guys what do you guys think do you think
00:17:35
that the the new chipsets that they
00:17:37
announced this year the 509s and
00:17:39
everything do you think that
00:17:43
they like my opinion is like it was a
00:17:45
six or a seven out of 10 a five wow yeah
00:17:49
mid right kind of disappointing I think
00:17:52
that this is a problem also is that
00:17:53
Nvidia has been like such a meteoric
00:17:56
success and because it's been such a
00:17:58
meteor oric success people expect that
00:18:01
to continue right they expect it to just
00:18:03
keep having these like crazy updates and
00:18:05
everything and when that doesn't happen
00:18:08
then people freak out and they're like
00:18:09
oh my God maybe the gravy train is
00:18:11
ending so I I personally think that
00:18:13
Nvidia might be a little bit overvalued
00:18:16
I I think so I think that there's too
00:18:17
much speculation but
00:18:20
um maybe not who knows
