January 27, 2025 • 20 mins
Watch Alix and Paul LIVE every day on YouTube: http://bit.ly/3vTiACF.
Gene Munster, Managing Partner at Deepwater Asset Management, joins to discuss Chinese AI startup DeepSeek's new model raising questions about the dominance of US tech companies like Nvidia. Kim Forrest, Founder and CIO of Bokeh Capital Partners, discusses today's DeepSeek news and market impact. Josh Pantony, CEO of Boosted.ai, discusses DeepSeek news and the AI space.
Hosts: Paul Sweeney and Alix Steel
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
Bloomberg Audio Studios, podcasts, radio news. You're listening to the
Bloomberg Intelligence Podcast. Catch us live weekdays at ten am
Eastern on Applecarplay and Android Auto with the Bloomberg Business App,
Listen on demand wherever you get your podcasts, or watch
us live on YouTube.
Speaker 2 (00:23):
Let's take it one step further. Gene Munster, managing partner,
co founder for Loop Ventures, really one of the leading
voices on all things technology really for the last twenty
five thirty years. We appreciate getting a few minutes of
his time. Gene, I'm just gonna speak for myself, but
I probably speak for most of my radio and YouTube audience.
I did not know what deep Sick was, deep Sik
(00:45):
was before this morning, but now I'm learning. What should
I be learning? What should I takeaway be today?
Speaker 3 (00:54):
I think that the highest level this whole chapter is about.
The first takeaway is that investors are nervous that this
eighty five percent moving the NASDAC over the past couple
of years is creating an environment where any sort of
ripple in what that AI trade looks like is going
to cause a significant impact on some of these valuations.
(01:14):
So I think that's I think this is a temperature
in terms of where the market is at as far
as kind of the company itself is concerned deep seek.
What's most important here is that they are advancing or
presumably advancing the cost of training these models you've talked
a lot about on the show today. And why that's
important is there's a question about how much hardware is
(01:35):
needed for that, and that obviously impacts the big hardware companies,
and then separately, how does that accelerate the adoption? So
there is the truth is always several layers below the surface,
and I think when it comes to Dee Sep there
are still some pretty significant unanswered questions, and I'll start
with one of probably the biggest one is that this
five to six million dollar training number that has the
(01:57):
market upside down today, that was not their most recent model.
They have an R model. It's unclear, you know, maybe
that was twenty five fifty one hundred million dollars to
train it. It probably was below what the most recent
version of Opening Eyes model, which is anticipated or estimated
to be about a five hundred million dollars in terms
(02:19):
of training. So I think big picture here is that
how we are delivering AI is evolving and there may
be some pretty significant improvements in terms of the ability
to do that at a lower cost.
Speaker 4 (02:33):
So to that pointing, how much if we look at
Envidia in particular, it's down about two hundred and eighty
three points. What part of that is justified?
Speaker 3 (02:44):
So I think the big picture here, I think that
this is an overreaction. I think when you look at
a company like Nvidia, it's a company we do own,
and I think that this is an overreaction, and just
specifically around that, if we can kind of zero in
on this concept that these models are becoming more efficient,
let's just take deep seek at face value that they've
(03:05):
had some sort of improvement. Is that that improvement if
you believe that the US tech companies are competent, that
improvement is around some architectures that have been talked about
for the last couple months. So those companies, the big
tech companies, the companies that are behind all the announcements
(03:25):
last week Meta increasing their CAPEX spend, and I think
that that is all that is a belief that they
still need this hardware. So when it comes back to
AI and comes back to the Nvidia trade, is that
I believe if they are reducing the costs to train
these models that I actually can build. I think a
(03:49):
credible case that that could increase the demand for some
of this hardware if we are getting closer to general intelligence.
If that potential is even closer, it's just going to
be an insatiable arms race. It's going to continue. And
so I understand why and video is down today. I
think it's actually a healthy thing. But ultimately I think
that their business is going to be just fine. We
(04:09):
have to wait twenty one trading days before we hear
from at least in VIDI on that, but I think
it's going to be fine.
Speaker 2 (04:15):
So, Gene, was it a coincidence that on Friday, Meta
upsets capex forecast dramatically, which by a magnitude I've never seen,
from fifty billion to sixty five billion, and then today
we get this news on deep sick? Was Deep seek?
Was that kind of I don't know, coincidence?
Speaker 3 (04:35):
I think, what's I think that's just purely coincidence. But
I do kind of go back to what we're just
talking about a minute ago. Is I think it's important
to note is that that announcement from Meta, if you
assume a Meta is Meta is competent, and I believe
that they are. That announcement came with the full knowledge
of what deep Seek was doing, so it's new to
(04:56):
most of us today. But for these companies who have
been making these huge investments, the concept of what deep
Seak had been trying to build has been They've had
a model that's been out. This has been something that's
been aware for the last couple of months, and so
I think that is important. Is that I think again
that this commentary from on Friday from Zuckerberg about the
(05:16):
increased capex factors in some of these potential breakthroughs that
we've seen with deep Seek. I want to stop short
of saying it's definitively a breakthrough because a lot of
unknowns around how good the models are and what the
true costs are. But I think that the AI investment
phase is still alive and well, this is still very early.
Speaker 4 (05:38):
So if this could be an overreaction, we don't know
fair enough. In Nvidia, obviously an overreaction that stocked. The
way you look at it, does it make you think
though a rating of forty one times estimated PE.
Speaker 3 (05:55):
So from where my head's at and thinking about calendar
twenty six at this point, and based on the calendar
twenty six numbers, I think it's closer to like a
twenty five, and I think that they can grow earnings
more than that. This is where the basically the rubber
hits the road, is that ultimately what happens in calendar
twenty six, all this stuff we're talking about with deep
seek today really impacts Ynvidia in twenty six. And if
(06:17):
in fact this dramatically changes how the kind of hardware
that people need less hardware, then I'm going to be wrong.
But if this does create this accelerated arms race, then
I think that they're going to grow faster. And so
I consider a one peg on that twenty six earnings
is actually attractive valuation relative to the rest of the group.
So I'm comfortable with the evaluation.
Speaker 2 (06:38):
Just a red headline crossing the Bloomberg criminals we speak,
deep Seek says it is subject to large scale malicious attack.
So that's a headline. We'll have more reporting on that
going forward. Gene, just real quickly, we're going to hear
from some of the other big tech companies, Microsoft, Meta, Tesla,
how do you think they're going to address this issue?
Speaker 3 (06:57):
It's obviously this is going to be front and center
the topic, and I think they're probably going to address
it by saying they still have plans to invest meaningfully
more in counter twenty five over twenty four when it
comes to Capex, and I think that that will play
part of a for reassuring we're long ways away from that.
We got two and a half trading days awave before
we start to get some of that commentary before the
(07:18):
mic gets turned over to the other side of the
equation here.
Speaker 2 (07:21):
All right, Gene, thank you so much for joining us.
I know you're super busy today. Appreciate getting a few
minutes of your time. Gen Munster, managing partner co founder
Lupet Ventures, taking a little bit of a I guess
a longer term view, in a more broader view of
what this can mean for the indusuay.
Speaker 1 (07:36):
You're listening to the Bloomberg Intelligence Podcast. Catch the program
live weekdays at ten am Eastern on Applecarplay and Android
Auto with the Bloomberg Business app. You can also listen
live on Amazon Alexa from our flagship New York station,
Just say Alexa play Bloomberg eleven thirty.
Speaker 4 (07:54):
All right, let's get more on this text sell off
here Kim Forrest, founder and CIO of Boca Capital Partners.
She's also a former Chips person, like she did the stuff,
so she knows the things.
Speaker 5 (08:04):
Clearly.
Speaker 4 (08:04):
The deep Seek news is upending most of the tech industry, right.
You get the socks and DEXes down hard, you get
all the hyperscalers are down hard, all the chip makers,
all the power makers all down hard. Is this the
reckoning or is this a by the dip moment?
Speaker 5 (08:21):
Well, that is the question, isn't it. I would say
it's a little bit of both. I think this is
a good level set, re leveling, resetting our expectations. So
the first thing is, we have to really see if
this is true. Now, apparently the company is in deep
Seek is an open source company. They have released how
(08:46):
they've done this, but you know, replication is key, so
we're gonna have to see if this actually works. That's
thing number one, Like, don't get too upset about major
developments until you know that they're real major developments, right,
So that's thing one. Thing two is and I think
this has been troubling me for a very long time,
(09:09):
especially with like last week's Stargate announcement. It seems as
if people have been expecting AI, which we don't know
what the payoff is yet we have a pretty good
idea it might help us out. But we're going to
cover the earth essentially in data centers and use far
(09:29):
more power than we create right now. So those two
things are kind of troubling that people have been making
investments in companies going to produce far more than they
can produce now. And do you see what I'm saying here?
We had these really big expectations, and I didn't think
that they were going to come to happen in short
or long term because of the well physical limitations of it.
Speaker 2 (09:53):
So Kim, at this very early stage, I think I'm
probably representative of most of our listeners and viewers had
not heard of deep Seek before this morning. Sure, and
so that's calling into question kind of what we think
we understood about AI and its implication. It's not just
for technology preference stock market in general. My only question
that I think I have now that has any relevance
is do I have to rethink my spending associated with AI.
Speaker 5 (10:18):
It depends on how long your timeframe is. If it's
the very short term, probably not you you know you'll
get rewarded because well orders are in and all that
kind of good stuff. And by short term I'm talking
a year, but let's say three to five years maybe,
And why is that? Well, if we can really train
models more rapidly with less input, which is a good thing, right,
(10:42):
where were we going to get all this electricity? That
was like my biggest question. But anyhow, back to your question,
I think you know this is part of being an
investors understanding how long was I play? I'm keeping this,
So that's part of the problem. We have two more
(11:05):
problems that are not being clarified by deep Seek. One is,
and remember I used to do this, so I'm way
down the road from most investors in thinking about things.
But as I use it, I'm not getting the results
that I need, And by that I mean it's not
right enough. The error rate that I get back from
(11:26):
the questions that I'm asking are anywhere between five and
maybe thirty percent, which that's huge. Like I can't just
use AI. I have to babysit AI. So these are
problems that aren't solved by anybody at this point. And
then there is how are we actually going to use it?
I think there was I think a Bloomberg report that
(11:48):
was saying that probably two hundred thousand people could be
laid off those lower level entry level people into finance
and then that mid tier that are now you know,
kind of like the AI for investment banking and equity research.
But if we were that wrong, we wouldn't last right, Like,
(12:10):
if you're five percent wrong in investment banking, your career
is really short. So you know, these are problems that
aren't being solved. So that's another issue for investors. How
right How long is it going to be till these
get right enough to replace a person?
Speaker 4 (12:26):
That's such a great point because I was also reading
an article that talked about the limitations so far that
have been seen about Gee, what is it deep Seek?
I want to say, geek see deep Seek was its
lack of inability to discuss jijenping or to give real
handswers on Tanam and square, just those topics that are
(12:47):
near and dear to China's heart. And that's in part
some of the issue that you're talking about. So who
wins in this right now?
Speaker 5 (12:56):
Who wins? Well, I think investors are losing in the
short term. But if you like me think that AI
is a path to productivity, and productivity always wins. I
think you could be a winner. I think maybe companies
like open Ai and Microsoft, you know, by dint of
(13:19):
you know, investing in them, and then everybody else who's
developing AI, they might be winners. Maybe the what is
it Stargate? I don't know why. I can never remember.
That's my break up here. I can ever remember that
because it has nothing to do with stars or gates,
that Stargate might be the loser in those people because
we're not necessarily going to need to cover the earth
(13:42):
in data centers.
Speaker 4 (13:43):
Right, yep, all.
Speaker 2 (13:45):
Right, Kim, thanks so much for joining us. Always appreciate
getting your perspective of your technology background. Kim Farst, founder
and chief investment officer of Book Capital Partners, joining us
from Pittsburgh via the zoom Thing.
Speaker 1 (14:00):
Listening to the Bloomberg Intelligence Podcast. Catch us live weekdays
at ten am Eastern on Applecarplay and Android Auto with
the Bloomberg Business App. Listen on demand wherever you get
your podcasts, or watch us live on YouTube.
Speaker 4 (14:14):
Happy mindy, everybody, Alex Deal here alongside Paul Sweeny. This
is Bloomberg Intelligence Radio. We are broadcasting to live for
Interactive Brookers Studio right here in Midtown Manhattan. Also check
us out on YouTube dot com. Or clearly have the
tech sel off underway really concentrated in certain names the
mag seven obviously, in Vidia, the chip stocks, although Apple's up,
I should say, but in Vidia obviously getting pounded, the
(14:35):
chip stocks getting pounded as well. So we wanted to
continue the conversation about AI and you really need to
rethink this investment thesis. Joining us now is Josh Pantoni.
He's CEO boosted dot Ai Now. The company has helped
dozens of investment managers who's AUM totals over one trillion
dollars and how to implement machine learning in their portfolios.
(14:56):
First off, Josh, have you seen this new geek seek? Geeks?
Why do I keep saying geek seek? It's not geeks
seek deep see see if you try deep seak? Have
you talked to your clients about it? What do you think?
Speaker 6 (15:13):
Yeah? So, I mean it came out I think last Monday.
We've already ran a whole suite of benchmarks against it.
Try that out over a bunch of things, and I'd
say we're pretty familiar with it. I think it's impressive
in a number of ways. It's impressive how how much
power you get for the cost. I think the flip
side is it's actually not quite something that we could
even consider implementing it to our process. And I think
(15:36):
there's a lot of clients that will reach the same
conclusion after they start working with it. Why is that.
Speaker 5 (15:42):
So?
Speaker 6 (15:42):
A lot of how we help our clients is with
things like trying to identify risk and trying to identify
to build out these different workflows for different parts of
the financial process. And a really big part of that
is related to trust. You need to be able to
give answers that you're very confident in. You need to
give answers that you can sort of deeply into it
where it's coming from. And I think one of the
(16:03):
challenge of this model is that it has some very
clear political biases right off the bat. So if I'm
trying to use it to say understand what my portfolio
exposure is to China's going to be a challenge to
use it for things like that. Also, a lot of
the things we're most excited about are things like being
able to do computer use, being able to sort of
teach the machine how to like interact with different apps
(16:24):
and things on the computer, image recognition, trying to extract
out charts and things like that, and a lot of
those kind of capabilities that actually doesn't really have it
and seem to be as sophisticated in so it can
do very well in certain benchmarks, but when it actually
comes to trying to apply it for at least most
of the use cases that we're doing, it's not quite
there yet.
Speaker 4 (16:42):
What is there for you right now? Like what does work?
Speaker 6 (16:46):
I think the most interesting thing for me is actually
on the reasoning side. So of course, like one of
the really big pushes Opening Eyes had right now is
like with one and three, you've got these models coming
in where in theory they can take a task, break
it into subcomponents, and then start executing those components. And
I think in that area it's actually quite good. I
also think, you know, just from a cost basis, it's
(17:08):
extremely impressive they've been able to do. There's been some
controversy about exactly how much it costs to actually train
the model. What I can confirm though, is the actual
inference cost. The cost of running the model is extremely
cheap compared to what you're getting with like oh one
and oh three, and so just the fact is even
possible to do that as a major technology breakthrough.
Speaker 2 (17:26):
If nothing else does Deep Seek just highlight the cost
issue and potentially the I don't know, the movement down
in costs of implementing AI.
Speaker 6 (17:40):
Yeah, the way I tend to think about it is
the cheaper it is to train AI systems, the more
advanced capabilities you can train.
Speaker 4 (17:49):
In the stress you lost your audio Josh, you there, Yes,
I am.
Speaker 5 (17:52):
Here, still hear me. We're good?
Speaker 6 (17:54):
Okay, perfect? Yeah. What the way I like to think
about it is the cheaper it is to train these
AI systems, Then the more advanced capabilities you can train,
the more advanced capabilities you can train, the more use
cases you unlock. The more use cases unlock, the more
you actually see it accelerate. So I actually think this
model will probably cause an acceleration in the capabilities of
(18:16):
other models that are getting built as more folks start
to adopt it. So for me, I actually see it
as completely positive.
Speaker 4 (18:22):
What outside of deep Seek and other models, like, have
you worked with that you did?
Speaker 6 (18:27):
Like yeah, so, I mean behind the scenes, we work
with in thropic models we work with open AI models.
We work with a bunch of fine tune models that
we build in house ourselves. You know. I like to
sort of describe it as an orchestra of different types
of models. And one of the things we've sort of
noticed is there's a lot of differences in capabilities. So
(18:48):
something like the inthroduct models and thropic models tend to
be better at very long context window where you try
and handle like huge amounts of text, whereas something like
the GPD form models tend to be a little bit
better at like foreign language and just sort of the
general verbiage it uses as it's giving outputs. So we
use a whole bunch.
Speaker 2 (19:08):
Where how are your clients and the asset management business,
Josh using AI these days, in these early days.
Speaker 6 (19:15):
Yeah, So the way I like to think about it
is we give them the ability to create this sort
of team of little AI workers where you teach them
how to do some kind of task, and then they're
going to do that task on a continuous basis. So
let's say you to do something like write an investment
thesis on a company, or let's say you had to
do something like write an ESG report or Let's say
you wanted to continuously monitor the world for any kinds
(19:38):
of updates to something that might happen in the air space.
These are all examples of sort of workflows you can
teach the system and have the system start to automate.
Speaker 4 (19:48):
Well, we really appreciate your time. It was so good
to get that perspective. It's good to kind of get
the user mindset in for this. Josh Josh Pantoni, CEO
a boosted dot AI on deep Seek and sort of
the pros and the hans there.
Speaker 1 (20:01):
This is the Bloomberg Intelligence podcast, available on Apple, Spotify,
and anywhere else you get your podcasts. Listen live each
weekday ten am to noon Eastern on Bloomberg dot com,
the iHeartRadio app, tune In, and the Bloomberg Business app.
You can also watch us live every weekday on YouTube
and always on the Bloomberg terminal.
Bloomberg Audio Studios, podcasts, radio news. You're listening to the
Bloomberg Intelligence Podcast. Catch us live weekdays at ten am
Eastern on Applecarplay and Android Auto with the Bloomberg Business App,
Listen on demand wherever you get your podcasts, or watch
us live on YouTube.
Speaker 2 (00:23):
Let's take it one step further. Gene Munster, managing partner,
co founder for Loop Ventures, really one of the leading
voices on all things technology really for the last twenty
five thirty years. We appreciate getting a few minutes of
his time. Gene, I'm just gonna speak for myself, but
I probably speak for most of my radio and YouTube audience.
I did not know what deep Sick was, deep Sik
(00:45):
was before this morning, but now I'm learning. What should
I be learning? What should I takeaway be today?
Speaker 3 (00:54):
I think that the highest level this whole chapter is about.
The first takeaway is that investors are nervous that this
eighty five percent moving the NASDAC over the past couple
of years is creating an environment where any sort of
ripple in what that AI trade looks like is going
to cause a significant impact on some of these valuations.
(01:14):
So I think that's I think this is a temperature
in terms of where the market is at as far
as kind of the company itself is concerned deep seek.
What's most important here is that they are advancing or
presumably advancing the cost of training these models you've talked
a lot about on the show today. And why that's
important is there's a question about how much hardware is
(01:35):
needed for that, and that obviously impacts the big hardware companies,
and then separately, how does that accelerate the adoption? So
there is the truth is always several layers below the surface,
and I think when it comes to Dee Sep there
are still some pretty significant unanswered questions, and I'll start
with one of probably the biggest one is that this
five to six million dollar training number that has the
(01:57):
market upside down today, that was not their most recent model.
They have an R model. It's unclear, you know, maybe
that was twenty five fifty one hundred million dollars to
train it. It probably was below what the most recent
version of Opening Eyes model, which is anticipated or estimated
to be about a five hundred million dollars in terms
(02:19):
of training. So I think big picture here is that
how we are delivering AI is evolving and there may
be some pretty significant improvements in terms of the ability
to do that at a lower cost.
Speaker 4 (02:33):
So to that pointing, how much if we look at
Envidia in particular, it's down about two hundred and eighty
three points. What part of that is justified?
Speaker 3 (02:44):
So I think the big picture here, I think that
this is an overreaction. I think when you look at
a company like Nvidia, it's a company we do own,
and I think that this is an overreaction, and just
specifically around that, if we can kind of zero in
on this concept that these models are becoming more efficient,
let's just take deep seek at face value that they've
(03:05):
had some sort of improvement. Is that that improvement if
you believe that the US tech companies are competent, that
improvement is around some architectures that have been talked about
for the last couple months. So those companies, the big
tech companies, the companies that are behind all the announcements
(03:25):
last week Meta increasing their CAPEX spend, and I think
that that is all that is a belief that they
still need this hardware. So when it comes back to
AI and comes back to the Nvidia trade, is that
I believe if they are reducing the costs to train
these models that I actually can build. I think a
(03:49):
credible case that that could increase the demand for some
of this hardware if we are getting closer to general intelligence.
If that potential is even closer, it's just going to
be an insatiable arms race. It's going to continue. And
so I understand why and video is down today. I
think it's actually a healthy thing. But ultimately I think
that their business is going to be just fine. We
(04:09):
have to wait twenty one trading days before we hear
from at least in VIDI on that, but I think
it's going to be fine.
Speaker 2 (04:15):
So, Gene, was it a coincidence that on Friday, Meta
upsets capex forecast dramatically, which by a magnitude I've never seen,
from fifty billion to sixty five billion, and then today
we get this news on deep sick? Was Deep seek?
Was that kind of I don't know, coincidence?
Speaker 3 (04:35):
I think, what's I think that's just purely coincidence. But
I do kind of go back to what we're just
talking about a minute ago. Is I think it's important
to note is that that announcement from Meta, if you
assume a Meta is Meta is competent, and I believe
that they are. That announcement came with the full knowledge
of what deep Seek was doing, so it's new to
(04:56):
most of us today. But for these companies who have
been making these huge investments, the concept of what deep
Seak had been trying to build has been They've had
a model that's been out. This has been something that's
been aware for the last couple of months, and so
I think that is important. Is that I think again
that this commentary from on Friday from Zuckerberg about the
(05:16):
increased capex factors in some of these potential breakthroughs that
we've seen with deep Seek. I want to stop short
of saying it's definitively a breakthrough because a lot of
unknowns around how good the models are and what the
true costs are. But I think that the AI investment
phase is still alive and well, this is still very early.
Speaker 4 (05:38):
So if this could be an overreaction, we don't know
fair enough. In Nvidia, obviously an overreaction that stocked. The
way you look at it, does it make you think
though a rating of forty one times estimated PE.
Speaker 3 (05:55):
So from where my head's at and thinking about calendar
twenty six at this point, and based on the calendar
twenty six numbers, I think it's closer to like a
twenty five, and I think that they can grow earnings
more than that. This is where the basically the rubber
hits the road, is that ultimately what happens in calendar
twenty six, all this stuff we're talking about with deep
seek today really impacts Ynvidia in twenty six. And if
(06:17):
in fact this dramatically changes how the kind of hardware
that people need less hardware, then I'm going to be wrong.
But if this does create this accelerated arms race, then
I think that they're going to grow faster. And so
I consider a one peg on that twenty six earnings
is actually attractive valuation relative to the rest of the group.
So I'm comfortable with the evaluation.
Speaker 2 (06:38):
Just a red headline crossing the Bloomberg criminals we speak,
deep Seek says it is subject to large scale malicious attack.
So that's a headline. We'll have more reporting on that
going forward. Gene, just real quickly, we're going to hear
from some of the other big tech companies, Microsoft, Meta, Tesla,
how do you think they're going to address this issue?
Speaker 3 (06:57):
It's obviously this is going to be front and center
the topic, and I think they're probably going to address
it by saying they still have plans to invest meaningfully
more in counter twenty five over twenty four when it
comes to Capex, and I think that that will play
part of a for reassuring we're long ways away from that.
We got two and a half trading days awave before
we start to get some of that commentary before the
(07:18):
mic gets turned over to the other side of the
equation here.
Speaker 2 (07:21):
All right, Gene, thank you so much for joining us.
I know you're super busy today. Appreciate getting a few
minutes of your time. Gen Munster, managing partner co founder
Lupet Ventures, taking a little bit of a I guess
a longer term view, in a more broader view of
what this can mean for the indusuay.
Speaker 1 (07:36):
You're listening to the Bloomberg Intelligence Podcast. Catch the program
live weekdays at ten am Eastern on Applecarplay and Android
Auto with the Bloomberg Business app. You can also listen
live on Amazon Alexa from our flagship New York station,
Just say Alexa play Bloomberg eleven thirty.
Speaker 4 (07:54):
All right, let's get more on this text sell off
here Kim Forrest, founder and CIO of Boca Capital Partners.
She's also a former Chips person, like she did the stuff,
so she knows the things.
Speaker 5 (08:04):
Clearly.
Speaker 4 (08:04):
The deep Seek news is upending most of the tech industry, right.
You get the socks and DEXes down hard, you get
all the hyperscalers are down hard, all the chip makers,
all the power makers all down hard. Is this the
reckoning or is this a by the dip moment?
Speaker 5 (08:21):
Well, that is the question, isn't it. I would say
it's a little bit of both. I think this is
a good level set, re leveling, resetting our expectations. So
the first thing is, we have to really see if
this is true. Now, apparently the company is in deep
Seek is an open source company. They have released how
(08:46):
they've done this, but you know, replication is key, so
we're gonna have to see if this actually works. That's
thing number one, Like, don't get too upset about major
developments until you know that they're real major developments, right,
So that's thing one. Thing two is and I think
this has been troubling me for a very long time,
(09:09):
especially with like last week's Stargate announcement. It seems as
if people have been expecting AI, which we don't know
what the payoff is yet we have a pretty good
idea it might help us out. But we're going to
cover the earth essentially in data centers and use far
(09:29):
more power than we create right now. So those two
things are kind of troubling that people have been making
investments in companies going to produce far more than they
can produce now. And do you see what I'm saying here?
We had these really big expectations, and I didn't think
that they were going to come to happen in short
or long term because of the well physical limitations of it.
Speaker 2 (09:53):
So Kim, at this very early stage, I think I'm
probably representative of most of our listeners and viewers had
not heard of deep Seek before this morning. Sure, and
so that's calling into question kind of what we think
we understood about AI and its implication. It's not just
for technology preference stock market in general. My only question
that I think I have now that has any relevance
is do I have to rethink my spending associated with AI.
Speaker 5 (10:18):
It depends on how long your timeframe is. If it's
the very short term, probably not you you know you'll
get rewarded because well orders are in and all that
kind of good stuff. And by short term I'm talking
a year, but let's say three to five years maybe,
And why is that? Well, if we can really train
models more rapidly with less input, which is a good thing, right,
(10:42):
where were we going to get all this electricity? That
was like my biggest question. But anyhow, back to your question,
I think you know this is part of being an
investors understanding how long was I play? I'm keeping this,
So that's part of the problem. We have two more
(11:05):
problems that are not being clarified by deep Seek. One is,
and remember I used to do this, so I'm way
down the road from most investors in thinking about things.
But as I use it, I'm not getting the results
that I need, And by that I mean it's not
right enough. The error rate that I get back from
(11:26):
the questions that I'm asking are anywhere between five and
maybe thirty percent, which that's huge. Like I can't just
use AI. I have to babysit AI. So these are
problems that aren't solved by anybody at this point. And
then there is how are we actually going to use it?
I think there was I think a Bloomberg report that
(11:48):
was saying that probably two hundred thousand people could be
laid off those lower level entry level people into finance
and then that mid tier that are now you know,
kind of like the AI for investment banking and equity research.
But if we were that wrong, we wouldn't last right, Like,
(12:10):
if you're five percent wrong in investment banking, your career
is really short. So you know, these are problems that
aren't being solved. So that's another issue for investors. How
right How long is it going to be till these
get right enough to replace a person?
Speaker 4 (12:26):
That's such a great point because I was also reading
an article that talked about the limitations so far that
have been seen about Gee, what is it deep Seek?
I want to say, geek see deep Seek was its
lack of inability to discuss jijenping or to give real
handswers on Tanam and square, just those topics that are
(12:47):
near and dear to China's heart. And that's in part
some of the issue that you're talking about. So who
wins in this right now?
Speaker 5 (12:56):
Who wins? Well, I think investors are losing in the
short term. But if you like me think that AI
is a path to productivity, and productivity always wins. I
think you could be a winner. I think maybe companies
like open Ai and Microsoft, you know, by dint of
(13:19):
you know, investing in them, and then everybody else who's
developing AI, they might be winners. Maybe the what is
it Stargate? I don't know why. I can never remember.
That's my break up here. I can ever remember that
because it has nothing to do with stars or gates,
that Stargate might be the loser in those people because
we're not necessarily going to need to cover the earth
(13:42):
in data centers.
Speaker 4 (13:43):
Right, yep, all.
Speaker 2 (13:45):
Right, Kim, thanks so much for joining us. Always appreciate
getting your perspective of your technology background. Kim Farst, founder
and chief investment officer of Book Capital Partners, joining us
from Pittsburgh via the zoom Thing.
Speaker 1 (14:00):
Listening to the Bloomberg Intelligence Podcast. Catch us live weekdays
at ten am Eastern on Applecarplay and Android Auto with
the Bloomberg Business App. Listen on demand wherever you get
your podcasts, or watch us live on YouTube.
Speaker 4 (14:14):
Happy mindy, everybody, Alex Deal here alongside Paul Sweeny. This
is Bloomberg Intelligence Radio. We are broadcasting to live for
Interactive Brookers Studio right here in Midtown Manhattan. Also check
us out on YouTube dot com. Or clearly have the
tech sel off underway really concentrated in certain names the
mag seven obviously, in Vidia, the chip stocks, although Apple's up,
I should say, but in Vidia obviously getting pounded, the
(14:35):
chip stocks getting pounded as well. So we wanted to
continue the conversation about AI and you really need to
rethink this investment thesis. Joining us now is Josh Pantoni.
He's CEO boosted dot Ai Now. The company has helped
dozens of investment managers who's AUM totals over one trillion
dollars and how to implement machine learning in their portfolios.
(14:56):
First off, Josh, have you seen this new geek seek? Geeks?
Why do I keep saying geek seek? It's not geeks
seek deep see see if you try deep seak? Have
you talked to your clients about it? What do you think?
Speaker 6 (15:13):
Yeah? So, I mean it came out I think last Monday.
We've already ran a whole suite of benchmarks against it.
Try that out over a bunch of things, and I'd
say we're pretty familiar with it. I think it's impressive
in a number of ways. It's impressive how how much
power you get for the cost. I think the flip
side is it's actually not quite something that we could
even consider implementing it to our process. And I think
(15:36):
there's a lot of clients that will reach the same
conclusion after they start working with it. Why is that.
Speaker 5 (15:42):
So?
Speaker 6 (15:42):
A lot of how we help our clients is with
things like trying to identify risk and trying to identify
to build out these different workflows for different parts of
the financial process. And a really big part of that
is related to trust. You need to be able to
give answers that you're very confident in. You need to
give answers that you can sort of deeply into it
where it's coming from. And I think one of the
(16:03):
challenge of this model is that it has some very
clear political biases right off the bat. So if I'm
trying to use it to say understand what my portfolio
exposure is to China's going to be a challenge to
use it for things like that. Also, a lot of
the things we're most excited about are things like being
able to do computer use, being able to sort of
teach the machine how to like interact with different apps
(16:24):
and things on the computer, image recognition, trying to extract
out charts and things like that, and a lot of
those kind of capabilities that actually doesn't really have it
and seem to be as sophisticated in so it can
do very well in certain benchmarks, but when it actually
comes to trying to apply it for at least most
of the use cases that we're doing, it's not quite
there yet.
Speaker 4 (16:42):
What is there for you right now? Like what does work?
Speaker 6 (16:46):
I think the most interesting thing for me is actually
on the reasoning side. So of course, like one of
the really big pushes Opening Eyes had right now is
like with one and three, you've got these models coming
in where in theory they can take a task, break
it into subcomponents, and then start executing those components. And
I think in that area it's actually quite good. I
also think, you know, just from a cost basis, it's
(17:08):
extremely impressive they've been able to do. There's been some
controversy about exactly how much it costs to actually train
the model. What I can confirm though, is the actual
inference cost. The cost of running the model is extremely
cheap compared to what you're getting with like oh one
and oh three, and so just the fact is even
possible to do that as a major technology breakthrough.
Speaker 2 (17:26):
If nothing else does Deep Seek just highlight the cost
issue and potentially the I don't know, the movement down
in costs of implementing AI.
Speaker 6 (17:40):
Yeah, the way I tend to think about it is
the cheaper it is to train AI systems, the more
advanced capabilities you can train.
Speaker 4 (17:49):
In the stress you lost your audio Josh, you there, Yes,
I am.
Speaker 5 (17:52):
Here, still hear me. We're good?
Speaker 6 (17:54):
Okay, perfect? Yeah. What the way I like to think
about it is the cheaper it is to train these
AI systems, Then the more advanced capabilities you can train,
the more advanced capabilities you can train, the more use
cases you unlock. The more use cases unlock, the more
you actually see it accelerate. So I actually think this
model will probably cause an acceleration in the capabilities of
(18:16):
other models that are getting built as more folks start
to adopt it. So for me, I actually see it
as completely positive.
Speaker 4 (18:22):
What outside of deep Seek and other models, like, have
you worked with that you did?
Speaker 6 (18:27):
Like yeah, so, I mean behind the scenes, we work
with in thropic models we work with open AI models.
We work with a bunch of fine tune models that
we build in house ourselves. You know. I like to
sort of describe it as an orchestra of different types
of models. And one of the things we've sort of
noticed is there's a lot of differences in capabilities. So
(18:48):
something like the inthroduct models and thropic models tend to
be better at very long context window where you try
and handle like huge amounts of text, whereas something like
the GPD form models tend to be a little bit
better at like foreign language and just sort of the
general verbiage it uses as it's giving outputs. So we
use a whole bunch.
Speaker 2 (19:08):
Where how are your clients and the asset management business,
Josh using AI these days, in these early days.
Speaker 6 (19:15):
Yeah, So the way I like to think about it
is we give them the ability to create this sort
of team of little AI workers where you teach them
how to do some kind of task, and then they're
going to do that task on a continuous basis. So
let's say you to do something like write an investment
thesis on a company, or let's say you had to
do something like write an ESG report or Let's say
you wanted to continuously monitor the world for any kinds
(19:38):
of updates to something that might happen in the air space.
These are all examples of sort of workflows you can
teach the system and have the system start to automate.
Speaker 4 (19:48):
Well, we really appreciate your time. It was so good
to get that perspective. It's good to kind of get
the user mindset in for this. Josh Josh Pantoni, CEO
a boosted dot AI on deep Seek and sort of
the pros and the hans there.
Speaker 1 (20:01):
This is the Bloomberg Intelligence podcast, available on Apple, Spotify,
and anywhere else you get your podcasts. Listen live each
weekday ten am to noon Eastern on Bloomberg dot com,
the iHeartRadio app, tune In, and the Bloomberg Business app.
You can also watch us live every weekday on YouTube
and always on the Bloomberg terminal.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Amy Robach & T.J. Holmes present: Aubrey O’Day, Covering the Diddy Trial
Introducing… Aubrey O’Day Diddy’s former protege, television personality, platinum selling music artist, Danity Kane alum Aubrey O’Day joins veteran journalists Amy Robach and TJ Holmes to provide a unique perspective on the trial that has captivated the attention of the nation. Join them throughout the trial as they discuss, debate, and dissect every detail, every aspect of the proceedings. Aubrey will offer her opinions and expertise, as only she is qualified to do given her first-hand knowledge. From her days on Making the Band, as she emerged as the breakout star, the truth of the situation would be the opposite of the glitz and glamour. Listen throughout every minute of the trial, for this exclusive coverage. Amy Robach and TJ Holmes present Aubrey O’Day, Covering the Diddy Trial, an iHeartRadio podcast.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.