WEBVTT - Global Tech Selloff, DeepSeek Buzz 

0:00:02.720 --> 0:00:10.560
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news. You're listening to the

0:00:10.560 --> 0:00:14.480
<v Speaker 1>Bloomberg Intelligence Podcast. Catch us live weekdays at ten am

0:00:14.560 --> 0:00:18.479
<v Speaker 1>Eastern on Applecarplay and Android Auto with the Bloomberg Business App,

0:00:18.600 --> 0:00:21.799
<v Speaker 1>Listen on demand wherever you get your podcasts, or watch

0:00:21.920 --> 0:00:23.400
<v Speaker 1>us live on YouTube.

0:00:23.560 --> 0:00:26.880
<v Speaker 2>Let's take it one step further. Gene Munster, managing partner,

0:00:26.920 --> 0:00:29.360
<v Speaker 2>co founder for Loop Ventures, really one of the leading

0:00:29.880 --> 0:00:33.000
<v Speaker 2>voices on all things technology really for the last twenty

0:00:33.040 --> 0:00:37.040
<v Speaker 2>five thirty years. We appreciate getting a few minutes of

0:00:37.120 --> 0:00:39.640
<v Speaker 2>his time. Gene, I'm just gonna speak for myself, but

0:00:39.680 --> 0:00:43.120
<v Speaker 2>I probably speak for most of my radio and YouTube audience.

0:00:43.320 --> 0:00:45.520
<v Speaker 2>I did not know what deep Sick was, deep Sik

0:00:45.680 --> 0:00:50.200
<v Speaker 2>was before this morning, but now I'm learning. What should

0:00:50.200 --> 0:00:52.120
<v Speaker 2>I be learning? What should I takeaway be today?

0:00:54.400 --> 0:00:58.840
<v Speaker 3>I think that the highest level this whole chapter is about.

0:00:59.000 --> 0:01:02.520
<v Speaker 3>The first takeaway is that investors are nervous that this

0:01:03.120 --> 0:01:05.800
<v Speaker 3>eighty five percent moving the NASDAC over the past couple

0:01:05.840 --> 0:01:08.480
<v Speaker 3>of years is creating an environment where any sort of

0:01:09.319 --> 0:01:11.840
<v Speaker 3>ripple in what that AI trade looks like is going

0:01:11.880 --> 0:01:14.399
<v Speaker 3>to cause a significant impact on some of these valuations.

0:01:14.400 --> 0:01:16.640
<v Speaker 3>So I think that's I think this is a temperature

0:01:16.640 --> 0:01:18.520
<v Speaker 3>in terms of where the market is at as far

0:01:18.520 --> 0:01:21.399
<v Speaker 3>as kind of the company itself is concerned deep seek.

0:01:21.760 --> 0:01:26.920
<v Speaker 3>What's most important here is that they are advancing or

0:01:27.120 --> 0:01:30.399
<v Speaker 3>presumably advancing the cost of training these models you've talked

0:01:30.400 --> 0:01:32.800
<v Speaker 3>a lot about on the show today. And why that's

0:01:32.840 --> 0:01:35.920
<v Speaker 3>important is there's a question about how much hardware is

0:01:35.959 --> 0:01:38.920
<v Speaker 3>needed for that, and that obviously impacts the big hardware companies,

0:01:38.920 --> 0:01:42.160
<v Speaker 3>and then separately, how does that accelerate the adoption? So

0:01:42.800 --> 0:01:46.479
<v Speaker 3>there is the truth is always several layers below the surface,

0:01:46.520 --> 0:01:48.840
<v Speaker 3>and I think when it comes to Dee Sep there

0:01:48.880 --> 0:01:52.400
<v Speaker 3>are still some pretty significant unanswered questions, and I'll start

0:01:52.400 --> 0:01:54.360
<v Speaker 3>with one of probably the biggest one is that this

0:01:54.880 --> 0:01:57.520
<v Speaker 3>five to six million dollar training number that has the

0:01:57.560 --> 0:02:02.600
<v Speaker 3>market upside down today, that was not their most recent model.

0:02:02.600 --> 0:02:05.800
<v Speaker 3>They have an R model. It's unclear, you know, maybe

0:02:05.800 --> 0:02:08.440
<v Speaker 3>that was twenty five fifty one hundred million dollars to

0:02:08.480 --> 0:02:12.600
<v Speaker 3>train it. It probably was below what the most recent

0:02:12.680 --> 0:02:16.200
<v Speaker 3>version of Opening Eyes model, which is anticipated or estimated

0:02:16.200 --> 0:02:19.120
<v Speaker 3>to be about a five hundred million dollars in terms

0:02:19.120 --> 0:02:22.280
<v Speaker 3>of training. So I think big picture here is that

0:02:23.280 --> 0:02:26.760
<v Speaker 3>how we are delivering AI is evolving and there may

0:02:26.760 --> 0:02:30.639
<v Speaker 3>be some pretty significant improvements in terms of the ability

0:02:30.680 --> 0:02:32.919
<v Speaker 3>to do that at a lower cost.

0:02:33.400 --> 0:02:36.200
<v Speaker 4>So to that pointing, how much if we look at

0:02:36.280 --> 0:02:38.720
<v Speaker 4>Envidia in particular, it's down about two hundred and eighty

0:02:38.720 --> 0:02:41.920
<v Speaker 4>three points. What part of that is justified?

0:02:44.080 --> 0:02:46.240
<v Speaker 3>So I think the big picture here, I think that

0:02:46.320 --> 0:02:49.440
<v Speaker 3>this is an overreaction. I think when you look at

0:02:49.440 --> 0:02:52.200
<v Speaker 3>a company like Nvidia, it's a company we do own,

0:02:52.320 --> 0:02:55.080
<v Speaker 3>and I think that this is an overreaction, and just

0:02:55.240 --> 0:02:58.120
<v Speaker 3>specifically around that, if we can kind of zero in

0:02:58.200 --> 0:03:02.120
<v Speaker 3>on this concept that these models are becoming more efficient,

0:03:02.240 --> 0:03:05.040
<v Speaker 3>let's just take deep seek at face value that they've

0:03:05.040 --> 0:03:09.200
<v Speaker 3>had some sort of improvement. Is that that improvement if

0:03:09.240 --> 0:03:13.360
<v Speaker 3>you believe that the US tech companies are competent, that

0:03:13.400 --> 0:03:16.720
<v Speaker 3>improvement is around some architectures that have been talked about

0:03:16.760 --> 0:03:20.720
<v Speaker 3>for the last couple months. So those companies, the big

0:03:21.160 --> 0:03:25.520
<v Speaker 3>tech companies, the companies that are behind all the announcements

0:03:25.600 --> 0:03:31.239
<v Speaker 3>last week Meta increasing their CAPEX spend, and I think

0:03:31.280 --> 0:03:35.160
<v Speaker 3>that that is all that is a belief that they

0:03:35.240 --> 0:03:37.560
<v Speaker 3>still need this hardware. So when it comes back to

0:03:37.840 --> 0:03:41.000
<v Speaker 3>AI and comes back to the Nvidia trade, is that

0:03:41.080 --> 0:03:45.200
<v Speaker 3>I believe if they are reducing the costs to train

0:03:45.280 --> 0:03:49.000
<v Speaker 3>these models that I actually can build. I think a

0:03:49.040 --> 0:03:52.880
<v Speaker 3>credible case that that could increase the demand for some

0:03:52.960 --> 0:03:57.000
<v Speaker 3>of this hardware if we are getting closer to general intelligence.

0:03:57.000 --> 0:03:59.920
<v Speaker 3>If that potential is even closer, it's just going to

0:03:59.960 --> 0:04:02.840
<v Speaker 3>be an insatiable arms race. It's going to continue. And

0:04:02.880 --> 0:04:05.440
<v Speaker 3>so I understand why and video is down today. I

0:04:05.480 --> 0:04:07.800
<v Speaker 3>think it's actually a healthy thing. But ultimately I think

0:04:07.840 --> 0:04:09.480
<v Speaker 3>that their business is going to be just fine. We

0:04:09.520 --> 0:04:11.680
<v Speaker 3>have to wait twenty one trading days before we hear

0:04:11.840 --> 0:04:13.960
<v Speaker 3>from at least in VIDI on that, but I think

0:04:13.960 --> 0:04:14.760
<v Speaker 3>it's going to be fine.

0:04:15.440 --> 0:04:20.240
<v Speaker 2>So, Gene, was it a coincidence that on Friday, Meta

0:04:20.520 --> 0:04:24.640
<v Speaker 2>upsets capex forecast dramatically, which by a magnitude I've never seen,

0:04:24.680 --> 0:04:28.520
<v Speaker 2>from fifty billion to sixty five billion, and then today

0:04:28.560 --> 0:04:31.440
<v Speaker 2>we get this news on deep sick? Was Deep seek?

0:04:31.520 --> 0:04:34.400
<v Speaker 2>Was that kind of I don't know, coincidence?

0:04:35.800 --> 0:04:39.359
<v Speaker 3>I think, what's I think that's just purely coincidence. But

0:04:39.560 --> 0:04:41.840
<v Speaker 3>I do kind of go back to what we're just

0:04:41.880 --> 0:04:43.799
<v Speaker 3>talking about a minute ago. Is I think it's important

0:04:43.800 --> 0:04:47.359
<v Speaker 3>to note is that that announcement from Meta, if you

0:04:47.400 --> 0:04:50.000
<v Speaker 3>assume a Meta is Meta is competent, and I believe

0:04:50.000 --> 0:04:53.640
<v Speaker 3>that they are. That announcement came with the full knowledge

0:04:53.640 --> 0:04:56.159
<v Speaker 3>of what deep Seek was doing, so it's new to

0:04:56.240 --> 0:04:59.080
<v Speaker 3>most of us today. But for these companies who have

0:04:59.200 --> 0:05:02.719
<v Speaker 3>been making these huge investments, the concept of what deep

0:05:02.760 --> 0:05:05.200
<v Speaker 3>Seak had been trying to build has been They've had

0:05:05.240 --> 0:05:07.839
<v Speaker 3>a model that's been out. This has been something that's

0:05:07.880 --> 0:05:10.040
<v Speaker 3>been aware for the last couple of months, and so

0:05:10.080 --> 0:05:13.159
<v Speaker 3>I think that is important. Is that I think again

0:05:13.200 --> 0:05:16.600
<v Speaker 3>that this commentary from on Friday from Zuckerberg about the

0:05:16.640 --> 0:05:23.120
<v Speaker 3>increased capex factors in some of these potential breakthroughs that

0:05:23.160 --> 0:05:25.119
<v Speaker 3>we've seen with deep Seek. I want to stop short

0:05:25.120 --> 0:05:27.960
<v Speaker 3>of saying it's definitively a breakthrough because a lot of

0:05:28.040 --> 0:05:30.240
<v Speaker 3>unknowns around how good the models are and what the

0:05:30.279 --> 0:05:34.800
<v Speaker 3>true costs are. But I think that the AI investment

0:05:34.839 --> 0:05:37.840
<v Speaker 3>phase is still alive and well, this is still very early.

0:05:38.920 --> 0:05:41.880
<v Speaker 4>So if this could be an overreaction, we don't know

0:05:42.160 --> 0:05:45.760
<v Speaker 4>fair enough. In Nvidia, obviously an overreaction that stocked. The

0:05:45.800 --> 0:05:48.520
<v Speaker 4>way you look at it, does it make you think

0:05:48.640 --> 0:05:56.560
<v Speaker 4>though a rating of forty one times estimated PE.

0:05:55.839 --> 0:05:58.760
<v Speaker 3>So from where my head's at and thinking about calendar

0:05:58.800 --> 0:06:01.400
<v Speaker 3>twenty six at this point, and based on the calendar

0:06:01.400 --> 0:06:03.359
<v Speaker 3>twenty six numbers, I think it's closer to like a

0:06:03.400 --> 0:06:06.320
<v Speaker 3>twenty five, and I think that they can grow earnings

0:06:06.360 --> 0:06:08.839
<v Speaker 3>more than that. This is where the basically the rubber

0:06:08.880 --> 0:06:12.119
<v Speaker 3>hits the road, is that ultimately what happens in calendar

0:06:12.200 --> 0:06:14.039
<v Speaker 3>twenty six, all this stuff we're talking about with deep

0:06:14.080 --> 0:06:17.320
<v Speaker 3>seek today really impacts Ynvidia in twenty six. And if

0:06:17.360 --> 0:06:20.719
<v Speaker 3>in fact this dramatically changes how the kind of hardware

0:06:20.720 --> 0:06:22.880
<v Speaker 3>that people need less hardware, then I'm going to be wrong.

0:06:23.000 --> 0:06:26.479
<v Speaker 3>But if this does create this accelerated arms race, then

0:06:26.520 --> 0:06:28.120
<v Speaker 3>I think that they're going to grow faster. And so

0:06:28.520 --> 0:06:31.279
<v Speaker 3>I consider a one peg on that twenty six earnings

0:06:31.360 --> 0:06:36.119
<v Speaker 3>is actually attractive valuation relative to the rest of the group.

0:06:36.160 --> 0:06:38.000
<v Speaker 3>So I'm comfortable with the evaluation.

0:06:38.800 --> 0:06:41.280
<v Speaker 2>Just a red headline crossing the Bloomberg criminals we speak,

0:06:41.480 --> 0:06:46.200
<v Speaker 2>deep Seek says it is subject to large scale malicious attack.

0:06:46.360 --> 0:06:48.480
<v Speaker 2>So that's a headline. We'll have more reporting on that

0:06:48.640 --> 0:06:51.520
<v Speaker 2>going forward. Gene, just real quickly, we're going to hear

0:06:51.520 --> 0:06:54.760
<v Speaker 2>from some of the other big tech companies, Microsoft, Meta, Tesla,

0:06:54.760 --> 0:06:56.440
<v Speaker 2>how do you think they're going to address this issue?

0:06:57.360 --> 0:06:59.719
<v Speaker 3>It's obviously this is going to be front and center

0:06:59.760 --> 0:07:02.600
<v Speaker 3>the topic, and I think they're probably going to address

0:07:02.600 --> 0:07:06.200
<v Speaker 3>it by saying they still have plans to invest meaningfully

0:07:06.200 --> 0:07:08.440
<v Speaker 3>more in counter twenty five over twenty four when it

0:07:08.480 --> 0:07:10.920
<v Speaker 3>comes to Capex, and I think that that will play

0:07:11.000 --> 0:07:14.000
<v Speaker 3>part of a for reassuring we're long ways away from that.

0:07:14.040 --> 0:07:15.920
<v Speaker 3>We got two and a half trading days awave before

0:07:15.920 --> 0:07:17.760
<v Speaker 3>we start to get some of that commentary before the

0:07:18.440 --> 0:07:20.160
<v Speaker 3>mic gets turned over to the other side of the

0:07:20.400 --> 0:07:21.080
<v Speaker 3>equation here.

0:07:21.520 --> 0:07:23.360
<v Speaker 2>All right, Gene, thank you so much for joining us.

0:07:23.400 --> 0:07:25.640
<v Speaker 2>I know you're super busy today. Appreciate getting a few

0:07:25.640 --> 0:07:28.560
<v Speaker 2>minutes of your time. Gen Munster, managing partner co founder

0:07:28.880 --> 0:07:31.520
<v Speaker 2>Lupet Ventures, taking a little bit of a I guess

0:07:31.560 --> 0:07:33.720
<v Speaker 2>a longer term view, in a more broader view of

0:07:33.720 --> 0:07:37.840
<v Speaker 2>what this can mean for the indusuay.

0:07:36.680 --> 0:07:40.560
<v Speaker 1>You're listening to the Bloomberg Intelligence Podcast. Catch the program

0:07:40.680 --> 0:07:44.080
<v Speaker 1>live weekdays at ten am Eastern on Applecarplay and Android

0:07:44.080 --> 0:07:47.080
<v Speaker 1>Auto with the Bloomberg Business app. You can also listen

0:07:47.200 --> 0:07:50.440
<v Speaker 1>live on Amazon Alexa from our flagship New York station,

0:07:51.000 --> 0:07:53.680
<v Speaker 1>Just say Alexa play Bloomberg eleven thirty.

0:07:54.240 --> 0:07:56.120
<v Speaker 4>All right, let's get more on this text sell off

0:07:56.120 --> 0:07:59.040
<v Speaker 4>here Kim Forrest, founder and CIO of Boca Capital Partners.

0:07:59.080 --> 0:08:02.560
<v Speaker 4>She's also a former Chips person, like she did the stuff,

0:08:02.600 --> 0:08:03.720
<v Speaker 4>so she knows the things.

0:08:04.120 --> 0:08:04.480
<v Speaker 5>Clearly.

0:08:04.520 --> 0:08:08.240
<v Speaker 4>The deep Seek news is upending most of the tech industry, right.

0:08:08.600 --> 0:08:10.600
<v Speaker 4>You get the socks and DEXes down hard, you get

0:08:10.600 --> 0:08:13.720
<v Speaker 4>all the hyperscalers are down hard, all the chip makers,

0:08:13.760 --> 0:08:17.840
<v Speaker 4>all the power makers all down hard. Is this the

0:08:17.880 --> 0:08:20.200
<v Speaker 4>reckoning or is this a by the dip moment?

0:08:21.880 --> 0:08:26.000
<v Speaker 5>Well, that is the question, isn't it. I would say

0:08:26.000 --> 0:08:28.160
<v Speaker 5>it's a little bit of both. I think this is

0:08:28.200 --> 0:08:33.800
<v Speaker 5>a good level set, re leveling, resetting our expectations. So

0:08:34.240 --> 0:08:36.320
<v Speaker 5>the first thing is, we have to really see if

0:08:36.360 --> 0:08:41.720
<v Speaker 5>this is true. Now, apparently the company is in deep

0:08:41.800 --> 0:08:46.560
<v Speaker 5>Seek is an open source company. They have released how

0:08:46.600 --> 0:08:50.439
<v Speaker 5>they've done this, but you know, replication is key, so

0:08:50.520 --> 0:08:53.360
<v Speaker 5>we're gonna have to see if this actually works. That's

0:08:53.440 --> 0:08:57.760
<v Speaker 5>thing number one, Like, don't get too upset about major

0:08:57.840 --> 0:09:01.680
<v Speaker 5>developments until you know that they're real major developments, right,

0:09:01.760 --> 0:09:05.400
<v Speaker 5>So that's thing one. Thing two is and I think

0:09:05.440 --> 0:09:08.040
<v Speaker 5>this has been troubling me for a very long time,

0:09:09.120 --> 0:09:15.439
<v Speaker 5>especially with like last week's Stargate announcement. It seems as

0:09:15.480 --> 0:09:21.040
<v Speaker 5>if people have been expecting AI, which we don't know

0:09:21.040 --> 0:09:23.719
<v Speaker 5>what the payoff is yet we have a pretty good

0:09:23.760 --> 0:09:26.120
<v Speaker 5>idea it might help us out. But we're going to

0:09:26.160 --> 0:09:29.760
<v Speaker 5>cover the earth essentially in data centers and use far

0:09:29.840 --> 0:09:32.800
<v Speaker 5>more power than we create right now. So those two

0:09:32.840 --> 0:09:35.440
<v Speaker 5>things are kind of troubling that people have been making

0:09:35.480 --> 0:09:39.199
<v Speaker 5>investments in companies going to produce far more than they

0:09:39.240 --> 0:09:42.400
<v Speaker 5>can produce now. And do you see what I'm saying here?

0:09:42.440 --> 0:09:47.080
<v Speaker 5>We had these really big expectations, and I didn't think

0:09:47.080 --> 0:09:49.160
<v Speaker 5>that they were going to come to happen in short

0:09:49.240 --> 0:09:52.520
<v Speaker 5>or long term because of the well physical limitations of it.

0:09:53.160 --> 0:09:55.200
<v Speaker 2>So Kim, at this very early stage, I think I'm

0:09:55.200 --> 0:09:58.120
<v Speaker 2>probably representative of most of our listeners and viewers had

0:09:58.160 --> 0:10:00.679
<v Speaker 2>not heard of deep Seek before this morning. Sure, and

0:10:00.720 --> 0:10:03.120
<v Speaker 2>so that's calling into question kind of what we think

0:10:03.160 --> 0:10:06.560
<v Speaker 2>we understood about AI and its implication. It's not just

0:10:06.600 --> 0:10:09.920
<v Speaker 2>for technology preference stock market in general. My only question

0:10:09.960 --> 0:10:11.720
<v Speaker 2>that I think I have now that has any relevance

0:10:11.800 --> 0:10:16.160
<v Speaker 2>is do I have to rethink my spending associated with AI.

0:10:18.280 --> 0:10:21.280
<v Speaker 5>It depends on how long your timeframe is. If it's

0:10:21.320 --> 0:10:24.760
<v Speaker 5>the very short term, probably not you you know you'll

0:10:24.760 --> 0:10:27.920
<v Speaker 5>get rewarded because well orders are in and all that

0:10:28.000 --> 0:10:30.320
<v Speaker 5>kind of good stuff. And by short term I'm talking

0:10:30.320 --> 0:10:33.080
<v Speaker 5>a year, but let's say three to five years maybe,

0:10:33.920 --> 0:10:36.720
<v Speaker 5>And why is that? Well, if we can really train

0:10:36.880 --> 0:10:42.600
<v Speaker 5>models more rapidly with less input, which is a good thing, right,

0:10:42.679 --> 0:10:45.040
<v Speaker 5>where were we going to get all this electricity? That

0:10:45.240 --> 0:10:49.680
<v Speaker 5>was like my biggest question. But anyhow, back to your question,

0:10:50.240 --> 0:10:56.760
<v Speaker 5>I think you know this is part of being an

0:10:56.800 --> 0:11:00.960
<v Speaker 5>investors understanding how long was I play? I'm keeping this,

0:11:01.559 --> 0:11:05.080
<v Speaker 5>So that's part of the problem. We have two more

0:11:05.120 --> 0:11:09.200
<v Speaker 5>problems that are not being clarified by deep Seek. One is,

0:11:09.720 --> 0:11:12.080
<v Speaker 5>and remember I used to do this, so I'm way

0:11:12.080 --> 0:11:16.400
<v Speaker 5>down the road from most investors in thinking about things.

0:11:16.679 --> 0:11:20.200
<v Speaker 5>But as I use it, I'm not getting the results

0:11:20.240 --> 0:11:22.960
<v Speaker 5>that I need, And by that I mean it's not

0:11:23.200 --> 0:11:26.600
<v Speaker 5>right enough. The error rate that I get back from

0:11:26.640 --> 0:11:30.080
<v Speaker 5>the questions that I'm asking are anywhere between five and

0:11:30.160 --> 0:11:34.040
<v Speaker 5>maybe thirty percent, which that's huge. Like I can't just

0:11:34.280 --> 0:11:37.920
<v Speaker 5>use AI. I have to babysit AI. So these are

0:11:38.000 --> 0:11:41.520
<v Speaker 5>problems that aren't solved by anybody at this point. And

0:11:41.640 --> 0:11:44.760
<v Speaker 5>then there is how are we actually going to use it?

0:11:44.800 --> 0:11:48.040
<v Speaker 5>I think there was I think a Bloomberg report that

0:11:48.200 --> 0:11:51.319
<v Speaker 5>was saying that probably two hundred thousand people could be

0:11:51.440 --> 0:11:56.480
<v Speaker 5>laid off those lower level entry level people into finance

0:11:56.800 --> 0:12:00.480
<v Speaker 5>and then that mid tier that are now you know,

0:12:00.600 --> 0:12:05.720
<v Speaker 5>kind of like the AI for investment banking and equity research.

0:12:06.640 --> 0:12:09.839
<v Speaker 5>But if we were that wrong, we wouldn't last right, Like,

0:12:10.200 --> 0:12:13.439
<v Speaker 5>if you're five percent wrong in investment banking, your career

0:12:13.520 --> 0:12:17.440
<v Speaker 5>is really short. So you know, these are problems that

0:12:17.559 --> 0:12:22.000
<v Speaker 5>aren't being solved. So that's another issue for investors. How

0:12:22.360 --> 0:12:24.440
<v Speaker 5>right How long is it going to be till these

0:12:24.480 --> 0:12:26.160
<v Speaker 5>get right enough to replace a person?

0:12:26.640 --> 0:12:28.400
<v Speaker 4>That's such a great point because I was also reading

0:12:28.400 --> 0:12:30.640
<v Speaker 4>an article that talked about the limitations so far that

0:12:30.679 --> 0:12:35.520
<v Speaker 4>have been seen about Gee, what is it deep Seek?

0:12:35.520 --> 0:12:39.640
<v Speaker 4>I want to say, geek see deep Seek was its

0:12:40.080 --> 0:12:44.120
<v Speaker 4>lack of inability to discuss jijenping or to give real

0:12:44.200 --> 0:12:47.959
<v Speaker 4>handswers on Tanam and square, just those topics that are

0:12:47.960 --> 0:12:51.200
<v Speaker 4>near and dear to China's heart. And that's in part

0:12:51.240 --> 0:12:53.920
<v Speaker 4>some of the issue that you're talking about. So who

0:12:53.960 --> 0:12:55.480
<v Speaker 4>wins in this right now?

0:12:56.559 --> 0:13:00.760
<v Speaker 5>Who wins? Well, I think investors are losing in the

0:13:00.840 --> 0:13:04.800
<v Speaker 5>short term. But if you like me think that AI

0:13:05.080 --> 0:13:09.800
<v Speaker 5>is a path to productivity, and productivity always wins. I

0:13:09.840 --> 0:13:14.359
<v Speaker 5>think you could be a winner. I think maybe companies

0:13:14.520 --> 0:13:18.880
<v Speaker 5>like open Ai and Microsoft, you know, by dint of

0:13:19.679 --> 0:13:23.200
<v Speaker 5>you know, investing in them, and then everybody else who's

0:13:23.240 --> 0:13:28.000
<v Speaker 5>developing AI, they might be winners. Maybe the what is

0:13:28.040 --> 0:13:30.200
<v Speaker 5>it Stargate? I don't know why. I can never remember.

0:13:30.200 --> 0:13:32.560
<v Speaker 5>That's my break up here. I can ever remember that

0:13:33.000 --> 0:13:35.160
<v Speaker 5>because it has nothing to do with stars or gates,

0:13:35.520 --> 0:13:39.280
<v Speaker 5>that Stargate might be the loser in those people because

0:13:39.280 --> 0:13:42.880
<v Speaker 5>we're not necessarily going to need to cover the earth

0:13:42.880 --> 0:13:43.760
<v Speaker 5>in data centers.

0:13:43.920 --> 0:13:45.240
<v Speaker 4>Right, yep, all.

0:13:45.240 --> 0:13:47.559
<v Speaker 2>Right, Kim, thanks so much for joining us. Always appreciate

0:13:47.559 --> 0:13:51.000
<v Speaker 2>getting your perspective of your technology background. Kim Farst, founder

0:13:51.320 --> 0:13:56.080
<v Speaker 2>and chief investment officer of Book Capital Partners, joining us

0:13:56.160 --> 0:13:57.920
<v Speaker 2>from Pittsburgh via the zoom Thing.

0:14:00.040 --> 0:14:04.000
<v Speaker 1>Listening to the Bloomberg Intelligence Podcast. Catch us live weekdays

0:14:04.000 --> 0:14:07.040
<v Speaker 1>at ten am Eastern on Applecarplay and Android Auto with

0:14:07.120 --> 0:14:10.240
<v Speaker 1>the Bloomberg Business App. Listen on demand wherever you get

0:14:10.280 --> 0:14:13.520
<v Speaker 1>your podcasts, or watch us live on YouTube.

0:14:14.160 --> 0:14:16.720
<v Speaker 4>Happy mindy, everybody, Alex Deal here alongside Paul Sweeny. This

0:14:16.720 --> 0:14:19.400
<v Speaker 4>is Bloomberg Intelligence Radio. We are broadcasting to live for

0:14:19.560 --> 0:14:23.040
<v Speaker 4>Interactive Brookers Studio right here in Midtown Manhattan. Also check

0:14:23.120 --> 0:14:25.920
<v Speaker 4>us out on YouTube dot com. Or clearly have the

0:14:25.920 --> 0:14:29.880
<v Speaker 4>tech sel off underway really concentrated in certain names the

0:14:29.920 --> 0:14:33.040
<v Speaker 4>mag seven obviously, in Vidia, the chip stocks, although Apple's up,

0:14:33.080 --> 0:14:35.200
<v Speaker 4>I should say, but in Vidia obviously getting pounded, the

0:14:35.240 --> 0:14:37.400
<v Speaker 4>chip stocks getting pounded as well. So we wanted to

0:14:37.800 --> 0:14:40.960
<v Speaker 4>continue the conversation about AI and you really need to

0:14:41.000 --> 0:14:45.040
<v Speaker 4>rethink this investment thesis. Joining us now is Josh Pantoni.

0:14:45.120 --> 0:14:48.840
<v Speaker 4>He's CEO boosted dot Ai Now. The company has helped

0:14:48.920 --> 0:14:53.080
<v Speaker 4>dozens of investment managers who's AUM totals over one trillion

0:14:53.120 --> 0:14:56.440
<v Speaker 4>dollars and how to implement machine learning in their portfolios.

0:14:56.880 --> 0:15:01.720
<v Speaker 4>First off, Josh, have you seen this new geek seek? Geeks?

0:15:01.720 --> 0:15:04.000
<v Speaker 4>Why do I keep saying geek seek? It's not geeks

0:15:04.000 --> 0:15:10.680
<v Speaker 4>seek deep see see if you try deep seak? Have

0:15:10.720 --> 0:15:12.720
<v Speaker 4>you talked to your clients about it? What do you think?

0:15:13.720 --> 0:15:16.240
<v Speaker 6>Yeah? So, I mean it came out I think last Monday.

0:15:16.680 --> 0:15:19.200
<v Speaker 6>We've already ran a whole suite of benchmarks against it.

0:15:19.200 --> 0:15:20.480
<v Speaker 6>Try that out over a bunch of things, and I'd

0:15:20.520 --> 0:15:24.120
<v Speaker 6>say we're pretty familiar with it. I think it's impressive

0:15:24.440 --> 0:15:27.560
<v Speaker 6>in a number of ways. It's impressive how how much

0:15:27.600 --> 0:15:29.800
<v Speaker 6>power you get for the cost. I think the flip

0:15:29.840 --> 0:15:33.520
<v Speaker 6>side is it's actually not quite something that we could

0:15:33.560 --> 0:15:36.000
<v Speaker 6>even consider implementing it to our process. And I think

0:15:36.000 --> 0:15:37.360
<v Speaker 6>there's a lot of clients that will reach the same

0:15:37.400 --> 0:15:40.240
<v Speaker 6>conclusion after they start working with it. Why is that.

0:15:42.080 --> 0:15:42.160
<v Speaker 5>So?

0:15:42.480 --> 0:15:44.880
<v Speaker 6>A lot of how we help our clients is with

0:15:44.960 --> 0:15:48.640
<v Speaker 6>things like trying to identify risk and trying to identify

0:15:49.400 --> 0:15:51.440
<v Speaker 6>to build out these different workflows for different parts of

0:15:51.440 --> 0:15:53.840
<v Speaker 6>the financial process. And a really big part of that

0:15:54.200 --> 0:15:56.520
<v Speaker 6>is related to trust. You need to be able to

0:15:56.560 --> 0:15:58.480
<v Speaker 6>give answers that you're very confident in. You need to

0:15:58.480 --> 0:16:01.480
<v Speaker 6>give answers that you can sort of deeply into it

0:16:01.520 --> 0:16:03.200
<v Speaker 6>where it's coming from. And I think one of the

0:16:03.320 --> 0:16:05.200
<v Speaker 6>challenge of this model is that it has some very

0:16:05.280 --> 0:16:08.160
<v Speaker 6>clear political biases right off the bat. So if I'm

0:16:08.200 --> 0:16:10.880
<v Speaker 6>trying to use it to say understand what my portfolio

0:16:10.880 --> 0:16:14.400
<v Speaker 6>exposure is to China's going to be a challenge to

0:16:14.520 --> 0:16:17.400
<v Speaker 6>use it for things like that. Also, a lot of

0:16:17.440 --> 0:16:20.080
<v Speaker 6>the things we're most excited about are things like being

0:16:20.120 --> 0:16:21.960
<v Speaker 6>able to do computer use, being able to sort of

0:16:21.960 --> 0:16:24.480
<v Speaker 6>teach the machine how to like interact with different apps

0:16:24.480 --> 0:16:27.960
<v Speaker 6>and things on the computer, image recognition, trying to extract

0:16:27.960 --> 0:16:30.520
<v Speaker 6>out charts and things like that, and a lot of

0:16:30.560 --> 0:16:32.760
<v Speaker 6>those kind of capabilities that actually doesn't really have it

0:16:32.760 --> 0:16:34.680
<v Speaker 6>and seem to be as sophisticated in so it can

0:16:34.720 --> 0:16:37.920
<v Speaker 6>do very well in certain benchmarks, but when it actually

0:16:37.960 --> 0:16:39.560
<v Speaker 6>comes to trying to apply it for at least most

0:16:39.600 --> 0:16:41.800
<v Speaker 6>of the use cases that we're doing, it's not quite

0:16:41.880 --> 0:16:42.440
<v Speaker 6>there yet.

0:16:42.640 --> 0:16:44.920
<v Speaker 4>What is there for you right now? Like what does work?

0:16:46.440 --> 0:16:48.400
<v Speaker 6>I think the most interesting thing for me is actually

0:16:48.440 --> 0:16:50.440
<v Speaker 6>on the reasoning side. So of course, like one of

0:16:50.440 --> 0:16:52.480
<v Speaker 6>the really big pushes Opening Eyes had right now is

0:16:52.520 --> 0:16:55.600
<v Speaker 6>like with one and three, you've got these models coming

0:16:55.640 --> 0:16:58.080
<v Speaker 6>in where in theory they can take a task, break

0:16:58.120 --> 0:17:01.520
<v Speaker 6>it into subcomponents, and then start executing those components. And

0:17:01.560 --> 0:17:05.239
<v Speaker 6>I think in that area it's actually quite good. I

0:17:05.280 --> 0:17:08.160
<v Speaker 6>also think, you know, just from a cost basis, it's

0:17:08.200 --> 0:17:10.680
<v Speaker 6>extremely impressive they've been able to do. There's been some

0:17:11.160 --> 0:17:14.000
<v Speaker 6>controversy about exactly how much it costs to actually train

0:17:14.040 --> 0:17:16.440
<v Speaker 6>the model. What I can confirm though, is the actual

0:17:16.480 --> 0:17:19.440
<v Speaker 6>inference cost. The cost of running the model is extremely

0:17:19.520 --> 0:17:21.199
<v Speaker 6>cheap compared to what you're getting with like oh one

0:17:21.240 --> 0:17:23.240
<v Speaker 6>and oh three, and so just the fact is even

0:17:23.320 --> 0:17:25.480
<v Speaker 6>possible to do that as a major technology breakthrough.

0:17:26.320 --> 0:17:31.600
<v Speaker 2>If nothing else does Deep Seek just highlight the cost

0:17:31.840 --> 0:17:36.840
<v Speaker 2>issue and potentially the I don't know, the movement down

0:17:37.000 --> 0:17:38.960
<v Speaker 2>in costs of implementing AI.

0:17:40.480 --> 0:17:42.920
<v Speaker 6>Yeah, the way I tend to think about it is

0:17:43.320 --> 0:17:46.760
<v Speaker 6>the cheaper it is to train AI systems, the more

0:17:47.040 --> 0:17:49.320
<v Speaker 6>advanced capabilities you can train.

0:17:49.359 --> 0:17:52.240
<v Speaker 4>In the stress you lost your audio Josh, you there, Yes,

0:17:52.320 --> 0:17:52.680
<v Speaker 4>I am.

0:17:52.560 --> 0:17:54.120
<v Speaker 5>Here, still hear me. We're good?

0:17:54.160 --> 0:17:57.199
<v Speaker 6>Okay, perfect? Yeah. What the way I like to think

0:17:57.200 --> 0:18:00.160
<v Speaker 6>about it is the cheaper it is to train these

0:18:00.400 --> 0:18:04.600
<v Speaker 6>AI systems, Then the more advanced capabilities you can train,

0:18:04.760 --> 0:18:07.360
<v Speaker 6>the more advanced capabilities you can train, the more use

0:18:07.400 --> 0:18:10.280
<v Speaker 6>cases you unlock. The more use cases unlock, the more

0:18:10.320 --> 0:18:12.520
<v Speaker 6>you actually see it accelerate. So I actually think this

0:18:12.560 --> 0:18:16.040
<v Speaker 6>model will probably cause an acceleration in the capabilities of

0:18:16.040 --> 0:18:18.720
<v Speaker 6>other models that are getting built as more folks start

0:18:18.760 --> 0:18:20.199
<v Speaker 6>to adopt it. So for me, I actually see it

0:18:20.200 --> 0:18:21.120
<v Speaker 6>as completely positive.

0:18:22.480 --> 0:18:26.159
<v Speaker 4>What outside of deep Seek and other models, like, have

0:18:26.280 --> 0:18:27.800
<v Speaker 4>you worked with that you did?

0:18:27.920 --> 0:18:31.280
<v Speaker 6>Like yeah, so, I mean behind the scenes, we work

0:18:31.320 --> 0:18:33.640
<v Speaker 6>with in thropic models we work with open AI models.

0:18:34.280 --> 0:18:36.760
<v Speaker 6>We work with a bunch of fine tune models that

0:18:36.800 --> 0:18:40.040
<v Speaker 6>we build in house ourselves. You know. I like to

0:18:40.040 --> 0:18:42.840
<v Speaker 6>sort of describe it as an orchestra of different types

0:18:42.840 --> 0:18:44.840
<v Speaker 6>of models. And one of the things we've sort of

0:18:44.880 --> 0:18:48.520
<v Speaker 6>noticed is there's a lot of differences in capabilities. So

0:18:48.520 --> 0:18:50.840
<v Speaker 6>something like the inthroduct models and thropic models tend to

0:18:50.880 --> 0:18:53.120
<v Speaker 6>be better at very long context window where you try

0:18:53.119 --> 0:18:56.600
<v Speaker 6>and handle like huge amounts of text, whereas something like

0:18:56.640 --> 0:18:58.479
<v Speaker 6>the GPD form models tend to be a little bit

0:18:58.480 --> 0:19:01.800
<v Speaker 6>better at like foreign language and just sort of the

0:19:03.240 --> 0:19:06.240
<v Speaker 6>general verbiage it uses as it's giving outputs. So we

0:19:06.520 --> 0:19:07.280
<v Speaker 6>use a whole bunch.

0:19:08.760 --> 0:19:11.280
<v Speaker 2>Where how are your clients and the asset management business,

0:19:12.240 --> 0:19:14.320
<v Speaker 2>Josh using AI these days, in these early days.

0:19:15.680 --> 0:19:17.320
<v Speaker 6>Yeah, So the way I like to think about it

0:19:17.359 --> 0:19:18.920
<v Speaker 6>is we give them the ability to create this sort

0:19:18.960 --> 0:19:22.399
<v Speaker 6>of team of little AI workers where you teach them

0:19:22.440 --> 0:19:24.200
<v Speaker 6>how to do some kind of task, and then they're

0:19:24.200 --> 0:19:26.439
<v Speaker 6>going to do that task on a continuous basis. So

0:19:26.880 --> 0:19:29.040
<v Speaker 6>let's say you to do something like write an investment

0:19:29.040 --> 0:19:31.080
<v Speaker 6>thesis on a company, or let's say you had to

0:19:31.119 --> 0:19:34.080
<v Speaker 6>do something like write an ESG report or Let's say

0:19:34.119 --> 0:19:38.800
<v Speaker 6>you wanted to continuously monitor the world for any kinds

0:19:38.840 --> 0:19:41.520
<v Speaker 6>of updates to something that might happen in the air space.

0:19:42.040 --> 0:19:45.119
<v Speaker 6>These are all examples of sort of workflows you can

0:19:45.119 --> 0:19:47.000
<v Speaker 6>teach the system and have the system start to automate.

0:19:48.200 --> 0:19:50.160
<v Speaker 4>Well, we really appreciate your time. It was so good

0:19:50.160 --> 0:19:52.080
<v Speaker 4>to get that perspective. It's good to kind of get

0:19:52.080 --> 0:19:55.600
<v Speaker 4>the user mindset in for this. Josh Josh Pantoni, CEO

0:19:55.640 --> 0:19:59.119
<v Speaker 4>a boosted dot AI on deep Seek and sort of

0:19:59.160 --> 0:20:00.560
<v Speaker 4>the pros and the hans there.

0:20:01.480 --> 0:20:06.199
<v Speaker 1>This is the Bloomberg Intelligence podcast, available on Apple, Spotify,

0:20:06.400 --> 0:20:09.840
<v Speaker 1>and anywhere else you get your podcasts. Listen live each

0:20:09.880 --> 0:20:13.640
<v Speaker 1>weekday ten am to noon Eastern on Bloomberg dot com,

0:20:13.760 --> 0:20:17.280
<v Speaker 1>the iHeartRadio app, tune In, and the Bloomberg Business app.

0:20:17.720 --> 0:20:20.800
<v Speaker 1>You can also watch us live every weekday on YouTube

0:20:21.040 --> 0:20:23.280
<v Speaker 1>and always on the Bloomberg terminal.