WEBVTT - Smart Talks with IBM: How Infrastructure is Powering the Age of AI 0:00:04.519 --> 0:00:12.719 Welcome to tech Stuff, a production from iHeartRadio. This season 0:00:12.800 --> 0:00:16.280 on smart Talks with IBM, Malcolm Gladwell and team are 0:00:16.320 --> 0:00:19.560 diving into the transformative world of artificial intelligence with a 0:00:19.600 --> 0:00:24.200 fresh perspective on the concept of open What does open 0:00:24.640 --> 0:00:28.000 really mean in the context of AI. It can mean 0:00:28.080 --> 0:00:31.680 open source code or open data, but it also encompasses 0:00:31.800 --> 0:00:36.120 fostering an ecosystem of ideas, ensuring diverse perspectives are heard, 0:00:36.240 --> 0:00:40.000 and enabling new levels of transparency. Join hosts from your 0:00:40.040 --> 0:00:43.560 favorite Pushkin podcasts as they explore how openness and AI 0:00:43.880 --> 0:00:49.120 is reshaping industries, driving innovation, and redefining what's possible. You'll 0:00:49.159 --> 0:00:52.440 hear from industry experts and leaders about the implications and 0:00:52.479 --> 0:00:56.240 possibilities of open AI, and of course, Malcolm Gladwell will 0:00:56.280 --> 0:00:58.280 be there to guide you through the season with his 0:00:58.440 --> 0:01:01.920 unique insights. Look out for new episodes of Smart Talks 0:01:02.000 --> 0:01:05.440 every other week on the iHeartRadio app, Apple Podcasts, or 0:01:05.440 --> 0:01:09.000 wherever you get your podcasts, and learn more at IBM 0:01:09.160 --> 0:01:12.440 dot com slash smart Talks. 0:01:13.480 --> 0:01:16.880 Hello, Hello, Welcome to Smart Talks with IBM, a podcast 0:01:17.160 --> 0:01:22.800 from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Godwell, this season, 0:01:22.840 --> 0:01:25.800 we're diving back into the world of artificial intelligence, but 0:01:25.880 --> 0:01:31.760 with a focus on the powerful concept of open its possibilities, implications, 0:01:31.800 --> 0:01:36.479 and misconceptions. On today's episode, our season finale, I'm joined 0:01:36.480 --> 0:01:40.640 by Rick Lewis, the senior vice president of Infrastructure at IBM. 0:01:41.360 --> 0:01:44.840 Rick has had a remarkable career focused around product innovation. 0:01:45.560 --> 0:01:48.520 He was actually a few years into retirement when IBM 0:01:48.560 --> 0:01:54.000 came calling with an opportunity he just couldn't turn down. Thankfully, 0:01:54.200 --> 0:01:57.520 Rick came out of retirement and today he oversees a 0:01:57.600 --> 0:02:03.080 vast portfolio from storage and software to global customer support operations, 0:02:03.480 --> 0:02:05.800 and he's engaged in one of the key problems facing 0:02:05.840 --> 0:02:10.720 companies today, an explosion of data. In talking with Rick, 0:02:10.800 --> 0:02:13.160 I can see that this problem of having so much 0:02:13.280 --> 0:02:17.360 data is also an incredible opportunity because if you're able 0:02:17.400 --> 0:02:20.079 to leverage that data to get the most value out 0:02:20.080 --> 0:02:22.480 of it, then you can use it to help bring 0:02:22.520 --> 0:02:26.280 your business into the future. We talked about the serious 0:02:26.320 --> 0:02:29.160 computing power needed to scale AI, as well as the 0:02:29.200 --> 0:02:33.400 ways that infrastructure storage solutions can be essential to enabling 0:02:33.480 --> 0:02:38.000 this new world of possibilities. It's a really great conversation 0:02:38.520 --> 0:02:44.520 so let's get to it. I'm here with Rick Lewis. 0:02:44.880 --> 0:02:45.079 Rick. 0:02:45.320 --> 0:02:49.560 Welcome, Thank Here. We are in the IBM's New York 0:02:49.560 --> 0:02:53.040 City headquarters at one Madison Avenue. I'm going to start 0:02:53.040 --> 0:02:54.959 with you're a hardware guy. 0:02:55.120 --> 0:02:58.760 I'm a hardware guy. I grew up doing hardware chip engineering. 0:02:59.200 --> 0:03:01.680 But like I tell a lot of people, a chip 0:03:01.680 --> 0:03:05.280 engineering project is actually a giant software project with a 0:03:05.400 --> 0:03:08.040 piece of hardware at the end of the project. I 0:03:08.080 --> 0:03:10.840 think if you have that analytical brain, you like to 0:03:10.919 --> 0:03:14.560 solve problems, you'd like to get things working. You can 0:03:14.600 --> 0:03:15.880 do that in soet desertwork. 0:03:15.960 --> 0:03:19.320 But as being someone coming from a hardware background mean 0:03:19.360 --> 0:03:21.959 that you think about problems in a different way. 0:03:22.600 --> 0:03:26.239 I think one thing that you do from a hardware background, 0:03:26.240 --> 0:03:29.359 and especially a chip background, is a chip spin and 0:03:29.480 --> 0:03:33.360 costs millions of dollars. So you're a lot more likely 0:03:33.480 --> 0:03:36.560 to make sure everything has a great chance of being 0:03:36.600 --> 0:03:39.080 perfect from the get go. Or if you start kind 0:03:39.080 --> 0:03:42.120 of from a software background, your general mindset is I 0:03:42.120 --> 0:03:43.760 don't know, try this, see if it works. I don't know. 0:03:43.800 --> 0:03:45.840 Try that is, if it work, and you're kind of iterated, 0:03:45.960 --> 0:03:48.520 or to iterate. Chip people are a little more uptight 0:03:48.600 --> 0:03:52.120 about Okay, if this first round of the chip breaks 0:03:52.360 --> 0:03:55.320 costs us from building another new round of the chip. 0:03:55.200 --> 0:03:58.680 Yeah, so you're a little more You guys are spend more. 0:03:58.520 --> 0:04:04.880 Time planning and planning verifying, tons of time verifying. Yeah. 0:04:05.320 --> 0:04:09.280 So you began your career as you look backward, yes, correct, 0:04:09.320 --> 0:04:10.440 And you were there for how many years? 0:04:10.520 --> 0:04:11.840 I was there for thirty two years? 0:04:12.520 --> 0:04:15.680 Yes, And your last job there was I was leading. 0:04:15.520 --> 0:04:18.200 The software defining cloud business. I had grown up a 0:04:18.240 --> 0:04:23.720 hardware guy. I had done all kinds of hardware projects, 0:04:23.760 --> 0:04:28.320 big complicated Unix servers and things like that, and then came, 0:04:28.640 --> 0:04:30.440 you know, grew out of R and D and more 0:04:30.440 --> 0:04:35.200 into the business realm, and then I'm much an innovator 0:04:35.240 --> 0:04:39.680 at heart. I really like innovating new concepts things like that. 0:04:39.960 --> 0:04:43.160 And what I learned is I enjoyed innovating business models 0:04:43.200 --> 0:04:46.320 and software projects as much as I did hardware products 0:04:47.120 --> 0:04:50.840 and projects, and so getting teams inspired towards doing that 0:04:51.240 --> 0:04:53.599 was really a deep fascination for me. So I ended 0:04:53.680 --> 0:04:57.400 up doing a fantastic variety of experiences and had a 0:04:57.440 --> 0:05:01.640 successful run and honestly retired in tending to retire and 0:05:02.000 --> 0:05:04.360 do some of my outside activities and things like that. 0:05:04.880 --> 0:05:07.280 And then how long did you stay retired before IBM 0:05:07.320 --> 0:05:07.760 can close? 0:05:08.040 --> 0:05:12.480 Almost two years? And when I first got at ALL, 0:05:12.720 --> 0:05:15.720 I thought, no, I'm having too much fun. But I 0:05:15.760 --> 0:05:20.360 would say three things really got me thinking hard about it. 0:05:20.440 --> 0:05:24.120 One the industry that we're in, the IT industry. I 0:05:24.160 --> 0:05:26.200 think it's the golden age. And what I mean by 0:05:26.240 --> 0:05:29.919 that is for twenty years of that career, it is 0:05:30.000 --> 0:05:32.200 kind of in the back office, say make sure that 0:05:32.240 --> 0:05:35.520 stuff doesn't crash, and can you please reduce the cost 0:05:35.560 --> 0:05:37.599 as much as possible, because it's not that important to 0:05:37.640 --> 0:05:41.080 the main business. It's just a back office function. You 0:05:41.120 --> 0:05:43.400 can see it right now. It is at the forefront 0:05:43.440 --> 0:05:47.719 of all business revolution. It happened with the Internet. It 0:05:47.800 --> 0:05:51.480 happened again with cloud and how that changed every ounce 0:05:51.520 --> 0:05:54.560 of business, not just IT business, but all business. And 0:05:54.600 --> 0:05:57.719 I think it's happening again with AI. So to be 0:05:57.839 --> 0:06:00.440 in that career that long and to miss the kind 0:06:00.440 --> 0:06:02.920 of this age where it's like this is front and center. 0:06:03.000 --> 0:06:06.680 This changes everything about all businesses, not just technology businesses. 0:06:07.400 --> 0:06:11.200 I was kind of feeling like, Gosh, you trained to 0:06:11.279 --> 0:06:16.320 be in these really awesome environments, why wouldn't you do 0:06:16.400 --> 0:06:18.640 that for a little while longer while you still can 0:06:18.720 --> 0:06:22.359 do it. That combined with IBM and IBM seeing the 0:06:22.440 --> 0:06:26.640 talent pool, the brilliant people at IBM, I worked with 0:06:26.680 --> 0:06:29.359 a ton of brilliant people before I saw a chance 0:06:29.400 --> 0:06:32.000 to work with even a larger staff of brilliant people. 0:06:32.120 --> 0:06:35.320 And then the assets that IBM had, which is, you know, 0:06:35.320 --> 0:06:38.080 they'd already been doing a lot of experimentation in AI, 0:06:38.200 --> 0:06:43.400 they're working in quantum, the deep, rich heritage of successful projects. 0:06:43.400 --> 0:06:45.440 I thought, who wouldn't want to kind of see if 0:06:45.440 --> 0:06:48.560 they could be part of that next great wave of IBM. 0:06:48.680 --> 0:06:51.200 And so I kind of decided, all right, I'm going 0:06:51.240 --> 0:06:53.279 to put the outside interest on hold for a while 0:06:53.279 --> 0:06:54.320 and get back in the game. 0:06:54.440 --> 0:06:57.760 Along between the phone call, the first phone call and 0:06:57.800 --> 0:06:58.800 you say, yes. 0:06:58.600 --> 0:07:03.520 It was a while, was probably six months. Arvind's our CEO, 0:07:03.720 --> 0:07:05.839 teases me about that a lot. Yeah, he was like, 0:07:05.920 --> 0:07:07.800 I don't think six months. Is that long? It took 0:07:07.800 --> 0:07:11.680 a while you're retirement, I know. Yeah, it's one thing 0:07:11.720 --> 0:07:14.360 to compare. I'm working here and doing this stuff versus 0:07:14.360 --> 0:07:17.720 working there. It's really hard to compare. I'm doing exactly 0:07:17.760 --> 0:07:19.720 as I want to do every single day. When I 0:07:19.760 --> 0:07:21.440 wake up and now I'm not going to get to 0:07:21.440 --> 0:07:24.240 do that again. It took a while for me to 0:07:24.240 --> 0:07:27.040 get over and I thought, I can't miss this wave, 0:07:27.080 --> 0:07:31.360 and I'm really really happy that I did, because we're 0:07:31.400 --> 0:07:34.920 doing some amazing fun things and I'm getting challenged in 0:07:34.960 --> 0:07:36.880 ways that I never did, so it's really fun. 0:07:37.120 --> 0:07:40.160 Talk a little bit about your job here at IBM. 0:07:40.360 --> 0:07:44.480 You oversee a kind of massive portfolio. 0:07:44.640 --> 0:07:47.560 It's a big group, so I run the Infrastructure organization. 0:07:47.600 --> 0:07:50.920 There's three main groups of products at IBM. There's the 0:07:51.000 --> 0:07:54.720 Infrastructure group, which I run, the Software group, and the 0:07:54.760 --> 0:07:59.880 Consulting group. And infrastructure is built up of mainframes, which 0:07:59.880 --> 0:08:04.160 is called our Z portfolio, our servers which is our 0:08:04.200 --> 0:08:08.520 power portfolio storage. By the way, those businesses include the 0:08:08.560 --> 0:08:10.880 supply chain to build all of that stuff, so that's 0:08:10.880 --> 0:08:14.680 in the group. Then I have the worldwide Customer Support Organization. 0:08:14.760 --> 0:08:18.560 It's called TLS Technology life Cycle Services, which is a 0:08:18.560 --> 0:08:21.960 network of about thirteen thousand people around the globe that 0:08:22.040 --> 0:08:24.120 make sure that everything runs and works when you buy 0:08:24.200 --> 0:08:27.800 IBM products. And then also our IBM Cloud, which is 0:08:28.360 --> 0:08:33.080 how we host applications and deliver as a service products 0:08:33.080 --> 0:08:35.839 for our client base, so there's a lot. I think 0:08:35.840 --> 0:08:37.640 it's about forty five thousand people total. 0:08:38.040 --> 0:08:42.600 Do those components of the infrastructure group are they aligned 0:08:42.600 --> 0:08:46.079 in their trajectory or do they are they on different paths? 0:08:46.120 --> 0:08:48.760 And I'm just curious what so navigattle of both. 0:08:48.800 --> 0:08:51.400 It's interesting you would ask that because I think of 0:08:51.679 --> 0:08:55.840 all of the challenges coming to the new company, there 0:08:55.840 --> 0:08:58.560 were things I expected, things that they didn't expect. But 0:08:59.640 --> 0:09:04.000 getting that culture right in that group has been a 0:09:04.040 --> 0:09:10.120 big challenge. IBM has a great culture toward quality products, 0:09:10.200 --> 0:09:14.200 toward emphasizing passion for the client and making sure that 0:09:14.240 --> 0:09:17.840 the client is happy, and for delivering innovation on a 0:09:17.880 --> 0:09:21.239 scale that you know, for more than one hundred years 0:09:21.000 --> 0:09:26.079 has been extremely powerful. But with success comes some challenges. 0:09:26.440 --> 0:09:28.560 And with that success you can tend to get a 0:09:28.600 --> 0:09:32.160 little bit insular, like you don't keep an eye on 0:09:32.200 --> 0:09:34.920 the competition as well, you can get more siloed, where 0:09:35.280 --> 0:09:37.319 you know, this is my business unit, this is my 0:09:37.480 --> 0:09:40.680 business unit, I compete with the other business unit. That's 0:09:40.720 --> 0:09:44.280 not a good thing when when you're a company and 0:09:44.320 --> 0:09:48.000 you can get really risk averse, meaning you feel like, hey, 0:09:48.040 --> 0:09:50.080 this is a successful business I don't want to do 0:09:50.120 --> 0:09:52.200 anything to mess it up, so I don't need to 0:09:52.200 --> 0:09:55.079 try new things. Well, that's exactly the recipe to kind 0:09:55.080 --> 0:09:57.360 of be shrinking, and infrastructure had been shrinking for a 0:09:57.400 --> 0:10:01.120 little while, and so a lot of what the challenge 0:10:01.200 --> 0:10:04.319 was for me was to invigorate that risk taking and 0:10:05.480 --> 0:10:07.840 get to a growth mindset where you're trying new things 0:10:07.880 --> 0:10:10.840 and seeing what works and what doesn't work, and changing 0:10:10.840 --> 0:10:13.160 some of the models, like investing a little bit less 0:10:13.160 --> 0:10:16.640 in hardware for some software differentiation that goes into the hardware. 0:10:16.720 --> 0:10:20.199 So it's been very successful so far, and it's been 0:10:20.200 --> 0:10:22.080 a good journey. It's almost four years now. 0:10:22.440 --> 0:10:25.400 Give me an example of what was a really hard 0:10:25.440 --> 0:10:27.160 problem that you've dealt with in those four years. 0:10:27.160 --> 0:10:30.079 So, boy, a really hard problem? 0:10:30.160 --> 0:10:33.120 An interesting and are you interesting is a better word 0:10:33.160 --> 0:10:33.480 than art. 0:10:33.760 --> 0:10:35.760 One of the first things that I kind of chewed 0:10:35.800 --> 0:10:38.440 on a little bit is I talked about how we 0:10:38.480 --> 0:10:41.600 have Z power and storage. The Z and power product 0:10:41.640 --> 0:10:44.400 lines are well known in the industry. Is is really 0:10:44.440 --> 0:10:47.760 fit for purpose computing that have strengths that you know 0:10:47.880 --> 0:10:51.800 Z runs you know most of the world's economic backbone, 0:10:51.840 --> 0:10:55.440 and if you use a credit card, ninety percent of 0:10:55.480 --> 0:10:58.959 credit card transactions for the globe. Go through these Z 0:10:59.440 --> 0:11:02.000 mainframes there in every bank there. You know, it's a 0:11:02.000 --> 0:11:04.800 big business. It's well known in the industry. Same with power, 0:11:05.240 --> 0:11:10.160 very tuned and optimized for smaller operations than our giant 0:11:10.280 --> 0:11:16.240 Z mainframes, but really mission critical workloads for retail, for insurance, 0:11:16.240 --> 0:11:19.440 for banking, for all of that. Our storage business not 0:11:19.559 --> 0:11:22.360 so well known. In fact, when I came I thought, 0:11:22.600 --> 0:11:25.000 did they have storage? Well, I have storage when I 0:11:25.160 --> 0:11:27.600 come into I. So I got online and I thought, 0:11:27.679 --> 0:11:29.480 it's still hard for me to tell did they have 0:11:29.600 --> 0:11:31.920 storage or not? Now I own a storage business. So 0:11:32.000 --> 0:11:33.560 one of the things was not just to get the 0:11:33.600 --> 0:11:37.200 market perception up, but to invest in that business. Because 0:11:37.240 --> 0:11:39.920 if you look at infrastructure overall around the globe, it's 0:11:39.960 --> 0:11:43.520 growing at five percent a year. The infrastructure business had 0:11:43.559 --> 0:11:46.240 been kind of flat to declining, and so a challenge 0:11:46.280 --> 0:11:48.240 was how do we grab onto the growth. Well, one 0:11:48.240 --> 0:11:50.960 of the biggest growth areas due to the explosion of 0:11:51.040 --> 0:11:53.640 data in the world is storage. So what do you 0:11:53.679 --> 0:11:55.880 do to kind of get on that growth rate. So 0:11:55.880 --> 0:11:59.520 we did a lot of reinvigoration of the innovation in 0:11:59.559 --> 0:12:02.439 that a lot of software value, add a lot of 0:12:02.760 --> 0:12:05.839 doubling down on the things that are working. Portfolio rationalization, 0:12:05.920 --> 0:12:08.160 where you segment the market and you say, okay, we're 0:12:08.280 --> 0:12:10.080 going to do less of this and really go big 0:12:10.120 --> 0:12:13.000 in these areas. And that's been probably the most dramatic 0:12:13.000 --> 0:12:16.360 turnaround inside the group. Is our storage thing. When you 0:12:16.400 --> 0:12:19.120 say it's a hard problem, it's not just oh, you know, 0:12:19.160 --> 0:12:23.720 how do we do the math? No, it's cultural. It's strategy, 0:12:23.800 --> 0:12:25.960 and how do you get the strategy set. It's segmentation, 0:12:26.080 --> 0:12:29.199 it's product strategy at a granular level across a bunch 0:12:29.240 --> 0:12:32.559 of dimensions, and then putting the investment behind it. It's 0:12:32.600 --> 0:12:34.800 a big challenge. It takes a long time, but it's working. 0:12:34.880 --> 0:12:36.839 So we're happy with Yeah. 0:12:36.520 --> 0:12:38.480 Tell me give me a little bit of perspective on 0:12:39.040 --> 0:12:42.840 you've been there four years. Imagine we're having this conversation 0:12:42.920 --> 0:12:46.760 four years ago. Yeah, what sorts of things have happened 0:12:46.800 --> 0:12:50.360 over the last four years that have surprised you that 0:12:50.400 --> 0:12:53.360 you didn't see come as we had exactly the same 0:12:53.360 --> 0:12:54.520 conversation four years ago. 0:12:55.920 --> 0:12:57.640 No, because I didn't know what was in I'll tell 0:12:57.640 --> 0:13:01.480 you some of the biggest surprises I thought from the outside, 0:13:01.920 --> 0:13:05.119 and you know, you hear from a lot of customers, 0:13:05.280 --> 0:13:08.360 especially ten years ago. We're all going to cloud. We're 0:13:08.400 --> 0:13:10.760 all so I thought, well, I wonder if the mainframe 0:13:10.840 --> 0:13:13.800 business is struggling. When I get inside of there, I 0:13:13.840 --> 0:13:16.240 found the opposite to be true. The mainframe business is 0:13:16.280 --> 0:13:20.400 actually flourishing because transaction demand across the globe has done 0:13:20.480 --> 0:13:23.960 nothing but grow. And even more surprising was the level 0:13:23.960 --> 0:13:27.000 of innovation that the team was already doing in mainframes 0:13:27.440 --> 0:13:31.920 before I got here was astounding. For example, we have AI. 0:13:32.800 --> 0:13:37.280 They were building AI technology into the mainframe processors three 0:13:37.360 --> 0:13:40.640 years before chat GPT made everybody talk about it in 0:13:40.679 --> 0:13:44.760 the industry, So that was really pleasantly surprising. So that 0:13:44.920 --> 0:13:49.760 was wonderful. Other surprises. I knew about the kind of 0:13:49.760 --> 0:13:52.880 the IP of IBM and the mystique in that, and 0:13:52.960 --> 0:13:55.440 I used to joke with people, especially on the outside, 0:13:55.440 --> 0:13:56.880 I said, I can't wait to get in there and 0:13:56.880 --> 0:13:59.680 see what's in the big blue toolbox? Right, what are 0:13:59.720 --> 0:14:03.040 all the things they have going on? I way underestimated 0:14:03.240 --> 0:14:05.640 the size of the big blue toolbox and what was 0:14:05.679 --> 0:14:10.280 in there, meaning the amount of really hardcore research that 0:14:10.320 --> 0:14:13.839 we're still doing into how to build chips and how 0:14:13.880 --> 0:14:16.960 to get to things beyond two nanimeter and that kind 0:14:16.960 --> 0:14:22.960 of capability packaging industry leading packaging technologies, and that's in 0:14:23.000 --> 0:14:25.760 my hardware kind of patch quantum. The next thing that 0:14:25.840 --> 0:14:29.720 will come after we're done talking about AI. You know, 0:14:30.440 --> 0:14:33.480 all of those things were surprising, But it wasn't just that. 0:14:33.560 --> 0:14:36.120 It was then the software innovations that are going on 0:14:36.280 --> 0:14:41.080 heavy investment in AI technologies before it was really popular 0:14:41.120 --> 0:14:43.760 to be talking about that. But as I saw that, 0:14:43.840 --> 0:14:46.480 I thought this is going to be really fun. Because 0:14:46.520 --> 0:14:49.000 I had a good feel for where the industry was going. 0:14:49.840 --> 0:14:52.360 I just didn't and I knew, man, I know that 0:14:52.480 --> 0:14:55.120 talent is really good and there's brilliant people there, but 0:14:55.160 --> 0:14:59.080 I didn't know the level of IP frankly that IBM 0:14:59.200 --> 0:15:02.080 had at its disposal, And now you're seeing that in 0:15:02.160 --> 0:15:05.359 things like Watson X and things like AI in mainframes, 0:15:05.400 --> 0:15:05.800 et cetera. 0:15:06.200 --> 0:15:09.160 Building on that, since you brought up AI, can you 0:15:09.200 --> 0:15:13.080 walk me through what has to happen from your perspective, 0:15:13.120 --> 0:15:19.200 from the infrastructure perspective to make the AI explosion work? Yeah, 0:15:19.240 --> 0:15:21.360 so everyone wants to do more of this stuff. Yes, 0:15:21.440 --> 0:15:23.920 clearly there has to be some underpinning of it. 0:15:24.320 --> 0:15:28.440 Yeah, I would tell you, you know, I think that people 0:15:28.600 --> 0:15:30.680 feel like where we're at right now in the AI 0:15:30.760 --> 0:15:33.480 journey had to do with one specific piece of software. 0:15:33.840 --> 0:15:37.960 I think the inflection point for that whole thing really 0:15:38.880 --> 0:15:42.600 at its root was around hardware, meaning the algorithms needed 0:15:42.640 --> 0:15:45.240 to do larger language models. And all of that had 0:15:45.280 --> 0:15:47.920 been around, they'd been talked about in the industry, but 0:15:47.960 --> 0:15:51.000 at some point you hit a tipping point of hardware 0:15:51.040 --> 0:15:53.680 capability where it's like, oh, now we can do this 0:15:53.800 --> 0:15:57.200 in a broof force way, massive amounts of matrix math 0:15:57.280 --> 0:16:00.200 to get weights correct so that you can do you know, 0:16:00.280 --> 0:16:03.600 the right level of predictions that enable large language models. 0:16:03.960 --> 0:16:06.520 And once we got to that horsepower. And that's why 0:16:06.560 --> 0:16:09.320 you hear about giant GPUs that are driving this and 0:16:09.680 --> 0:16:11.920 the sales of those, et cetera. It's because we just 0:16:12.000 --> 0:16:13.920 barely got over the hump where you can do these 0:16:13.960 --> 0:16:19.240 big hard things in terms of hardware capability to do it. 0:16:19.360 --> 0:16:22.200 Give me a layman, give me a sense of when 0:16:22.240 --> 0:16:24.400 you say there was a kind of threshold where suddenly 0:16:24.400 --> 0:16:25.560 these things became possible. 0:16:25.680 --> 0:16:28.480 Yeah, I don't know if there's an exact number, But 0:16:29.640 --> 0:16:31.720 and more basic question that I get from a lot 0:16:31.760 --> 0:16:34.440 of people. You know, my friends and family outside is 0:16:34.760 --> 0:16:39.480 why GPUs. What does a GPU, a graphics processor have 0:16:39.600 --> 0:16:44.000 to do with AI. It's not, Well, graphics processors are 0:16:44.040 --> 0:16:47.760 really good at this thing matrix math, because they're figuring 0:16:47.840 --> 0:16:51.200 out how do I map a pixel? And as I 0:16:51.320 --> 0:16:55.480 move an object across the screen, it's essentially matrix math 0:16:55.560 --> 0:16:58.240 to figure out, Okay, what does what does this pixel 0:16:58.320 --> 0:17:01.040 on a screen look like? And what's it's doing? And 0:17:01.080 --> 0:17:04.560 as you know, we've gotten more high resolution graphics, more 0:17:04.640 --> 0:17:07.680 high resolution monitors, et cetera. It's a lot more pixels 0:17:07.720 --> 0:17:09.640 and a lot more math and a lot more matrix 0:17:09.680 --> 0:17:12.480 math about how you compute that. The first big thing 0:17:12.560 --> 0:17:15.200 that kind of started to look like that, it turns out, 0:17:15.400 --> 0:17:18.880 was crypto and crypto mining, and so you saw graphics 0:17:18.920 --> 0:17:22.119 companies starting to sell to crypto. The technology got to 0:17:22.160 --> 0:17:24.480 a certain point and there were use cases like bitcoin 0:17:24.560 --> 0:17:26.679 and that that kind of said, hey, we need to 0:17:26.720 --> 0:17:29.000 do a lot of this matrix math to be able 0:17:29.000 --> 0:17:32.040 to do that. So graphic chips were a natural fit 0:17:32.119 --> 0:17:34.800 and that kind of sustain But meanwhile, behind the scenes, 0:17:34.800 --> 0:17:39.240 a lot of this AI AI is about numeric calculations 0:17:39.320 --> 0:17:43.000 having to do with weights and matrices that say you know, 0:17:43.440 --> 0:17:46.879 giant consolidated things that predict what's going to kind of 0:17:46.920 --> 0:17:49.320 happen based on what other things have happened, just like 0:17:49.400 --> 0:17:53.920 predicting where pixel goes. But it's really about being able 0:17:54.000 --> 0:17:56.800 to do enough data in jest to be able to 0:17:56.840 --> 0:17:59.919 do and then the calculations to be able to simple 0:18:00.080 --> 0:18:04.240 five things like entire sets of language or giant chunks 0:18:04.280 --> 0:18:06.639 of the Internet, to get enough waitings in there to 0:18:06.640 --> 0:18:09.320 be able to say, Okay, we can predict what you 0:18:09.320 --> 0:18:12.720 would say in this language based on all of the 0:18:12.800 --> 0:18:15.720 volumes of stuff that we've seen that when you start 0:18:15.720 --> 0:18:18.960 talking like this, the next word is likely, oh it's this. Yeah. 0:18:19.000 --> 0:18:21.240 So, but my point is to get to that point. 0:18:21.359 --> 0:18:25.159 That's threshold. We got there because was it because we 0:18:25.240 --> 0:18:28.080 simply threw a lot more resources at the problem or 0:18:28.119 --> 0:18:32.399 is it because the underlying technology got suddenly or gradually 0:18:32.560 --> 0:18:33.720 so much more efficient. 0:18:33.840 --> 0:18:37.080 It's always yes and yes. But you know, the industry 0:18:37.160 --> 0:18:39.320 for a lot of years would talk about Moore's law. 0:18:39.800 --> 0:18:43.560 Well, quick, will you define for us More's law for 0:18:43.600 --> 0:18:45.159 those of those who's forgotten it. 0:18:45.400 --> 0:18:48.560 Yeah, So Gordon Moore at Intel coined this thing. It 0:18:48.600 --> 0:18:53.320 was basically that the horsepower I'm going to translate it 0:18:53.400 --> 0:18:59.040 roughly of technology will double every couple of years. We're 0:18:59.080 --> 0:19:01.760 still on Moore's law. More's law changed a little bit. 0:19:02.160 --> 0:19:05.000 For a while, it was always about frequency. Things would 0:19:05.000 --> 0:19:08.159 go faster, faster, faster. That kind of petered out. But 0:19:08.240 --> 0:19:11.440 what happened is, rather than faster, faster, faster, we did 0:19:11.480 --> 0:19:14.760 more and more and more. So rather than one operating 0:19:14.880 --> 0:19:18.720 unit going a lot faster on its throughput, you put 0:19:18.760 --> 0:19:21.199 ten operating units on a chip, now you put one 0:19:21.280 --> 0:19:24.320 hundred operating units on a chip, now a thousand. Some 0:19:24.400 --> 0:19:29.040 of these problems, the matrix math problems scaled parallel extremely well. 0:19:29.040 --> 0:19:31.320 You don't have to do something really fast, you just 0:19:31.440 --> 0:19:33.440 have to do a lot of the similar things in 0:19:33.480 --> 0:19:36.439 parallel at the same time. So again that kind of 0:19:36.440 --> 0:19:39.159 that extension of Moore's law, more and more hardware on 0:19:39.240 --> 0:19:40.640 a chip to be able to do more and more 0:19:40.640 --> 0:19:43.879 of those calculations in parallel and come up with it. 0:19:44.000 --> 0:19:47.120 And we said, yeah, was that threshold predictable? In other words, 0:19:47.119 --> 0:19:50.000 see people in the industry, like you sit down X 0:19:50.040 --> 0:19:52.080 number of years ago and say, when we get here, 0:19:53.080 --> 0:19:54.840 AI is going to become much more of a. 0:19:55.480 --> 0:20:02.080 It's funny the horsepower that very pretty. The use cases 0:20:02.560 --> 0:20:05.160 not always so easy to kind of figure out. That's 0:20:05.200 --> 0:20:08.840 where the human spirit kind of gets involved. I think 0:20:08.880 --> 0:20:10.800 for some people that say, oh, I saw that coming, 0:20:11.080 --> 0:20:14.199 but people have been predicting kind of the rise of 0:20:14.280 --> 0:20:17.199 AI for twenty five years. Oh well, then when we 0:20:17.280 --> 0:20:19.119 get to this next gener oh when we get here, 0:20:19.320 --> 0:20:22.280 it kind of hadn't happened. There's always a magic point 0:20:23.200 --> 0:20:25.520 where you kind of get to where the technology and 0:20:25.600 --> 0:20:28.000 the use case and somebody does something to kind of 0:20:28.040 --> 0:20:30.760 make it catch on. And I think we're at one 0:20:30.760 --> 0:20:32.679 of those moments in AI for sure right now. And 0:20:32.720 --> 0:20:34.919 I don't think it's you know, people that have said, oh, 0:20:34.920 --> 0:20:38.760 this is just the latest wave of you know, I've 0:20:38.800 --> 0:20:41.280 heard this about a lot of technologies, but AI is 0:20:41.320 --> 0:20:43.960 the technology the future, and it always will be. I 0:20:44.080 --> 0:20:46.760 used to hear that. You're not hearing that now, right, 0:20:46.800 --> 0:20:50.480 It's like, no, it's primetime. It will change everything, just 0:20:50.640 --> 0:20:52.680 like some of these other things changed everything. 0:20:52.920 --> 0:20:57.359 I noticed it if personally when I speak somewhere or 0:20:57.400 --> 0:21:01.000 I'm listening in an audience somewhere. Over the last let's 0:21:01.000 --> 0:21:05.520 say twelve months, there's always a whole bunch of AI 0:21:05.640 --> 0:21:08.520 questions yes, And if I go back two years ago, 0:21:08.560 --> 0:21:09.760 there were no AI questions. 0:21:09.840 --> 0:21:10.040 Yes. 0:21:10.640 --> 0:21:12.960 Now my question is, so there's been this explosion on 0:21:13.000 --> 0:21:17.399 the in popular fascination with what's going on AI. It 0:21:17.480 --> 0:21:20.480 seems like the last year. I agree with you in 0:21:20.560 --> 0:21:26.840 your world, when did the explosion of conversation around this start. 0:21:27.560 --> 0:21:36.880 It's I love this question because IBM had a fairly 0:21:37.040 --> 0:21:42.119 big effort and business called Watson before Watson X. And 0:21:42.160 --> 0:21:45.080 this is going back kind of ten years. I'll give 0:21:45.080 --> 0:21:47.920 you another kind of example. I knew about a lot 0:21:47.920 --> 0:21:51.359 of tablet technology before there was an iPad, a lot. 0:21:51.600 --> 0:21:53.960 For ten years, there were a lot, but it kind 0:21:53.960 --> 0:21:57.640 of takes a magic combination of the technology, the user experienced, 0:21:57.680 --> 0:22:00.520 the software, and the need and the market for it 0:22:00.560 --> 0:22:02.719 to kind of go. Now it's the thing. Now we 0:22:02.760 --> 0:22:05.359