1 00:00:00,080 --> 00:00:03,520 If we can advance quantum a little bit faster, while 2 00:00:03,600 --> 00:00:07,240 quantum comes with a power requirement in the terms 3 00:00:07,240 --> 00:00:10,800 of cooling, the actual cost to run the 4 00:00:10,800 --> 00:00:14,520 QPU is almost zero, right? It 5 00:00:14,520 --> 00:00:17,360 really doesn't cost a whole lot to run a QPU. 6 00:00:18,000 --> 00:00:21,440 AI may be approaching its limits, but quantum 7 00:00:21,440 --> 00:00:23,920 computing could be the next leap forward. 8 00:00:41,597 --> 00:00:45,090 Hello and welcome back to Impact Quantum, the podcast where we explore the emerging 9 00:00:45,090 --> 00:00:48,890 field of quantum computing. And you don't need to be a PhD, you just 10 00:00:48,890 --> 00:00:52,730 need to be a little bit curious. And with me is the most quantum curious 11 00:00:52,730 --> 00:00:56,270 person I know, Candace Cahouli. How's it going, Candace? It's great. 12 00:00:56,350 --> 00:00:59,310 Today's a wonderful day. I'm really, really excited. 13 00:00:59,950 --> 00:01:03,790 We are going to be speaking with Danny Wall, who 14 00:01:03,790 --> 00:01:07,230 is the founder, CEO, and CTO 15 00:01:07,710 --> 00:01:11,510 at OA Quantum Labs. Hi, 16 00:01:11,510 --> 00:01:15,150 Danny. How are you today? I am fantastic. 17 00:01:15,390 --> 00:01:19,070 How about yourself? Doing all right. It's always 18 00:01:19,070 --> 00:01:22,510 good to hear from folks in a state warmer and 19 00:01:22,510 --> 00:01:26,360 sunnier than where I am. We had our first winter storm warning 20 00:01:26,360 --> 00:01:29,480 here for the season here in the Baltimore, D.C. area. 21 00:01:30,040 --> 00:01:33,840 And kids, kids were— had a late start to school and that always 22 00:01:33,840 --> 00:01:37,320 throws things off. But Candace is an old hat at 23 00:01:37,320 --> 00:01:41,000 snow. It's Montreal. They probably already had like 20 feet already for the season. 24 00:01:41,000 --> 00:01:44,660 No, seriously, like it's true because it always starts. It 25 00:01:44,660 --> 00:01:48,200 usually always starts Halloween. Like you get a little bit in Halloween 26 00:01:48,360 --> 00:01:52,080 just to have a taste. So if you're— if your costume does not fit 27 00:01:52,080 --> 00:01:55,780 over your winter coat, It is not an acceptable costume 28 00:01:55,780 --> 00:01:59,620 here in Montreal, Quebec. But yeah, it's snowing every day. 29 00:01:59,700 --> 00:02:03,420 Like, it just snows every day. But that's just how it is. But you 30 00:02:03,420 --> 00:02:07,060 learn how to deal with it. And so it's just fine. Just very pretty. We 31 00:02:07,060 --> 00:02:10,900 actually get our first snow overnight tonight. Oh, 32 00:02:11,380 --> 00:02:15,220 nice. Oh, you must be in the altitude then. About 33 00:02:15,220 --> 00:02:18,820 5,000, a little over 5,000 feet. Yeah. Oh, okay. Okay. 34 00:02:20,790 --> 00:02:24,470 You're coming to us from sunny New Mexico, or normally sunny New Mexico. 35 00:02:24,630 --> 00:02:28,350 And so tell us, what are you doing? We, in the virtual 36 00:02:28,350 --> 00:02:31,750 green room, we spoke briefly, working on building something really cool. 37 00:02:32,390 --> 00:02:35,990 Yeah, so I'm building a quantum lab 38 00:02:35,990 --> 00:02:39,590 out here. So OA 39 00:02:39,670 --> 00:02:43,430 Quantum Labs is not just a quantum lab, so we don't do 40 00:02:43,430 --> 00:02:46,470 just research. All of our research is 41 00:02:46,550 --> 00:02:48,630 100% geared towards 42 00:02:51,620 --> 00:02:55,300 creating true commercial application of quantum 43 00:02:55,300 --> 00:02:58,900 technology. So a good example is we are also 44 00:02:58,980 --> 00:03:01,220 the owners of multiple AI companies, 45 00:03:03,100 --> 00:03:06,820 which we have now acquired. So as 46 00:03:06,820 --> 00:03:08,820 part of that, we are applying 47 00:03:10,260 --> 00:03:14,020 quantum computing in its current state of the 48 00:03:14,020 --> 00:03:17,350 science to multiple different 49 00:03:18,070 --> 00:03:21,430 components within the AI 50 00:03:21,430 --> 00:03:24,790 ecosystem. Interesting. Okay, 51 00:03:24,950 --> 00:03:28,550 okay. How so? Like what particularly, like, I'm curious 52 00:03:29,030 --> 00:03:32,870 to see what the intersection of quantum and AI, sorry. Okay, so the very first 53 00:03:33,110 --> 00:03:36,390 things that we did was reducing 54 00:03:36,950 --> 00:03:40,790 training cost and time. That was the 55 00:03:40,790 --> 00:03:44,350 easiest place where quantum could make the 56 00:03:45,060 --> 00:03:48,660 biggest impact. And this was back when we 57 00:03:48,660 --> 00:03:51,220 were still on, you know, sub-100 58 00:03:51,940 --> 00:03:55,700 qubit systems, really in the 50s somewhere, logical qubits. 59 00:03:58,100 --> 00:04:00,820 Now what we are doing is we're also improving 60 00:04:01,460 --> 00:04:04,020 AI inference in a number of areas. 61 00:04:05,220 --> 00:04:07,860 So if you, I'm going to 62 00:04:08,020 --> 00:04:11,460 oversimplify this a little bit to the point of it almost 63 00:04:11,540 --> 00:04:14,740 being wrong, but it provides a good analogy. 64 00:04:15,540 --> 00:04:19,220 One of the things that quantum computing is really, really, really 65 00:04:19,220 --> 00:04:22,740 good at is math, right? It does 66 00:04:22,980 --> 00:04:26,820 math and complex math very, very quickly. So 67 00:04:26,820 --> 00:04:30,660 if you think of a quantum computer almost 68 00:04:30,820 --> 00:04:32,980 like a super ridiculous 69 00:04:33,940 --> 00:04:37,460 calculator, you can use AI 70 00:04:38,100 --> 00:04:41,860 for all of its inference, but when math needs to take place, 71 00:04:41,860 --> 00:04:45,570 you throw the math to the quantum computer, get the math back, and 72 00:04:45,570 --> 00:04:49,370 done, and then it comes back. Where this works the best is 73 00:04:49,450 --> 00:04:52,970 in materials. When you're, when you're doing anything with 74 00:04:52,970 --> 00:04:54,330 materials or molecules. 75 00:04:56,330 --> 00:05:00,050 Interesting. I mean, that 76 00:05:00,050 --> 00:05:03,730 makes sense, right? Because there's definitely a tight correlation between 77 00:05:03,730 --> 00:05:07,530 quantum effects and chemistry. Yes. And it sounds a 78 00:05:07,530 --> 00:05:11,130 bit like, like a GPU, right? In a sense, right? Like almost. You— 79 00:05:11,530 --> 00:05:14,670 that's what I said, a bit like, right? Like you're sending off whether it's a 80 00:05:14,670 --> 00:05:18,430 video game, whether it's AI or neural network training, 81 00:05:18,430 --> 00:05:22,030 you're just saying, here's a bunch of stuff, GPU, go for it, right? 82 00:05:22,030 --> 00:05:25,190 Yes. And then you come back with an answer. Yeah, yeah. Only you're saying 83 00:05:25,190 --> 00:05:28,510 QPU, go for it. QPU, yeah, right. Yeah, 84 00:05:29,070 --> 00:05:32,791 right. I'm hoping that term catches. I'm hoping that term catches on. Yeah, yeah, 85 00:05:32,791 --> 00:05:36,590 exactly. We do. We hear it a lot. Yeah. So QPU 86 00:05:36,590 --> 00:05:40,190 is actually already a term on that thing, but 87 00:05:40,990 --> 00:05:44,670 Quantum computing, the architecture of quantum computers is different. 88 00:05:45,230 --> 00:05:48,750 It's not really the same architecture where you have 89 00:05:49,230 --> 00:05:52,630 a central processing unit and then memory 90 00:05:52,630 --> 00:05:56,350 sits somewhere else, and then you have 91 00:05:56,509 --> 00:06:00,230 buses between your CPU and your— it's not like 92 00:06:00,230 --> 00:06:03,950 that in the quantum world. Interesting. The 93 00:06:04,430 --> 00:06:07,270 memory is, let's call it, on-chip. 94 00:06:09,650 --> 00:06:13,450 Right. Well, there's also kind of— there's also the thing, like, 95 00:06:13,450 --> 00:06:16,890 once you read the memory, do you collapse the quantum state? I know that once 96 00:06:16,890 --> 00:06:20,610 you get into kind of the brass tacks of, you know, beyond 97 00:06:20,610 --> 00:06:24,290 like the theoretical, like you start to get some— it starts to get weird 98 00:06:24,290 --> 00:06:27,930 real fast, right? Because like, you know, how do you— debugging a 99 00:06:27,930 --> 00:06:31,250 quantum system, right? We've already talked with some other guests about that. Like, that's, you 100 00:06:31,250 --> 00:06:33,730 know, how do you, you know, if you— how do you step through the code, 101 00:06:33,730 --> 00:06:36,170 right? And like you peek at the variables. Well, as soon as you do that 102 00:06:36,170 --> 00:06:40,020 in a quantum system, You're collapsing. You're collapsing and 103 00:06:40,020 --> 00:06:43,640 you kind of lose the advantage of quantum. Yeah, yeah, yeah. So like, I, I 104 00:06:43,640 --> 00:06:47,380 would imagine that there's a lot of these little gotchas that nobody's really fully kind 105 00:06:47,380 --> 00:06:51,140 of worked through just yet. Um, so there are, and this 106 00:06:51,140 --> 00:06:54,940 is why, um, as long 107 00:06:55,020 --> 00:06:58,460 as you understand what the limitations are, 108 00:06:59,100 --> 00:07:02,700 quantum has some really significant advantages 109 00:07:02,940 --> 00:07:06,620 right now. This is the reason 110 00:07:06,620 --> 00:07:10,100 why JP Morgan, as an example, is spending $1.5 billion 111 00:07:11,100 --> 00:07:14,780 on quantum computing, because there are certain things, certain 112 00:07:14,780 --> 00:07:18,300 mathematics, QAOA, right? Quantum Approximate 113 00:07:19,340 --> 00:07:22,860 Optimization Algorithms, right? Where you're using 114 00:07:23,260 --> 00:07:26,780 quantum to do 115 00:07:26,780 --> 00:07:30,510 certain mathematical functions that just take too 116 00:07:30,510 --> 00:07:34,030 long, and they're, you know, take a second or two on quantum, 117 00:07:34,110 --> 00:07:37,910 they take minutes on classical. And, and when 118 00:07:37,910 --> 00:07:41,630 you're in the world of finance, you know, a minute is too 119 00:07:41,630 --> 00:07:43,950 long. Same goes with, 120 00:07:46,350 --> 00:07:49,990 um, um, advanced correlation algorithms. You get into quantum advanced 121 00:07:49,990 --> 00:07:53,070 correlation algorithms, and those run really 122 00:07:53,070 --> 00:07:56,440 ridiculously a lot faster. Right. 123 00:07:56,840 --> 00:08:00,560 But not for all problems, just certain. Yes, that's what I'm saying. Yeah. 124 00:08:00,560 --> 00:08:04,200 When you, when you understand what problem domains quantum is 125 00:08:04,200 --> 00:08:07,800 really good at, it becomes a lot easier, 126 00:08:07,800 --> 00:08:11,560 faster to start applying commercial application 127 00:08:11,639 --> 00:08:15,480 to it. Gotcha. So you 128 00:08:15,480 --> 00:08:19,320 talked about the financial sector. What other, what 129 00:08:19,320 --> 00:08:22,880 other industries are you think, do you think are the most primed to benefit 130 00:08:22,880 --> 00:08:26,150 first? Okay, so where 131 00:08:27,270 --> 00:08:31,110 it's already benefiting is anything where you need molecular 132 00:08:31,270 --> 00:08:35,110 or quantum knowledge or effects or whatever, right? 133 00:08:35,670 --> 00:08:38,710 So material sciences is a big one. 134 00:08:39,589 --> 00:08:43,150 In partnership with Ursulaing Quantum 135 00:08:43,150 --> 00:08:46,910 Innovations, we have created the single 136 00:08:46,910 --> 00:08:50,670 most advanced materials, let's call 137 00:08:50,670 --> 00:08:54,470 it, engineering platform in the 138 00:08:54,470 --> 00:08:57,790 world, right? Our nearest 139 00:08:57,790 --> 00:09:01,550 competitor is— oh my gosh, I was just going to 140 00:09:01,550 --> 00:09:02,030 say them. 141 00:09:05,790 --> 00:09:09,590 They just got this massive amount of money and I totally 142 00:09:09,590 --> 00:09:12,910 spaced their name. Cusp AI, I think that's what it is. 143 00:09:14,030 --> 00:09:17,750 They— so they're supposed to be a material science platform. They need 6 144 00:09:17,750 --> 00:09:21,150 months and an entire team of material sciences 145 00:09:21,740 --> 00:09:25,500 scientists to do almost anything. And we 146 00:09:25,500 --> 00:09:29,340 were spinning out new materials at the pace of a new one every 147 00:09:29,340 --> 00:09:33,140 2 weeks. Oh, wow. Okay. Yeah. Like, 148 00:09:33,460 --> 00:09:37,140 we have— we got— we created a material that 149 00:09:37,140 --> 00:09:40,620 is stronger and harder than 150 00:09:42,260 --> 00:09:46,060 carbon fiber, but about half the price to manufacture. 151 00:09:47,180 --> 00:09:50,860 We created brand new heat shielding that 152 00:09:51,820 --> 00:09:55,400 survives multiple multiple reentries and is 153 00:09:55,480 --> 00:09:58,760 far less expensive to produce than what SpaceX is using today. 154 00:10:00,440 --> 00:10:03,880 We created a new material, a new 155 00:10:03,880 --> 00:10:05,560 advanced material for 156 00:10:07,480 --> 00:10:11,240 heat management. It basically pulls heat away to use as like heat sinks 157 00:10:11,240 --> 00:10:15,040 and those kinds of things that is far better than 158 00:10:15,040 --> 00:10:18,840 anything that exists. So we finally— we were creating so 159 00:10:18,840 --> 00:10:22,400 many new materials so fast that we overran the sales team's ability 160 00:10:22,400 --> 00:10:26,020 to keep up, so we spun that out into a brand new company, 161 00:10:26,980 --> 00:10:29,780 and now that, that guy is off to the races. 162 00:10:31,530 --> 00:10:34,500 Um, and, uh, so the other place where, 163 00:10:35,780 --> 00:10:39,300 um, it helps a lot is again in modeling, uh, quantum 164 00:10:39,300 --> 00:10:42,980 effects. I was able to create a whole brand new 165 00:10:42,980 --> 00:10:46,540 GPU kernel that is far better than Flash 166 00:10:46,540 --> 00:10:50,180 Attention V2 because I modeled how 167 00:10:51,210 --> 00:10:54,890 electrons flow through a GPU and therefore was 168 00:10:54,890 --> 00:10:58,450 able to optimize the code for how the attention 169 00:10:58,450 --> 00:11:00,170 mechanisms work on inference. 170 00:11:01,850 --> 00:11:02,570 Interesting. 171 00:11:05,050 --> 00:11:05,770 Interesting. 172 00:11:09,210 --> 00:11:11,930 What sorts of hardware does this run on? I'm sorry, 173 00:11:13,050 --> 00:11:16,090 Candace. No, no, go ahead. What sorts of hardware? Is it hardware agnostic? Oh, no, 174 00:11:16,090 --> 00:11:19,610 no, no. So I mean, I wrote it to be very specific 175 00:11:20,330 --> 00:11:23,600 to the NVIDIA H100, 176 00:11:23,840 --> 00:11:27,520 A100, and above better, right? That makes sense. Yeah, this 177 00:11:27,520 --> 00:11:30,840 is, this, I, when I wrote it, it was when the, 178 00:11:32,100 --> 00:11:35,800 uh, uh, X was coming out with all the news about their brand 179 00:11:35,800 --> 00:11:39,600 new Colossus supercluster, blah, blah, blah. And I was like, I wonder 180 00:11:39,600 --> 00:11:43,400 if I could, since I can model, um, molecules 181 00:11:43,400 --> 00:11:47,160 and all that kind of, and quantum effects and all that other kinds of 182 00:11:47,160 --> 00:11:50,800 good stuff, can I model how things flow 183 00:11:50,800 --> 00:11:54,580 through a GPU and therefore improve on 184 00:11:55,860 --> 00:11:59,220 improve on how the attention mechanism 185 00:11:59,300 --> 00:12:03,100 works within a GPU, and it's better by a 186 00:12:03,100 --> 00:12:05,820 lot. Interesting. Between 1.5 and 187 00:12:06,380 --> 00:12:09,780 3x improved inference. Oh, wow. 188 00:12:10,180 --> 00:12:13,980 Yeah. Depending on where you 189 00:12:13,980 --> 00:12:16,900 are in the stack, do you need sparse attention or— 190 00:12:19,540 --> 00:12:22,740 all of a sudden I drew a blank on the name— sparse attention or 191 00:12:26,950 --> 00:12:30,590 Wow. I deal with this every day and all of a sudden I blanked. It 192 00:12:30,590 --> 00:12:32,990 happens to the best of us. Of course, attention is the thing that they— that 193 00:12:32,990 --> 00:12:35,190 you really kind of need the least of. 194 00:12:37,110 --> 00:12:40,310 But yeah, so I got you. Okay. I mean, that makes sense. 195 00:12:41,910 --> 00:12:45,550 Look, honestly, everything you're talking about is so incredibly exciting. So how 196 00:12:45,550 --> 00:12:48,550 do newcomers interested in quantum and AI 197 00:12:48,950 --> 00:12:51,990 researchers, entrepreneurs, investors, 198 00:12:52,810 --> 00:12:56,650 What advice or first steps would you recommend today to get 199 00:12:56,650 --> 00:13:00,170 them involved meaningfully? Okay, so 200 00:13:00,570 --> 00:13:04,410 that's really gonna depend on which one of those you're talking 201 00:13:04,410 --> 00:13:07,530 about. For investors, 202 00:13:08,090 --> 00:13:11,610 the biggest thing that I would say is to, 203 00:13:13,530 --> 00:13:17,050 is in two areas. Number one, look for people that 204 00:13:17,050 --> 00:13:20,570 don't necessarily have the pedigree. It becomes really, 205 00:13:20,970 --> 00:13:24,600 really easy in quantum to assume 206 00:13:25,080 --> 00:13:28,600 that somebody must— that you got to have the PhD, and the more 207 00:13:28,600 --> 00:13:32,360 PhDs on the team, the better. And you see a lot of that. 208 00:13:32,440 --> 00:13:36,160 You look at D-Wave, Quantinuum, Qera, right? You 209 00:13:36,160 --> 00:13:39,399 look at all of these guys, and what you see is this long list of 210 00:13:39,399 --> 00:13:43,200 PhDs. And the truth is, is that companies like mine are 211 00:13:43,200 --> 00:13:46,840 completely blowing their doors off. Like, I don't— 212 00:13:46,840 --> 00:13:49,280 I'm rapidly getting to the point I don't even know how they're going to keep 213 00:13:49,280 --> 00:13:52,770 up. We have created a quantum 214 00:13:52,770 --> 00:13:56,450 error correction algorithm that 215 00:13:56,450 --> 00:13:59,850 reduces physical to logical overhead by 100. Wow. 216 00:14:02,410 --> 00:14:06,090 So yeah, it's better. It's better by so much. 217 00:14:06,970 --> 00:14:08,090 It's almost 218 00:14:11,730 --> 00:14:15,410 hard to believe. And we had to run it on 219 00:14:15,410 --> 00:14:19,250 the IBM Lima and Bellum benchmarks. We had to run 220 00:14:19,250 --> 00:14:23,070 that thing 3 times because we sort of assumed that 221 00:14:23,070 --> 00:14:26,710 it couldn't have been right the first time. You know, like, how 222 00:14:27,070 --> 00:14:30,230 is this good? So, 223 00:14:31,430 --> 00:14:34,790 so number one, look for people that are actually doing it. 224 00:14:35,750 --> 00:14:39,350 Number two, look for people that, that don't just need a check. 225 00:14:39,990 --> 00:14:43,510 Right, right. So for— too, 226 00:14:45,190 --> 00:14:48,950 too often investors are giving money to people either because 227 00:14:48,950 --> 00:14:51,910 of pedigree or because they go, oh, this guy has got, 228 00:14:54,050 --> 00:14:57,570 you know, two successful exits. So probably they can do a third one. But you 229 00:14:57,570 --> 00:15:01,330 look back at history and that's not true at all, right? Look at, 230 00:15:01,330 --> 00:15:05,090 look at Pets.com. Pets.com from, you know, way back in 231 00:15:05,090 --> 00:15:08,770 the dot-bomb era, right? It was started by multiple 232 00:15:08,770 --> 00:15:12,530 people that had done multiple different successful exits. And that 233 00:15:12,530 --> 00:15:16,090 thing was a disaster, right? No, it's true. 234 00:15:16,090 --> 00:15:19,930 And, you know, you mentioned that and 3DO, do you remember? Speaking 235 00:15:19,930 --> 00:15:23,770 of the '90s. Yes. 3DO, 3DO was like, I remember the 236 00:15:23,770 --> 00:15:27,510 Wired magazine article. Cover was like the digital start of the 237 00:15:27,510 --> 00:15:31,150 rise of the digital supergroup. And aside from like a handful of people 238 00:15:31,150 --> 00:15:34,830 who remember the '90s, no one knows what 3DL was, right? 239 00:15:34,830 --> 00:15:38,670 No, it was just like— and you're right, like pedigree. I think, I 240 00:15:38,670 --> 00:15:41,430 think there's a temptation. I think this brings up a deep point. Like, there's a 241 00:15:41,430 --> 00:15:43,790 temptation to overbuy 242 00:15:45,070 --> 00:15:48,670 on pedigree. Yes. Whether, 243 00:15:48,750 --> 00:15:52,510 whether it's in quantum, the assumption that, well, how could you possibly 244 00:15:52,510 --> 00:15:55,710 understand quantum if you don't have a PhD? And the answer is 245 00:15:56,520 --> 00:16:00,240 look at the solutions that are created, right? And then, and 246 00:16:00,240 --> 00:16:03,800 then the second thing, if somebody— just because somebody says I have 247 00:16:03,800 --> 00:16:07,400 something, or maybe they actually do— I mean, and this is something 248 00:16:07,560 --> 00:16:11,400 most venture capitalists or investors are already pretty good at, 249 00:16:11,400 --> 00:16:15,000 is going, let me see your customers. Um, you 250 00:16:15,000 --> 00:16:18,280 know, is there actually market traction for it? 251 00:16:19,320 --> 00:16:23,040 At the end of the day, a company— uh, so OA Quantum 252 00:16:23,040 --> 00:16:26,810 Labs isn't looking for an investor. But assuming that we were, 253 00:16:28,170 --> 00:16:31,970 we have— we don't only have solutions, we have customers. So because we have 254 00:16:31,970 --> 00:16:35,730 solutions and customers, like, I don't need your money. If I was going to— 255 00:16:35,730 --> 00:16:39,370 if I was going to take money from an investor, 256 00:16:39,450 --> 00:16:43,130 it would only be because that investor was bringing me 257 00:16:43,610 --> 00:16:47,050 some kind of strategic alliance that 258 00:16:47,450 --> 00:16:51,290 I like, that it's worth more than the equity that I 259 00:16:51,290 --> 00:16:54,860 would give up. Does that make sense? No, I mean, that makes sense. Yeah, no, 260 00:16:55,020 --> 00:16:57,260 I think, I think it's an interesting point you bring up. Like, 261 00:16:58,700 --> 00:17:02,300 um, it's about selling solutions. Yeah, not the science, 262 00:17:02,300 --> 00:17:05,700 right? It's almost like you— we got to give you a free copy of our 263 00:17:05,700 --> 00:17:08,780 book on, uh, selling quantum solutions, right? 264 00:17:10,140 --> 00:17:13,340 Um, because like, it, it's almost like you've read it. Like, because you're basically saying 265 00:17:13,500 --> 00:17:17,180 effectively the same thing, just like it. Yeah, you know that you're right. 266 00:17:17,180 --> 00:17:20,079 Like, if you can if you could prove the value— and I forget what it 267 00:17:20,079 --> 00:17:23,919 was, it was like months versus weeks— like, you could prove that real value to 268 00:17:23,919 --> 00:17:27,639 a business, doesn't really matter how many PhDs you have. I mean, obviously, right, 269 00:17:27,639 --> 00:17:31,479 obviously somebody has to, you know, check the numbers and make sure the answers you 270 00:17:31,479 --> 00:17:35,199 get are, you know, legit. Uh, but I mean, at some point 271 00:17:35,519 --> 00:17:38,959 it's really where the rubber meets the road, right? Like, I would not have thought, 272 00:17:39,839 --> 00:17:42,479 uh, if you look at pets.com compared to Amazon, 273 00:17:44,440 --> 00:17:47,720 um, Who, I mean, in the '90s, people would have assumed Pets.com would have won. 274 00:17:47,720 --> 00:17:51,480 Barnes Noble, like the same story. Fun fact, I worked at 275 00:17:51,480 --> 00:17:55,240 BarnesandNoble.com. Oh, wow. I was the first 276 00:17:55,240 --> 00:17:55,800 webmaster there. 277 00:17:58,600 --> 00:18:01,800 Wow. Underestimating people who are relentless is a 278 00:18:02,200 --> 00:18:06,040 mistake. Yes. Yeah. So when 279 00:18:06,040 --> 00:18:09,720 it comes to, let's say it's a developer who's interested 280 00:18:09,720 --> 00:18:12,770 in quantum, I would 281 00:18:14,930 --> 00:18:18,610 say that the best way for a developer to 282 00:18:18,610 --> 00:18:22,210 get involved in quantum is to get a Quantum Cloud account 283 00:18:22,290 --> 00:18:25,330 and to start creating 284 00:18:26,449 --> 00:18:29,570 stuff. Don't mess around with research. 285 00:18:30,170 --> 00:18:33,410 Don't, like, I mean, yeah, take some time to learn, you 286 00:18:34,610 --> 00:18:37,890 know, Qisk or whatever the 287 00:18:41,990 --> 00:18:45,510 different, DLLs that get wrapped up into Python. 288 00:18:46,430 --> 00:18:50,270 But yeah, take some time to learn what you're writing. But as soon as 289 00:18:50,270 --> 00:18:53,110 you can learn something, start creating something from 290 00:18:54,229 --> 00:18:57,750 it. Don't sit around and wait, create something from 291 00:18:58,550 --> 00:19:02,150 it because there is 292 00:19:02,150 --> 00:19:05,770 no substitute for experience. The 293 00:19:05,770 --> 00:19:09,610 problem that most developers have is that they've spent their entire 294 00:19:09,770 --> 00:19:13,490 lives either A, in school being taught, or 295 00:19:13,490 --> 00:19:17,290 B, in their careers on classical binary 296 00:19:17,290 --> 00:19:20,929 digital computers that are very time-dependent 297 00:19:20,929 --> 00:19:24,530 and sequentially processed. Whereas quantum 298 00:19:24,530 --> 00:19:28,370 is non-time-dependent and simultaneously processed. 299 00:19:28,370 --> 00:19:31,630 And therefore the way you have to even think 300 00:19:31,950 --> 00:19:34,830 about how you architect a 301 00:19:35,630 --> 00:19:39,310 solution is different. How you think about how you're going to write the code is 302 00:19:39,630 --> 00:19:43,390 different, and you don't know those things, or it's— I would 303 00:19:43,390 --> 00:19:47,190 be better to say it's hard to understand those things until 304 00:19:47,190 --> 00:19:50,270 you start writing the code and start seeing what 305 00:19:51,070 --> 00:19:54,710 happens, right? Right. So that's a good way to 306 00:19:54,710 --> 00:19:58,400 put it. Yeah. Sorry, Candice, I'll be quiet now. No, I'm just 307 00:19:58,400 --> 00:20:01,640 thinking about, you know, you come from such a 308 00:20:01,640 --> 00:20:05,040 unique background because most leaders are, you know, they're either in 309 00:20:05,040 --> 00:20:08,120 the quantum world or they're in the AI world. 310 00:20:09,240 --> 00:20:12,840 But because you're in both, it gives you such a 311 00:20:12,840 --> 00:20:15,880 unique advantage to have this dual 312 00:20:18,040 --> 00:20:21,480 fluency. So how do you find that that affects, you know, you 313 00:20:21,480 --> 00:20:24,740 being founder and CTO and CEO of of your 314 00:20:26,820 --> 00:20:30,340 company? It definitely in a lot of 315 00:20:31,300 --> 00:20:34,900 ways makes the commercial potential 316 00:20:34,900 --> 00:20:36,740 and applicability of what I'm 317 00:20:38,420 --> 00:20:42,020 doing better or easier. It means 318 00:20:42,100 --> 00:20:45,620 that when I am selling solutions, I 319 00:20:45,620 --> 00:20:49,260 can articulate to people like, this 320 00:20:49,260 --> 00:20:52,340 is, this is why what, what we're doing works 321 00:20:53,570 --> 00:20:56,450 better., right? And I'm able to 322 00:20:57,730 --> 00:21:01,530 speak to, you know, the CTOs of people. AI is 323 00:21:01,530 --> 00:21:05,170 getting to be understood well 324 00:21:06,210 --> 00:21:08,690 enough now in the enterprise and all those kinds 325 00:21:10,290 --> 00:21:14,130 of things that when I start to explain, okay, this is where 326 00:21:14,130 --> 00:21:17,970 the AI is and this is where the quantum is and this is why the 327 00:21:18,920 --> 00:21:22,680 quantum matters. It's a pretty simple conversation 328 00:21:22,840 --> 00:21:26,280 to have these days, especially now that they 329 00:21:26,280 --> 00:21:29,400 know that 330 00:21:29,720 --> 00:21:33,480 I'm bringing quantum enhancement. I'm not saying this 331 00:21:33,560 --> 00:21:37,080 is a quantum solution, it's a 332 00:21:37,160 --> 00:21:40,200 quantum enhancement. And that's— it's a very 333 00:21:41,000 --> 00:21:44,670 subtle distinction, but the gap between them is, you know, 334 00:21:45,460 --> 00:21:48,900 about the distance from one side of the Grand Canyon to 335 00:21:49,860 --> 00:21:53,708 the other. Well, it also frames the conversation differently. Sorry, Ken. No, I 336 00:21:53,708 --> 00:21:57,460 was thinking, so does that mean that the AI accelerates 337 00:21:57,860 --> 00:22:01,620 the quantum? No, the other way around. So the quantum accelerates 338 00:22:01,860 --> 00:22:05,380 the AI? 339 00:22:07,220 --> 00:22:10,660 Yes. Yes. Yeah. And, and it's because it's quantum 340 00:22:10,660 --> 00:22:14,330 accelerating the AI By having the discussion 341 00:22:14,330 --> 00:22:16,610 in that way, it means that 342 00:22:18,450 --> 00:22:22,090 the business people can understand it better. It means I 343 00:22:22,090 --> 00:22:25,330 can now have a much 344 00:22:25,650 --> 00:22:29,490 quicker conversation about this is what it means to your bottom line, because at the 345 00:22:29,490 --> 00:22:33,090 end of the day, that's what really matters, right? 346 00:22:33,090 --> 00:22:36,690 Right. If you're going to go to any enterprise, you had better be 347 00:22:36,690 --> 00:22:40,400 able to answer be able to say that either A, 348 00:22:40,400 --> 00:22:44,080 my solution is going to improve revenue, or B, it's going to 349 00:22:44,080 --> 00:22:47,880 reduce your cost and therefore improve profit. If you can't say 350 00:22:47,880 --> 00:22:51,640 it's going to do A, B, or both, don't 351 00:22:51,640 --> 00:22:53,920 even bother having the discussion because it 352 00:22:55,760 --> 00:22:58,240 doesn't matter, right? It's all about 353 00:23:00,160 --> 00:23:03,040 solutioning. Yes. Oh yes, not tech for the sake of tech. I mean, right, tech 354 00:23:03,040 --> 00:23:06,160 for the sake of tech is 355 00:23:06,750 --> 00:23:10,590 an academic conversation, correct? And that's fine for 356 00:23:10,590 --> 00:23:14,190 academia, but not outside of academia, correct? And 357 00:23:14,270 --> 00:23:16,990 I'm— and to 358 00:23:17,470 --> 00:23:21,230 me, one, quantum has gone far enough 359 00:23:21,230 --> 00:23:24,670 now that it no longer even should be 360 00:23:25,230 --> 00:23:29,070 in academia. And this is why you're seeing, even though everybody's— there's 361 00:23:29,070 --> 00:23:32,670 been a lot of news stories lately about, you know, the bursting of the 362 00:23:33,550 --> 00:23:37,390 quantum bubble. Or whatever. And D-Wave 363 00:23:37,470 --> 00:23:41,070 and Quantinuum and Rigetti have all been, let's call it punished a 364 00:23:41,310 --> 00:23:44,830 little bit. But the truth is, is that 365 00:23:44,830 --> 00:23:48,430 as we start having more of a 366 00:23:48,909 --> 00:23:52,710 business discussion, this is the business problems that we are solving 367 00:23:52,710 --> 00:23:56,310 right now today, the more that 368 00:23:56,310 --> 00:24:00,030 discussion goes away because now quantum starts moving into the data center 369 00:24:00,510 --> 00:24:04,070 and it really needs to get there for there to 370 00:24:04,070 --> 00:24:07,890 be additional significant investment investment to improve the technology. 371 00:24:07,890 --> 00:24:11,730 That makes a lot of sense. 372 00:24:11,730 --> 00:24:14,970 Yeah, yeah, we got to get it out of the research lab and into 373 00:24:15,290 --> 00:24:19,050 the enterprise. Very important. 374 00:24:19,370 --> 00:24:22,970 So when, if you're to look back earlier in, 375 00:24:22,970 --> 00:24:26,610 in your work, when, when you had 376 00:24:26,610 --> 00:24:30,370 that moment where you realized that this isn't just something theoretical, 377 00:24:30,370 --> 00:24:32,710 but this is actually something that I 378 00:24:36,060 --> 00:24:37,180 can commercialize What clicked 379 00:24:39,740 --> 00:24:43,540 for you? Okay, so one of the companies that 380 00:24:43,540 --> 00:24:47,020 I acquired is HughieBT. I was originally the CTO 381 00:24:47,020 --> 00:24:49,900 of HughieBT. 382 00:24:50,940 --> 00:24:52,900 Um, so 383 00:24:55,500 --> 00:24:58,620 at HughieBT, we have the most 384 00:24:58,780 --> 00:25:02,590 advanced digital identity solution by a very wide margin. 385 00:25:02,590 --> 00:25:06,390 Nobody else is even close. And, and it's a— we use 386 00:25:06,390 --> 00:25:10,071 behavioral biometrics as a way of, uh, it's— 387 00:25:10,071 --> 00:25:13,630 we are 99 point and then add 388 00:25:14,990 --> 00:25:18,670 7 nines percent of ability 389 00:25:19,230 --> 00:25:22,950 to distinguish between one human and another human. 390 00:25:22,950 --> 00:25:26,430 And because it's that accurate, it means we are that 391 00:25:26,430 --> 00:25:30,130 accurate also distinguishing between a deepfake. I have had 392 00:25:30,130 --> 00:25:33,970 people create deepfakes of themselves and not be able to defeat, 393 00:25:33,970 --> 00:25:37,410 um, our solution. Okay. 394 00:25:38,130 --> 00:25:41,970 Yeah, so the— when it clicked was 395 00:25:41,970 --> 00:25:45,770 when, um, the training was taking too long, and I was like, okay, well, what 396 00:25:45,770 --> 00:25:49,490 can I do? What can I do to this 397 00:25:50,050 --> 00:25:52,530 stupid thing? Um, and this is one of 398 00:25:53,740 --> 00:25:54,220 these weird 399 00:25:57,660 --> 00:26:01,420 sort of, um, so AI is an odd thing in general. It's, it 400 00:26:01,420 --> 00:26:05,020 can be odd. I, I am well known 401 00:26:05,020 --> 00:26:08,860 for saying that AI is 402 00:26:08,860 --> 00:26:12,380 really some shockingly simple 403 00:26:12,620 --> 00:26:15,900 algorithms, and it is about as intelligent actually 404 00:26:15,900 --> 00:26:19,620 as your calculator, right? Everybody wants to talk 405 00:26:19,620 --> 00:26:23,420 about, you know, is AI sentient? Is AI conscious? Is AI, you 406 00:26:23,420 --> 00:26:25,961 know, and how soon are we going to get to AGI? I don't think we're 407 00:26:25,961 --> 00:26:29,400 going to get to AGI anytime soon. I really don't. 408 00:26:30,600 --> 00:26:34,360 In fact, they've tried 409 00:26:34,760 --> 00:26:38,560 to change where AGI is from it being 410 00:26:38,560 --> 00:26:42,040 able to reason as good as a human to simply 411 00:26:42,120 --> 00:26:44,040 being able— being as— what's the word 412 00:26:46,440 --> 00:26:50,280 they use? Not learn. It's like adaptability or something. Like, 413 00:26:50,360 --> 00:26:53,950 they changed the bar for what's going to be 414 00:26:54,110 --> 00:26:57,790 considered AGI from reasoning capability to adaptability 415 00:26:57,790 --> 00:27:00,830 or something like that. And I just rolled my eyes and went, well, this 416 00:27:01,870 --> 00:27:05,510 is stupid. To me, it's not. Yeah, if you can't 417 00:27:05,510 --> 00:27:09,070 reason as good as even, you know, 418 00:27:09,390 --> 00:27:13,230 an average IQ person, then that's 419 00:27:13,230 --> 00:27:17,070 not artificial general intelligence. It's just not. 420 00:27:17,790 --> 00:27:20,150 So anyway, I Just on 421 00:27:21,350 --> 00:27:23,990 a lark, I asked the 422 00:27:26,710 --> 00:27:30,150 AI to consider itself as a 423 00:27:30,310 --> 00:27:34,070 high IQ materials scientist and to give 424 00:27:34,470 --> 00:27:38,270 me ways that I could 425 00:27:38,270 --> 00:27:41,510 improve the speed of training 426 00:27:41,510 --> 00:27:45,350 of the application. What it came up with 427 00:27:46,470 --> 00:27:50,230 was basically use quantum computing and it also 428 00:27:50,230 --> 00:27:53,850 output a whole bunch bunch of Cirq code. And I was 429 00:27:53,850 --> 00:27:57,650 like, well, this is interesting. So just as 430 00:27:57,650 --> 00:27:59,730 a hint to your audience, I know this is a 431 00:28:01,890 --> 00:28:05,490 quantum thing, this, but, but just as a sort of trick with 432 00:28:05,490 --> 00:28:09,330 AI, if you tell an AI 433 00:28:09,410 --> 00:28:13,090 to act in a role that 434 00:28:13,490 --> 00:28:17,050 is only sort of tertiary to 435 00:28:17,050 --> 00:28:20,850 what its actual thing that you're asking it to do, let's say you wanted to 436 00:28:21,330 --> 00:28:24,770 review code, Tell it that it's a chemist and to review 437 00:28:26,210 --> 00:28:29,810 the code from the viewpoint of a chemist. It 438 00:28:30,210 --> 00:28:33,490 will actually be more, for lack of a 439 00:28:33,730 --> 00:28:37,329 better word, creative. I know AI isn't actually 440 00:28:37,330 --> 00:28:40,050 creative, but sort of. It comes up with some 441 00:28:40,770 --> 00:28:44,450 really interesting responses that I have 442 00:28:44,450 --> 00:28:46,680 found dramatically improves it often, 443 00:28:48,110 --> 00:28:51,870 its output. Because of its having to, like 444 00:28:51,870 --> 00:28:55,550 I said, for lack of a better word, be creative. But anyway, so the 445 00:28:55,550 --> 00:28:58,910 very first thing that I did was 446 00:28:59,150 --> 00:29:01,950 implement quantum to improve the training 447 00:29:03,550 --> 00:29:07,230 of HuGPT. Then that grew into improving the 448 00:29:07,230 --> 00:29:10,470 inference of HuGPT. But in improving the 449 00:29:10,470 --> 00:29:14,080 inference of HuGPT, I sort of, 450 00:29:14,400 --> 00:29:18,240 by accident, for lack of a better way of putting it, created 451 00:29:18,400 --> 00:29:22,200 this system for how molecules and all of those kinds 452 00:29:22,200 --> 00:29:25,600 of things are modeled. I created a physics-informed 453 00:29:27,600 --> 00:29:30,000 neural network. Let me rephrase that, a 454 00:29:31,200 --> 00:29:34,800 quantum-enhanced physics-informed neural network. That then grew to where I 455 00:29:35,680 --> 00:29:39,120 was using PINs, PINOs. So, 456 00:29:39,520 --> 00:29:42,720 PIN, physics-informed neural network, physics-informed neural 457 00:29:43,270 --> 00:29:47,070 operator, GAN, which is a graph 458 00:29:47,070 --> 00:29:49,510 neural network, and a GNO, which is a graph 459 00:29:51,310 --> 00:29:54,950 neural operator. I started putting all of these things together and stuck quantum in the 460 00:29:54,950 --> 00:29:58,790 middle of it for doing the math, and then that 461 00:29:58,869 --> 00:30:02,550 grew into materials and grew into molecular modeling 462 00:30:02,550 --> 00:30:06,390 and all those 463 00:30:06,750 --> 00:30:10,390 things. It came to me, for lack of a 464 00:30:11,120 --> 00:30:14,320 better word, by 465 00:30:14,640 --> 00:30:18,080 accident because of output from, from 466 00:30:18,480 --> 00:30:21,920 an AI. And then just from deep 467 00:30:22,000 --> 00:30:25,040 diving into quantum is that's how these 468 00:30:26,080 --> 00:30:29,400 things happened. 469 00:30:29,400 --> 00:30:33,120 Interesting. What misconceptions do you run into the most when people 470 00:30:33,280 --> 00:30:37,000 hear AI plus quantum, and how do you try to 471 00:30:37,000 --> 00:30:40,490 reframe the conversation so they understand what's actually 472 00:30:40,730 --> 00:30:44,530 possible? The biggest one is they think I'm running the AI 473 00:30:45,410 --> 00:30:49,170 on quantum. Right? That's the biggest one. They go, you're running an AI on quantum. 474 00:30:49,170 --> 00:30:52,930 And then, you know, we go back into the whole, you know, 475 00:30:53,010 --> 00:30:56,770 is it conscious or whatever thing? 476 00:30:56,770 --> 00:31:00,450 And which I admittedly have a pretty 477 00:31:00,450 --> 00:31:04,130 low tolerance for. It irritates me when I hear, 478 00:31:04,130 --> 00:31:07,960 you know, people wanting to talk about you know, how intelligent 479 00:31:07,960 --> 00:31:11,600 they are, that they might be conscious or might be sentient or 480 00:31:11,600 --> 00:31:15,320 whatever. That stuff really is a pet peeve. I don't know why it drives me 481 00:31:15,400 --> 00:31:16,200 so crazy, but 482 00:31:19,000 --> 00:31:22,680 it does. But so anyway, that's the first misconception. 483 00:31:22,680 --> 00:31:26,320 The second misconception, and it comes from people 484 00:31:26,320 --> 00:31:29,480 within the AI industry, is the 485 00:31:29,640 --> 00:31:32,920 belief that quantum doesn't really 486 00:31:32,920 --> 00:31:36,520 have commercial application, that it doesn't really apply 487 00:31:36,920 --> 00:31:40,500 to AI, and Oh, you're just playing a game. 488 00:31:40,660 --> 00:31:44,420 You're not, you're not really doing what you're saying. You're not 489 00:31:44,740 --> 00:31:48,540 really doing blah, blah, blah. I'm like, you know, it's kind of hard to 490 00:31:48,540 --> 00:31:51,700 argue with the results, right? At the end of 491 00:31:52,340 --> 00:31:56,180 the day, you ask, you give me a problem domain for a 492 00:31:56,180 --> 00:31:59,860 material and I can spin out that material in 2 weeks. You tell me how 493 00:31:59,860 --> 00:32:02,580 I'm doing that without quantum enhancing a lot of 494 00:32:03,640 --> 00:32:07,360 different things. Right. And my nearest competitor needs 6 495 00:32:07,360 --> 00:32:10,200 months. The nearest competitor from them needs 496 00:32:11,880 --> 00:32:15,240 18 months. Right. Schrödinger needs 497 00:32:17,720 --> 00:32:20,840 18 months. Right. There's a lot to that, right? Like, 498 00:32:21,240 --> 00:32:25,040 you know, there's this idea of 499 00:32:25,040 --> 00:32:28,680 speed that Grant Cardone, one of the, one of my favorite kind of sales authors. 500 00:32:28,680 --> 00:32:32,460 Yeah. I love him. Yeah, everyone, you either love him or you hate him. There's 501 00:32:32,460 --> 00:32:36,020 nobody in the middle. But, um, you know, he has a phrase that, that really 502 00:32:36,020 --> 00:32:39,420 stuck with me. It's called speed is the new big. 503 00:32:39,580 --> 00:32:43,340 Yes, yes, 100% believe that. The phrase I use all the 504 00:32:43,340 --> 00:32:46,940 time is money loves 505 00:32:46,940 --> 00:32:50,780 speed. Yep. Oh, I like that. That's true too. That's 506 00:32:50,780 --> 00:32:54,540 quotable, right? I, I actually think I got that one from 507 00:32:55,350 --> 00:32:58,670 Jay Abraham. I don't know if you remember him or not, but he's another big 508 00:32:58,670 --> 00:33:01,510 sales guy from, from like the 509 00:33:04,230 --> 00:33:07,750 '90s. Interesting. So you're building in a field where things 510 00:33:09,030 --> 00:33:12,310 are evolving daily. Yeah. How do you 511 00:33:14,070 --> 00:33:17,190 stay ahead? Okay, 512 00:33:17,990 --> 00:33:19,990 so here's— things 513 00:33:23,180 --> 00:33:26,940 evolve fast. But when you're in the field, 514 00:33:26,940 --> 00:33:30,060 some of the times you almost wish they would 515 00:33:31,660 --> 00:33:35,100 evolve faster, especially in 516 00:33:36,060 --> 00:33:37,900 quantum. So quantum error 517 00:33:40,060 --> 00:33:43,740 correction has two separate problems. One, you want 518 00:33:43,740 --> 00:33:47,300 to maintain coherence for as long as possible, and number two, you 519 00:33:47,300 --> 00:33:49,930 want to prevent decoherence. 520 00:33:50,730 --> 00:33:53,290 Right? Two sides of the same fence, let's 521 00:33:55,610 --> 00:33:59,370 call it. So because of 522 00:33:59,530 --> 00:34:03,330 the nature of qubits and quantum and all of that kind of— 523 00:34:03,330 --> 00:34:07,010 and all of that, it's a lot 524 00:34:07,010 --> 00:34:10,130 harder. Those two pieces of the puzzle are a lot harder than it sounds. So 525 00:34:10,130 --> 00:34:13,650 even though I've got this really great 526 00:34:13,650 --> 00:34:17,450 quantum error correction algorithm where we can 527 00:34:17,530 --> 00:34:21,190 maintain coherence for about 4x longer, so instead 528 00:34:21,190 --> 00:34:23,270 instead of about 529 00:34:24,710 --> 00:34:27,190 300 microseconds, we're getting 530 00:34:28,630 --> 00:34:31,910 about 1.3 milliseconds we can maintain 531 00:34:32,310 --> 00:34:35,910 coherence, right? The best that we have been able to do on 532 00:34:35,910 --> 00:34:39,550 the decoherence side is predicting, well, these 533 00:34:39,550 --> 00:34:43,390 qubits are likely to decohere, therefore we can ignore those on the other side 534 00:34:43,390 --> 00:34:46,630 of the gate, right? 'Cause why 535 00:34:48,859 --> 00:34:52,619 pay attention? Sort of cut down on the amount of 536 00:34:52,619 --> 00:34:56,139 noise because we're ignoring the, we're 537 00:34:56,139 --> 00:34:59,859 ignoring the qubits that decohered. So you're almost doing quality 538 00:34:59,859 --> 00:35:03,579 assurance or QA on the qubits? Yeah, that's, that's actually a really good way 539 00:35:03,579 --> 00:35:06,539 of putting it. 540 00:35:08,139 --> 00:35:11,939 But we— there, there still needs to be a lot more work done 541 00:35:11,939 --> 00:35:15,019 in the lab on 542 00:35:15,670 --> 00:35:19,030 this aspect of preventing decoherence 543 00:35:19,030 --> 00:35:22,830 and maintaining coherence, because there's only so much that can 544 00:35:22,830 --> 00:35:26,630 be done on the software side, so let's call it, or 545 00:35:26,630 --> 00:35:29,990 the kernel side, where 546 00:35:30,390 --> 00:35:33,750 for that preventing decoherence 547 00:35:33,750 --> 00:35:37,510 or maintaining coherence, right? I can help 548 00:35:37,510 --> 00:35:41,150 the maintaining of coherence some, right? Like I said, 549 00:35:41,150 --> 00:35:44,720 extend it about 4x, But that's the best 550 00:35:44,720 --> 00:35:48,360 I can possibly get out of it. I'm not gonna get— 'cause now it's 551 00:35:48,360 --> 00:35:51,080 a hardware issue, right? I can only do 552 00:35:52,200 --> 00:35:56,000 so much. And this has been a problem for really a very, 553 00:35:56,000 --> 00:35:59,840 very, very long— since quantum started. And it 554 00:35:59,840 --> 00:36:01,320 hasn't, really hasn't improved a 555 00:36:03,560 --> 00:36:07,240 whole lot. So that's a 556 00:36:10,270 --> 00:36:14,030 big one. We still need a lot more work in the lab 557 00:36:14,030 --> 00:36:17,390 because until we can solve the coherence 558 00:36:18,430 --> 00:36:22,190 decoherence issue, scaling beyond about where we are 559 00:36:22,430 --> 00:36:26,110 now is going to be near impossible because there's just too 560 00:36:28,110 --> 00:36:31,470 much noise. What do you think it's going to take to solve 561 00:36:33,790 --> 00:36:37,520 that problem? 562 00:36:37,520 --> 00:36:41,320 Materials. Like new materials to be developed that the qubits, the quantum systems, 563 00:36:41,320 --> 00:36:44,720 are made out of? 564 00:36:45,600 --> 00:36:48,160 Absolutely. So I personally am 565 00:36:50,320 --> 00:36:53,360 convinced that that really is the 566 00:36:54,000 --> 00:36:57,520 major issue, is that part of the reason why we're having 567 00:36:57,520 --> 00:37:01,200 these coherence problems is that the materials 568 00:37:01,520 --> 00:37:05,200 aren't sufficient. And I can say I can spin 569 00:37:05,200 --> 00:37:08,880 out new materials once every 2 weeks, but number one, I can 570 00:37:08,880 --> 00:37:12,680 only sell so many. And getting these 571 00:37:13,240 --> 00:37:17,080 new materials through into the companies that are doing 572 00:37:18,360 --> 00:37:21,800 the research, IBM, Google, 573 00:37:22,280 --> 00:37:26,040 Continuum, Rigetti, QuEra, those guys, 574 00:37:26,040 --> 00:37:29,720 that can only happen so fast. Unless I'm physically part 575 00:37:29,720 --> 00:37:33,080 of their engineering teams, which of course I'm not. I've got my 576 00:37:33,720 --> 00:37:37,480 own company, right? So, um, you know, I would 577 00:37:37,480 --> 00:37:41,160 like to have a much more in-depth discussion with these 578 00:37:41,720 --> 00:37:45,080 guys about why, why their 579 00:37:45,080 --> 00:37:48,680 materials are causing the decoherence, even though I suspect they kind of 580 00:37:49,320 --> 00:37:52,560 know it, um, so that way those kinds of problems can be solved. But even 581 00:37:52,560 --> 00:37:56,410 once the new materials have been engineered, then they've gotta get actually 582 00:37:56,410 --> 00:38:00,210 into the QPU. Like, there's process 583 00:38:00,770 --> 00:38:04,570 that happens with this. So while from the outside, 584 00:38:04,570 --> 00:38:08,370 to go back to something, Candice, you had said before, 585 00:38:08,450 --> 00:38:11,090 it seems like things are 586 00:38:12,210 --> 00:38:14,210 moving fast, in a lot of ways, they still need to 587 00:38:16,530 --> 00:38:20,180 move faster. Because quantum, we 588 00:38:20,180 --> 00:38:22,940 need, AI right now is starting to bounce 589 00:38:23,340 --> 00:38:25,020 up against 590 00:38:26,940 --> 00:38:30,700 theoretical maximums. So because it's starting to bounce up against 591 00:38:32,380 --> 00:38:35,180 theoretical maximums, it's— this is why once we 592 00:38:35,260 --> 00:38:38,980 hit about GPT-3, you can almost 593 00:38:38,980 --> 00:38:42,500 draw a line there and you can see that the pace 594 00:38:42,500 --> 00:38:46,060 of AI improvement started slowing down and 595 00:38:46,420 --> 00:38:50,100 we started we stopped going from, it almost seemed 596 00:38:50,100 --> 00:38:53,060 like every few months there was 597 00:38:53,620 --> 00:38:57,060 this massive new improvements that we were getting out 598 00:38:57,380 --> 00:39:01,100 of AI. And lately all you're getting is 599 00:39:01,100 --> 00:39:04,020 incremental, very slow incremental 600 00:39:04,580 --> 00:39:08,380 improvements. Yeah, context windows are getting a little bit longer. 601 00:39:08,380 --> 00:39:12,060 Yeah, now it can remember past conversations a little bit better, 602 00:39:12,060 --> 00:39:15,500 or its ability to reason through code is slightly 603 00:39:16,050 --> 00:39:19,890 improved. But that's all we're getting. And we're getting 604 00:39:19,890 --> 00:39:23,410 that at the expense of massive 605 00:39:23,490 --> 00:39:27,170 new power requirements. Yeah, that's the big 606 00:39:27,170 --> 00:39:30,610 issue, isn't it? Yeah. Whereas if we 607 00:39:30,610 --> 00:39:34,170 can advance quantum a little bit faster, while 608 00:39:34,170 --> 00:39:37,650 quantum comes with a power requirement in the terms 609 00:39:38,050 --> 00:39:41,730 of cooling, the actual cost to run the QPU 610 00:39:41,810 --> 00:39:45,540 is almost zero. Right? It really doesn't cost 611 00:39:45,540 --> 00:39:49,060 a whole lot to run a QPU, right? 612 00:39:49,060 --> 00:39:52,620 Beyond the cooling, right? The cooling is, of course, is 613 00:39:52,780 --> 00:39:56,500 expensive, right? I'm not saying that the dilution fridges, you 614 00:39:56,740 --> 00:39:59,620 know, they're not cheap to buy them alone, 615 00:39:59,620 --> 00:40:02,940 you know, $500,000 per. 616 00:40:03,420 --> 00:40:05,940 And then the cost to run them is, you know, you got to keep it, 617 00:40:05,940 --> 00:40:09,730 you know, we're in dot Kelvin 618 00:40:09,890 --> 00:40:13,410 range, But assuming we can get beyond those 619 00:40:13,730 --> 00:40:17,570 with materials, we can find that we're 620 00:40:17,570 --> 00:40:21,370 able to push AI forward a 621 00:40:21,370 --> 00:40:24,850 lot because we really need to start running some of these 622 00:40:25,090 --> 00:40:28,770 neural networks, especially 623 00:40:29,170 --> 00:40:32,370 convolutional, on 624 00:40:33,740 --> 00:40:37,500 quantum. Right. Interesting. And for those that are not familiar with convolutional neural networks, they're 625 00:40:37,500 --> 00:40:41,340 a type of neural network architecture that's really good for 626 00:40:41,420 --> 00:40:43,980 image processing, typically. Images and video. 627 00:40:45,900 --> 00:40:49,579 Yeah. Yeah. Somebody had an experimental use case for 628 00:40:49,579 --> 00:40:53,220 them. I'm sorry, go ahead, Candace. No, I was wondering, has there 629 00:40:53,220 --> 00:40:56,820 been a breakthrough or a micro-innovation 630 00:40:56,820 --> 00:41:00,580 inside your lab that may not have made headlines 631 00:41:00,580 --> 00:41:02,460 but signals like a major shift in 632 00:41:05,980 --> 00:41:09,700 what's possible? Uh, well, there's been a few of them. Um, the fact that I 633 00:41:09,700 --> 00:41:12,460 can engineer new materials in 2 weeks, um, is a 634 00:41:13,500 --> 00:41:16,460 big one. Uh, that's, that's a really, really, really 635 00:41:17,820 --> 00:41:21,580 big one. Um, and, and like I said, we completely overran, um, 636 00:41:21,580 --> 00:41:24,860 the, you know, the ability of a sales team to even possibly 637 00:41:25,900 --> 00:41:29,580 keep up. Um, we have some, some of the new materials 638 00:41:31,000 --> 00:41:31,800 we have, um 639 00:41:38,760 --> 00:41:42,600 going to North American Stainless or US Steel, 640 00:41:42,600 --> 00:41:46,200 and then also to Ford. So some of the 641 00:41:46,360 --> 00:41:50,120 heat management, like heat shielding and all that kind of good stuff, Ford will 642 00:41:50,120 --> 00:41:53,840 be looking at shortly. But when you're talking about 643 00:41:53,840 --> 00:41:57,400 materials, when you create, engineer a material, especially when 644 00:41:57,400 --> 00:42:01,240 it's in silico, then they say, okay, well, now you got 645 00:42:01,240 --> 00:42:03,220 to create the thing. And then once you create the thing, then you have to 646 00:42:03,220 --> 00:42:05,340 test it, and then once you test it, then it has to get rolled into 647 00:42:05,340 --> 00:42:08,420 production and blah, blah, blah, right? Right. Yeah, 648 00:42:10,900 --> 00:42:14,580 but, but, um, so the materials is really, really a 649 00:42:14,820 --> 00:42:18,340 big one. Um, we haven't, um, 650 00:42:18,340 --> 00:42:22,100 uh, announced it broadly at all, 651 00:42:22,180 --> 00:42:25,980 uh, partially because it would be, it would be too 652 00:42:25,980 --> 00:42:29,620 easy to be too overwhelmed too fast. 653 00:42:29,620 --> 00:42:33,260 Um, so that's a big one. Um, the, the quantum error 654 00:42:33,260 --> 00:42:36,890 encryption would definitely be another one. We aren't talking— 655 00:42:36,890 --> 00:42:40,530 the only people we've talked to about our QEC so far 656 00:42:40,930 --> 00:42:44,610 is IBM. And that's because we benchmark it. We did our benchmarking 657 00:42:44,610 --> 00:42:48,330 on their systems. So, you know, they're a pretty easy one to have 658 00:42:48,330 --> 00:42:52,090 that discussion with. But, you know, IBM is a behemoth, right? 659 00:42:52,090 --> 00:42:55,730 It takes them— even IBM Quantum, it 660 00:42:56,690 --> 00:43:00,410 takes forever, you know, to get anywhere. But, you 661 00:43:00,770 --> 00:43:04,610 know, they're an obvious first place. And our quantum encryption algorithm 662 00:43:04,610 --> 00:43:08,390 is even agnostic. The, the— it doesn't matter whose hardware 663 00:43:08,390 --> 00:43:11,670 it is, it's going— the, the error correction is going to work no 664 00:43:13,190 --> 00:43:16,230 matter what, um, you know, whether it's D-Wave or Continuum 665 00:43:17,590 --> 00:43:20,790 or whoever. Um, so 666 00:43:21,270 --> 00:43:25,090 those would— yeah, the error correction would be a big one, and 667 00:43:25,090 --> 00:43:28,670 the, our— the materials, the— I, I say it's 668 00:43:29,110 --> 00:43:32,630 a quantum-enhanced pin, but that's only to simplify the discussion. 669 00:43:33,270 --> 00:43:36,910 There 4 different neural networks, and then what I call a 670 00:43:36,910 --> 00:43:40,750 controlling neural network that sits above it, and sort of in the middle, to think 671 00:43:40,750 --> 00:43:44,470 about it architecture-wise, in the middle is 672 00:43:44,790 --> 00:43:48,470 where the quantum enhancement sits, and the various neural 673 00:43:48,470 --> 00:43:52,230 networks sort of all talk to each other and talk 674 00:43:52,630 --> 00:43:55,750 to the quantum as a way of making this thing run 675 00:43:57,350 --> 00:43:58,070 so 676 00:44:02,240 --> 00:44:05,040 fast. Interesting. What— I know Candace usually asks 677 00:44:06,160 --> 00:44:09,840 this question. What's the biggest misconception out 678 00:44:09,840 --> 00:44:13,600 there about your business and kind of what you're up to? So 679 00:44:13,600 --> 00:44:16,960 there'll be— there, I guess I have to say there's two of them, and it 680 00:44:17,120 --> 00:44:20,400 depends on who you're talking to. Number one is in 681 00:44:20,720 --> 00:44:24,520 the business, too much of business outside of finance. This isn't so 682 00:44:24,520 --> 00:44:28,000 much true of— in the finance sector, they're starting to 683 00:44:28,410 --> 00:44:32,250 understand quantum. Because of QAOA, they understand, they're starting to understand. But outside 684 00:44:32,250 --> 00:44:36,010 of the financial vertical, there's still 685 00:44:36,010 --> 00:44:39,850 too much belief that quantum is a 686 00:44:39,850 --> 00:44:43,170 laboratory and research effort and there's no real 687 00:44:43,170 --> 00:44:46,570 true commercial applicability to it. And that's just 688 00:44:47,450 --> 00:44:51,210 completely false. That's probably the biggest 689 00:44:51,370 --> 00:44:53,530 one. And if we're talking 690 00:44:55,000 --> 00:44:57,640 about developers, I like to give this sort of as 691 00:44:58,680 --> 00:45:01,960 a hint. Most developers are 692 00:45:01,960 --> 00:45:04,520 using quantum computing almost like a 693 00:45:05,720 --> 00:45:09,560 trinary system, and I've had quantum developers disagree with me 694 00:45:09,799 --> 00:45:12,960 on this. And then I say, okay, let's look at your code, and I prove 695 00:45:12,960 --> 00:45:16,600 it to them. Most people 696 00:45:16,680 --> 00:45:20,280 are using quantum computers almost like a 697 00:45:20,360 --> 00:45:24,100 trinary computational device. You're gonna have to explain that. I roughly know what that 698 00:45:24,500 --> 00:45:28,100 is, but, uh, okay, so I like to be enlightened on terms of the 699 00:45:28,180 --> 00:45:32,020 difference between that because I've had this discussion and I didn't have a good 700 00:45:32,020 --> 00:45:35,460 answer to counter the statement. Yeah, okay, so, so 701 00:45:35,700 --> 00:45:39,340 right now, digital computers, classical computers, are binary, 702 00:45:39,340 --> 00:45:43,020 all right? It's a 1 or a 0 and that's it, 703 00:45:43,020 --> 00:45:46,820 okay? Most people are using, uh, 704 00:45:47,970 --> 00:45:51,730 quantum computers the same way. So think of 1 or 0 as spin up, 705 00:45:52,290 --> 00:45:55,010 spin down, right? And then you 706 00:45:56,050 --> 00:45:59,410 have superposition. Okay, so there— so trinary is spin up, 707 00:45:59,410 --> 00:46:00,940 spin down, superposition. Those are the 708 00:46:02,690 --> 00:46:06,290 3. Oh, okay. And they don't go any deeper with 709 00:46:06,290 --> 00:46:09,410 superposition. They stop there. 710 00:46:09,890 --> 00:46:12,770 Superposition actually means more than 711 00:46:14,480 --> 00:46:17,920 just both. Which is what superposition sort of means, but 712 00:46:18,000 --> 00:46:21,600 it doesn't just mean only 713 00:46:21,600 --> 00:46:23,360 that. Um, 714 00:46:25,680 --> 00:46:28,280 so it's— how do I put this in a way that I don't give away 715 00:46:28,280 --> 00:46:32,080 too much of my own secret sauce? Um, 716 00:46:32,080 --> 00:46:34,960 so you could treat— you could treat it that way. You could treat like a 717 00:46:34,960 --> 00:46:38,400 trinary system and you wouldn't be wrong, but you're not taking advantage 718 00:46:38,720 --> 00:46:42,470 of the superposition. Advantage, yes. Right. So for most normal 719 00:46:42,470 --> 00:46:46,190 people, I think a good way to look at this as, 720 00:46:46,190 --> 00:46:49,790 um, like a checkbox on a, on an online form, right? It's 721 00:46:49,790 --> 00:46:52,110 either checked, unchecked, 722 00:46:52,990 --> 00:46:56,749 or some designers will have a third space means you never touched 723 00:46:56,749 --> 00:47:00,190 it, right? So it's either kind of like yes, no, or unknown would be another 724 00:47:00,190 --> 00:47:04,030 one, right? Right. So you could almost use it like a maybe. So where the— 725 00:47:04,430 --> 00:47:07,650 what most developers are effectively doing it is using it like 726 00:47:08,200 --> 00:47:11,800 a maybe. Yes, no, maybe, on, off, 727 00:47:11,960 --> 00:47:15,720 don't know, right? But strictly speaking, superposition 728 00:47:15,720 --> 00:47:19,360 is yes and no at the same time. Correct. 729 00:47:19,360 --> 00:47:23,160 And, and it's not that it's yes, so yes, it's yes and no at the 730 00:47:23,160 --> 00:47:26,760 same time. But so let's, let's— 731 00:47:27,080 --> 00:47:29,960 I don't know, this might be getting a little bit deep into the woods, but 732 00:47:29,960 --> 00:47:32,840 let's look at Schrödinger's cat for a minute, 733 00:47:35,370 --> 00:47:39,090 right? Okay. The— in the thought experiment, it's the 734 00:47:39,090 --> 00:47:42,850 cat can be thought of as alive and dead at the same 735 00:47:42,850 --> 00:47:46,410 time. But here's the truth: the cat could also be 736 00:47:46,410 --> 00:47:50,250 thought of as in the process of 737 00:47:50,730 --> 00:47:54,250 dying. Okay, so it's not just alive and dead at the 738 00:47:54,410 --> 00:47:58,050 same time, it's alive, dead, and in the process of dying. And if it's in 739 00:47:58,050 --> 00:48:01,850 the process of dying, how far along the process 740 00:48:01,850 --> 00:48:02,330 of dying 741 00:48:05,370 --> 00:48:08,250 is it? Oh, okay, 742 00:48:09,450 --> 00:48:13,210 okay. So, so there's— there is a saying from, 743 00:48:13,210 --> 00:48:16,850 um, a personal development guy that I really like a lot. 744 00:48:16,850 --> 00:48:20,690 Most people in the personal development space, the minute they say 745 00:48:20,690 --> 00:48:24,490 quantum physics says, the next words that come out of their mouth 746 00:48:24,890 --> 00:48:28,690 are nonsense. Okay, nonsense. Uh, 747 00:48:28,690 --> 00:48:32,450 Dr. Joe Dispenza does a really good job when he says quantum 748 00:48:32,450 --> 00:48:36,150 physics says, the next words that come out of his mouth mouth are probably going 749 00:48:36,150 --> 00:48:39,950 to be right. When they are wrong, 750 00:48:39,950 --> 00:48:43,430 it's usually he's in the early part of 751 00:48:43,510 --> 00:48:46,870 his discussion, he's trying to get you to understand something, and if you listen to 752 00:48:46,870 --> 00:48:50,150 him a little bit longer, he makes it correct. 753 00:48:50,390 --> 00:48:53,470 So superposition is not just both. A better way 754 00:48:53,470 --> 00:48:57,190 of describing superposition would be it is a 755 00:48:57,190 --> 00:49:00,960 definition of all possible 756 00:49:01,280 --> 00:49:04,880 possibilities. Gotcha. Okay, that's what superposition actually is. So if 757 00:49:04,880 --> 00:49:08,720 it's all possible possibilities, that opens up more 758 00:49:08,720 --> 00:49:12,160 than just trinary computation. And that is about 759 00:49:12,320 --> 00:49:16,159 as far deep into the woods on that subject as I will 760 00:49:16,160 --> 00:49:19,680 get, because that's fair. We're at the top, we're towards the top of the hour, 761 00:49:19,680 --> 00:49:21,760 so like, it's, it's probably— we'd love to have you 762 00:49:24,000 --> 00:49:27,820 back because, uh, um, yeah, no, I, I, I feel you, like There's 763 00:49:27,820 --> 00:49:30,820 a lot to unpack there. I can kind of sense like, 764 00:49:31,700 --> 00:49:34,380 oh, this— it's kind of like you pull a thread on a sweater, like, oh, 765 00:49:34,380 --> 00:49:37,900 it's not just this little bit. It's actually way more 766 00:49:37,900 --> 00:49:41,620 than I anticipated. Yeah. And then you start getting into the 767 00:49:41,620 --> 00:49:45,380 woods of what is superposition exactly and whether it's, 768 00:49:45,380 --> 00:49:48,940 you know, we're talking about when you're measuring spin up, 769 00:49:48,940 --> 00:49:52,270 spin down, it's because you have collapsed the particle 770 00:49:52,900 --> 00:49:56,580 into a into a particle, and superposition is actually just 771 00:49:57,780 --> 00:50:00,820 the waveform, right? And that's right. And then you start getting the double slit and 772 00:50:00,820 --> 00:50:04,580 all those kinds of fun things. Yeah, 773 00:50:04,580 --> 00:50:08,300 causality, you start— yeah, yeah, yeah. So anyway, I can, 774 00:50:08,579 --> 00:50:12,140 I can geek out on this for kind of— we'd love to have another— you 775 00:50:12,140 --> 00:50:15,940 back on the show. Yeah, absolutely. This has been a fantastic 776 00:50:15,940 --> 00:50:19,630 conversation. We really, really appreciate your time. I thank you. I've had a lot 777 00:50:42,350 --> 00:50:46,080 of fun. They're skanking, skanking in time. 778 00:50:46,080 --> 00:50:49,680 Black holes are wailing in a horn line so fine. From plank scales to 779 00:50:49,680 --> 00:50:53,280 planets, they're connecting the dots. Candace and Frank, they're 780 00:50:53,280 --> 00:50:54,360 the 781 00:50:58,920 --> 00:51:02,560 cosmic hotshot. Quantum Podcast, turn it up fast. Candace 782 00:51:02,560 --> 00:51:06,320 and Frank blowing my mind at last. Quantum Podcast, 783 00:51:06,320 --> 00:51:10,050 they're breaking the mold. Science has got beats. 784 00:51:10,290 --> 00:51:11,350 It's bold and it's gold.