1 00:00:24,600 --> 00:00:26,839 Speaker 1: I'm with Mitch Randall and welcome bitch. 2 00:00:27,120 --> 00:00:29,519 Speaker 2: Nice to see you, to see you, thanks for having. 3 00:00:29,280 --> 00:00:33,640 Speaker 1: Yeah, sure you're a speaker here, and you spoke yesterday. 4 00:00:33,679 --> 00:00:38,159 I missed your talk. But I'm very interested in this subject, 5 00:00:38,840 --> 00:00:42,240 and that's basically AI. You know, I just think that 6 00:00:42,359 --> 00:00:46,359 it's something of wonder and something a little bit scary 7 00:00:46,600 --> 00:00:51,079 both at the same time. And so if you could 8 00:00:51,079 --> 00:00:53,880 a little bit about your background first, and we'll cut 9 00:00:53,920 --> 00:00:56,920 you into AI and how long ago. 10 00:00:58,359 --> 00:01:02,960 Speaker 3: Oh, I have a pretty diverse engineering background, So I 11 00:01:02,960 --> 00:01:07,519 did radars with the National Center for Atmosphere Research from 12 00:01:07,560 --> 00:01:11,480 the nineties and then in the two thousands my own 13 00:01:11,480 --> 00:01:15,079 company with wireless power. So am very technical, you know, 14 00:01:15,680 --> 00:01:20,840 hardcore engineer, physicist kind of thing. But around twenty sixteen 15 00:01:20,840 --> 00:01:23,519 I got really interested in AI, and if you know 16 00:01:23,719 --> 00:01:27,480 what's happened there around there, they came up with convolutional 17 00:01:27,560 --> 00:01:31,000 neural nets and that's what really AI was kind of 18 00:01:31,000 --> 00:01:34,079 a toy before that, really, because it couldn't quite recognize 19 00:01:34,120 --> 00:01:35,840 what was in images, you know, the big thing was 20 00:01:35,920 --> 00:01:40,079 image classification at the time, and then it just skyrocketed. 21 00:01:40,359 --> 00:01:41,079 Speaker 2: They figured out. 22 00:01:41,000 --> 00:01:45,599 Speaker 3: The convolutional networks and then things started to roll much 23 00:01:45,640 --> 00:01:49,599 more rapidly, and I got interested. Then I can tell 24 00:01:49,640 --> 00:01:52,719 you the exact thing that really tweeked me is what 25 00:01:53,079 --> 00:01:57,719 is when Nvidia published a paper where they had just 26 00:01:57,760 --> 00:02:01,159 the pixels coming into the AI and also the steering 27 00:02:01,159 --> 00:02:06,319 wheel position, and they drove around town to train, so 28 00:02:06,439 --> 00:02:09,120 it just had the steering wheel position and pixels from 29 00:02:09,120 --> 00:02:12,159 the looking forward and then it would sort of drive. 30 00:02:12,960 --> 00:02:15,759 And that blew my mind because nothing like that was 31 00:02:16,159 --> 00:02:18,120 going on before that. I mean I couldn't believe that 32 00:02:18,199 --> 00:02:22,719 there was no program involved, nothing, you know, just paels 33 00:02:22,759 --> 00:02:23,879 coming in and then it would. 34 00:02:23,680 --> 00:02:29,080 Speaker 1: Tell turn was wow, Well, I just wrote an autonomous 35 00:02:29,360 --> 00:02:33,240 self driving taxi and San Francisco a few weeks ago, 36 00:02:33,360 --> 00:02:36,439 and that was quite amazing. I felt safe, actually felt safe, 37 00:02:36,560 --> 00:02:38,960 because that's like the first thing that people have said 38 00:02:38,960 --> 00:02:41,319 to me, well, wasn't it scary? And I said, actually no, 39 00:02:41,439 --> 00:02:45,120 I felt totally secure in it. But so anyway, that, yeah, 40 00:02:45,159 --> 00:02:49,520 it's fascinating and it seems like it's snowballing. I mean 41 00:02:49,599 --> 00:02:52,520 I watch a lot of you know, documentaries and just 42 00:02:52,599 --> 00:02:55,039 little clips here and there about it, do some reading 43 00:02:55,080 --> 00:02:58,719 about it, and it just seems like it's just growing 44 00:02:58,759 --> 00:03:01,400 by leaps and bounds and you know when you search 45 00:03:01,400 --> 00:03:04,680 on Google now it automatically comes up with AI answering 46 00:03:04,719 --> 00:03:08,400 your your whatever your question is. So I just got 47 00:03:08,400 --> 00:03:11,159 to tell you this quick story because before we started recording, 48 00:03:11,159 --> 00:03:13,439 I told you that I'm in the fine art and antique. 49 00:03:14,120 --> 00:03:16,960 So I was in a pretty sizable state in the 50 00:03:17,039 --> 00:03:20,360 coast of Massachusetts and there was a I could tell 51 00:03:20,400 --> 00:03:23,280 it was a really good painting, could not read the signature. 52 00:03:23,840 --> 00:03:26,520 So the signature was just you know, scribble. So I'm 53 00:03:26,560 --> 00:03:29,560 working on it, and I told the people, you know, 54 00:03:29,599 --> 00:03:31,120 it's going to take some extra time in this, but 55 00:03:31,159 --> 00:03:33,120 I can tell us a really good painting ends up. 56 00:03:33,120 --> 00:03:35,879 It was, but it took me about oh I don't know, 57 00:03:35,960 --> 00:03:39,400 maybe five or six hours of straight time trying to 58 00:03:39,759 --> 00:03:43,000 just and I finally figured it out with all cross referencing, 59 00:03:43,120 --> 00:03:46,240 and then I identified a signature on another painting. It 60 00:03:46,360 --> 00:03:50,319 was exactly the same. You can't read it. So I 61 00:03:50,319 --> 00:03:52,599 got I got done, and I got the price. It 62 00:03:52,680 --> 00:03:55,960 was a twenty thousand dollars painting not break. Oh yes, 63 00:03:56,479 --> 00:04:00,319 So my son is really using chat GBT as like 64 00:04:00,319 --> 00:04:04,479 a he subscribes. You know, it's a level up. So 65 00:04:04,599 --> 00:04:07,479 I said, oh, just out of curiosity. Here's a picture. 66 00:04:07,639 --> 00:04:11,759 Can you just throw that in there? Instantly? It had 67 00:04:11,800 --> 00:04:15,680 the artist and it said it's most likely this artist 68 00:04:15,719 --> 00:04:19,199 and that's exactly theirs. It was Wow, and so that 69 00:04:19,319 --> 00:04:23,160 to me is incredible. I mean that it can change 70 00:04:23,680 --> 00:04:26,399 what I do. And so as far as AI, one 71 00:04:26,439 --> 00:04:28,879 of the first things I think of, what is your 72 00:04:28,879 --> 00:04:31,399 opinion of what type of jobs people are going to 73 00:04:31,480 --> 00:04:32,240 start losing? 74 00:04:34,279 --> 00:04:36,399 Speaker 2: I mean, it's such a big subject to it. 75 00:04:36,439 --> 00:04:39,120 Speaker 3: There's so much going on you'd pick a little part 76 00:04:39,199 --> 00:04:42,000 and talk about it. I mean, what's you know, where 77 00:04:42,000 --> 00:04:45,040 we're headed for next? Agents? So right now we're having 78 00:04:45,040 --> 00:04:48,439 these large language models right where you put in text 79 00:04:48,720 --> 00:04:51,680 gives your sinulus response, right, and when you're not putting 80 00:04:51,720 --> 00:04:56,079 in stimulus, it's basically asleep, not doing anything. But agents 81 00:04:56,079 --> 00:04:59,079 are different. They're a loop where it's always trying to 82 00:04:59,120 --> 00:05:02,120 meet some kind of part or you know, achieve some 83 00:05:02,199 --> 00:05:05,160 kind of goal, and that's where it starts to that's 84 00:05:05,160 --> 00:05:06,319 where things really pick up. 85 00:05:06,759 --> 00:05:08,319 Speaker 1: So I mean, right now, of. 86 00:05:08,199 --> 00:05:11,800 Speaker 3: Course anything that's well, you know, moving back up to 87 00:05:11,920 --> 00:05:14,439 for a second, because they have found and there's curves 88 00:05:14,439 --> 00:05:18,160 to show that really clearly, the more data you know, 89 00:05:18,279 --> 00:05:21,800 the more data you put into the training, the smarter 90 00:05:21,920 --> 00:05:22,360 it gets. 91 00:05:22,839 --> 00:05:25,519 Speaker 2: And right now they've figured out that scale is everything 92 00:05:26,480 --> 00:05:28,959 for one thing. I mean, it's not really going to 93 00:05:29,000 --> 00:05:31,000 be everything. There's going to be innovations. 94 00:05:31,000 --> 00:05:32,560 Speaker 3: I'm sure they're going to make it so the scale 95 00:05:32,560 --> 00:05:34,399 doesn't have to be that crazy because I didn't have 96 00:05:34,439 --> 00:05:37,759 to read twelve trillion tokens, you know, be the smartest 97 00:05:37,759 --> 00:05:40,040 and I don't them, right, so we know that it's possible. 98 00:05:40,439 --> 00:05:42,480 But all they're doing is just turning up the crank. 99 00:05:42,600 --> 00:05:45,319 You know, Hey, if we put in more, it's smarter, 100 00:05:45,439 --> 00:05:48,120 you know. And the act set of curve that shows here, 101 00:05:48,199 --> 00:05:51,480 you know, here's the intelligence of the of each one 102 00:05:51,480 --> 00:05:53,839 of the models compared to the data coming in, and 103 00:05:53,920 --> 00:05:55,959 they can show it's a straight line in. 104 00:05:55,920 --> 00:05:59,199 Speaker 2: The log log curve. So that so then of course 105 00:05:59,240 --> 00:05:59,800 what do they do. 106 00:06:00,160 --> 00:06:02,360 Speaker 3: They turned up the grank they put in Now we're 107 00:06:02,399 --> 00:06:04,439 running out of data to put into AI. 108 00:06:05,079 --> 00:06:08,199 Speaker 1: I've heard that they've used the scraping even YouTube videos 109 00:06:08,240 --> 00:06:08,959 and stuff. 110 00:06:08,959 --> 00:06:12,439 Speaker 3: Right you know, but there you know they're talking about 111 00:06:12,600 --> 00:06:13,480 this is the end of AI. 112 00:06:13,680 --> 00:06:16,160 Speaker 2: We've hit the limit of intelligence or something like that. 113 00:06:16,240 --> 00:06:19,800 You know, it's just just means that that that easy trick, 114 00:06:19,879 --> 00:06:23,800 but just scaling that knob isn't working anymore, and they'll 115 00:06:23,800 --> 00:06:26,279 come up with somebody else. I mean, there's more than 116 00:06:26,319 --> 00:06:30,639 five thousand papers a month being read this Yeah, yeah, 117 00:06:30,879 --> 00:06:36,240 Actually six six years ago I looked specifically. 118 00:06:35,639 --> 00:06:37,519 Speaker 3: There was five hundred papers a month, and I was like, 119 00:06:37,639 --> 00:06:40,639 oh my god, I've never been in a field that 120 00:06:40,759 --> 00:06:43,319 has that kind of activity. You can't really keep track 121 00:06:43,360 --> 00:06:44,600 of what the new thing is. 122 00:06:45,079 --> 00:06:46,360 Speaker 1: And then just a few. 123 00:06:46,199 --> 00:06:48,240 Speaker 2: Years later it's five thousand a month. 124 00:06:48,639 --> 00:06:49,199 Speaker 1: Hung real. 125 00:06:49,319 --> 00:06:51,160 Speaker 2: In fact, it might be more than that, because that 126 00:06:51,639 --> 00:06:53,160 looked couple years supposed. 127 00:06:52,800 --> 00:06:57,199 Speaker 1: To get that number has chat chpt exactly. Yeah, wow, 128 00:06:57,519 --> 00:06:58,720 but I didn't answer your question. 129 00:06:58,839 --> 00:07:03,040 Speaker 2: Yes, I mean obviously now with Ellen Limbs, anything. 130 00:07:02,680 --> 00:07:06,160 Speaker 3: That's text for your example of searching, I mean is 131 00:07:06,319 --> 00:07:08,360 it's literally read every medical journal. 132 00:07:08,439 --> 00:07:10,000 Speaker 1: Probably it's read every. 133 00:07:10,399 --> 00:07:13,079 Speaker 3: I mean anything that you that's you know, you give 134 00:07:13,079 --> 00:07:15,959 it a stimulus and it gets a response that those 135 00:07:16,040 --> 00:07:16,800 doctor are threatened. 136 00:07:16,879 --> 00:07:17,279 Speaker 2: Of course. 137 00:07:17,839 --> 00:07:21,439 Speaker 4: Yeah, I'm gonna there's so much to say, but I'm 138 00:07:21,439 --> 00:07:24,240 gonna segway this per second. They just did a study. 139 00:07:24,240 --> 00:07:27,839 I read it just a couple of weeks ago. They 140 00:07:27,879 --> 00:07:30,879 wanted to see if AI would help doctors docmos. 141 00:07:31,439 --> 00:07:35,720 Speaker 3: So they took a set of doctors that didn't use AI, 142 00:07:35,800 --> 00:07:39,360 took set of doctors that did use AI, and the 143 00:07:39,399 --> 00:07:40,600 doctors that didn't use. 144 00:07:40,480 --> 00:07:43,519 Speaker 2: AII their diagnosis. Great. It was about seventy two percent 145 00:07:43,560 --> 00:07:46,839 in the study. So that's how accurate they were. The 146 00:07:46,879 --> 00:07:51,759 doctors that did use AI in seventy three percent. That's 147 00:07:51,759 --> 00:07:54,759 come on. But the actual punchline in the article is 148 00:07:55,120 --> 00:07:57,120 AI alone was ninety four percent. 149 00:07:57,959 --> 00:08:01,800 Speaker 3: It was like, well the doctors and doctors aren't listening 150 00:08:02,399 --> 00:08:03,519 because they'd just listened to. 151 00:08:03,519 --> 00:08:06,079 Speaker 1: Them, they would get nine four percent, right. Well, just 152 00:08:06,160 --> 00:08:08,759 the fact that that's true is unbelievable. 153 00:08:08,800 --> 00:08:09,000 Speaker 2: You know. 154 00:08:09,399 --> 00:08:13,480 Speaker 1: That's what I'm excited mostly about the medical aspect of this, 155 00:08:14,160 --> 00:08:16,560 you know, I mean, I think, you know, we could 156 00:08:16,600 --> 00:08:19,920 possibly have the cure for cancer insight at some point, 157 00:08:20,360 --> 00:08:23,360 and you know, on many and many other other things. 158 00:08:23,360 --> 00:08:25,079 Really exciting. So I just want to tell you something 159 00:08:25,160 --> 00:08:27,879 kind of funny too, going back to Chat GPT. I 160 00:08:28,000 --> 00:08:30,439 was on a road trip with my son for well 161 00:08:30,879 --> 00:08:33,879 about a week or so, and so he was on 162 00:08:34,159 --> 00:08:37,879 chat GPT and he says, oh, so he said, let's 163 00:08:37,879 --> 00:08:40,759 see what it knows about your show. So he went 164 00:08:40,799 --> 00:08:43,440 in and the thing. It did a wonderful job all 165 00:08:43,480 --> 00:08:46,440 about my show, and it's Martin Willis, host has six 166 00:08:46,519 --> 00:08:49,600 hundred plus you know, episodes and blah blah blah. It 167 00:08:49,720 --> 00:08:53,240 was very nice. And so then he goes, well, what 168 00:08:53,320 --> 00:08:55,879 do we know about the host? And he said, not 169 00:08:55,960 --> 00:08:58,240 much is known about Martin Willis. And so then he 170 00:08:58,320 --> 00:09:02,080 said to it, well why don't you how'd you say it? 171 00:09:02,320 --> 00:09:04,279 Why don't you imagine what the house would be like? 172 00:09:04,480 --> 00:09:07,600 And it was hilarious. It was saying that all these 173 00:09:07,919 --> 00:09:10,879 it says that I'm also into antiques and I maybe 174 00:09:10,960 --> 00:09:13,320 outrunning the state sales. Well I'm looking up in the 175 00:09:13,360 --> 00:09:15,559 sky and maybe sit in the back porch with a 176 00:09:15,600 --> 00:09:17,840 scotch and look up in the sky, which I don't drink. 177 00:09:17,840 --> 00:09:20,200 But anyway, I thought it was hilarious what it came 178 00:09:20,279 --> 00:09:22,799 up with. So I love you know, it has like 179 00:09:22,840 --> 00:09:26,279 an imagination, or it can have an imagination, which is 180 00:09:26,399 --> 00:09:29,840 really it's uh, you know, when is it going to 181 00:09:29,840 --> 00:09:33,759 be that? I mean, people are I've heard people say, well, 182 00:09:33,799 --> 00:09:35,720 it's going to be another twenty years before it'll think 183 00:09:35,919 --> 00:09:40,200 like the human brain. But the way it's going, it's 184 00:09:40,240 --> 00:09:40,679 going to be. 185 00:09:40,919 --> 00:09:44,120 Speaker 2: The essential principle on what it's called the singularity. 186 00:09:44,240 --> 00:09:48,919 Speaker 3: If you're not the singularity is it's when AI becomes 187 00:09:48,960 --> 00:09:51,879 smart and that everybody else. The reason they call it 188 00:09:51,879 --> 00:09:55,720 the singularity is for math, there's a function that's you know, 189 00:09:56,480 --> 00:10:00,639 one over acts as it grows the singularity of that function, 190 00:10:00,720 --> 00:10:04,759 it goes to infinity and essentially what there's the reason 191 00:10:04,799 --> 00:10:08,919 it does that is because not only are you you know, 192 00:10:08,960 --> 00:10:11,919 getting better at designing AI, because that's what everybody you know, 193 00:10:12,000 --> 00:10:13,080 the ride's figuring it out. 194 00:10:14,360 --> 00:10:16,200 Speaker 2: There's other factors. I'll just I'll throw that in a 195 00:10:16,360 --> 00:10:16,799 quick too. 196 00:10:18,360 --> 00:10:20,039 Speaker 3: You know, if you had a if you had a 197 00:10:20,159 --> 00:10:24,840 robot an AI system that in place of worker, you know, 198 00:10:25,200 --> 00:10:28,600 you money that's worth even well, let's say this. Let's 199 00:10:28,600 --> 00:10:30,320 say that there was a box that cost a million 200 00:10:30,360 --> 00:10:32,840 dollars that you can put in the driver's. 201 00:10:32,480 --> 00:10:36,320 Speaker 2: Seat of the truck. That's a If you work out 202 00:10:36,360 --> 00:10:38,440 the numbers, you're way ahead with that. 203 00:10:38,519 --> 00:10:40,440 Speaker 1: Not even truck. I can go twenty four to seven. 204 00:10:41,080 --> 00:10:42,440 Speaker 2: So that's three shifts, right. 205 00:10:42,639 --> 00:10:46,159 Speaker 3: Yeah, you don't have to manage any any workers, any 206 00:10:46,159 --> 00:10:49,440 worker whose comm no health insurance, no health insurance, you know, 207 00:10:49,600 --> 00:10:52,720 no HR department, that million dollars. 208 00:10:52,480 --> 00:10:56,159 Speaker 1: No taxes, brilliant no taxes because there's no it's all 209 00:10:56,279 --> 00:10:59,759 tax right off. Yeah, so that. 210 00:11:00,120 --> 00:11:03,639 Speaker 2: Oft comes incredibly quick and everyone knows that. 211 00:11:03,679 --> 00:11:06,320 Speaker 3: I mean, not only is AI you're going to and 212 00:11:06,360 --> 00:11:09,440 do a number of jobs with you're asking up. 213 00:11:10,200 --> 00:11:14,360 Speaker 2: But you know, it's also a way to control the world. 214 00:11:14,559 --> 00:11:17,000 Speaker 3: I mean, it's like the ultimate super weapon, right you 215 00:11:17,080 --> 00:11:21,080 can you can literally have an AI agent on every 216 00:11:21,080 --> 00:11:25,000 single person, following them, seeing exactly what they do and looking, 217 00:11:25,039 --> 00:11:28,840 you know, figuring out everything about you, like focused on you, 218 00:11:29,440 --> 00:11:30,720 the super intelligent vocals. 219 00:11:30,759 --> 00:11:35,919 Speaker 2: Anyway, The point is the economic incentive to achieve this 220 00:11:36,840 --> 00:11:38,519 is unbelievably high. 221 00:11:38,840 --> 00:11:43,000 Speaker 3: Yeah, so that's why that's another factor. It's making you 222 00:11:43,039 --> 00:11:45,080 of that curve. But if people realize wait a minute, 223 00:11:45,440 --> 00:11:47,879 this can actually start be placing up. That can drive 224 00:11:48,519 --> 00:11:51,120 you know, all these things they're finding out, that economic 225 00:11:51,159 --> 00:11:53,519 concentive just keeps growing and growing and growing. 226 00:11:53,879 --> 00:11:55,799 Speaker 1: But then beyond that, you. 227 00:11:55,759 --> 00:11:57,840 Speaker 2: Can take the AIS that you have and you can 228 00:11:57,960 --> 00:11:59,200 just explain how good they are. 229 00:12:00,240 --> 00:12:01,759 Speaker 1: Every single scientific. 230 00:12:01,279 --> 00:12:04,360 Speaker 3: Paper, yeah right, you know they can actually read those 231 00:12:04,399 --> 00:12:06,559 five thousand papers a month that are coming out when 232 00:12:06,559 --> 00:12:07,600 you never could, you know. 233 00:12:08,159 --> 00:12:11,679 Speaker 1: So they're using AI to design AI. Yeah, and you know, 234 00:12:12,120 --> 00:12:13,559 here's an example that. 235 00:12:13,759 --> 00:12:15,879 Speaker 2: I'll get all the numbers wrong, but it doesn't. 236 00:12:15,600 --> 00:12:19,519 Speaker 3: Matter because it's you know, I wouldn't tech ack knowledge 237 00:12:19,519 --> 00:12:21,600 that for a long time to see you know, n video. 238 00:12:21,919 --> 00:12:24,039 They've been around for a long time, and you know, 239 00:12:24,200 --> 00:12:28,240 like every year it's twenty percent better some you know, 240 00:12:28,440 --> 00:12:33,200 comprehensible amount better. Right, they're not using aonic to design 241 00:12:33,279 --> 00:12:35,399 their stuff. It went up like five times or ten. 242 00:12:35,480 --> 00:12:36,600 Speaker 2: I kind of remember the number. 243 00:12:36,600 --> 00:12:39,559 Speaker 3: I just remember my job dropping because these kind of 244 00:12:39,679 --> 00:12:43,399 improvements aren't something that we're used to do in these industries. 245 00:12:43,480 --> 00:12:48,360 It's coming from AI. But optimizing, you know, whatever, whatever 246 00:12:48,360 --> 00:12:49,799 it does, is just better at it, you know. 247 00:12:50,159 --> 00:12:53,039 Speaker 1: Right, So now I'm trying to remember if he was 248 00:12:53,519 --> 00:12:57,039 he was a child prodigy, like a chess player in 249 00:12:57,120 --> 00:13:02,000 England and he started eight. He didn't call it AGI 250 00:13:02,200 --> 00:13:07,759 maybe as artificial general, yeah, and he was trying to 251 00:13:07,799 --> 00:13:12,519 figure out some type of genome work, which but I 252 00:13:12,519 --> 00:13:15,639 mean this was going back like in the beginning. You know, 253 00:13:16,480 --> 00:13:21,159 he was really on this stuff. Brilliant that's what it is. 254 00:13:21,519 --> 00:13:24,200 That's what it is, You're right, and they couldn't quite 255 00:13:24,240 --> 00:13:26,320 get it. But anything when it comes to medical I 256 00:13:26,399 --> 00:13:29,000 understand takes a lot of there's a lot of failure 257 00:13:29,039 --> 00:13:31,159 when it comes to medicals, so it takes a lot. 258 00:13:31,240 --> 00:13:33,879 But I'm excited as far as that goes. But I'm 259 00:13:33,879 --> 00:13:39,399 also scared because you know, I just you know, there 260 00:13:39,480 --> 00:13:42,600 was I remember when I first heard about it. You're talking, 261 00:13:42,840 --> 00:13:46,720 like you said, twenty fifteen somewhere back there. My son's 262 00:13:46,759 --> 00:13:49,320 really into all this type of these type of things, 263 00:13:49,360 --> 00:13:51,840 and he said to me, you know, it's possible that 264 00:13:52,240 --> 00:13:55,919 there could be like a self replicating machine and that 265 00:13:55,960 --> 00:13:58,399 could be told to take over the world, you know, 266 00:13:58,480 --> 00:14:01,360 and just do it, you know. I mean, that's what 267 00:14:01,440 --> 00:14:04,080 it could come down to. And there's unfortunately, there's always 268 00:14:04,080 --> 00:14:06,360 a bad seat out there, just like there is with 269 00:14:06,519 --> 00:14:10,759 weapons that could do something you know, terrible and for 270 00:14:11,159 --> 00:14:13,200 you know, I mean, it's a it's a bottleneck of 271 00:14:13,279 --> 00:14:16,679 technology that we have to get through. 272 00:14:16,919 --> 00:14:19,720 Speaker 3: Don't I'm a little bit more on the business actually, 273 00:14:20,279 --> 00:14:23,360 the bad seat. Yeah, I feel that all that economic 274 00:14:23,440 --> 00:14:28,679 incented just is a incubator for yeah, I mean. 275 00:14:29,399 --> 00:14:32,720 Speaker 5: For a disease of of that, you know, And I 276 00:14:32,759 --> 00:14:33,159 think that. 277 00:14:35,279 --> 00:14:39,559 Speaker 3: You know, for example, but I a bunch of people 278 00:14:39,600 --> 00:14:42,960 from the safety department left because they said this is 279 00:14:43,200 --> 00:14:44,039 this is bs. 280 00:14:44,120 --> 00:14:46,440 Speaker 2: They're not paying attention to safety like they should be. 281 00:14:46,960 --> 00:14:50,120 Speaker 1: And then by the way, no guardrails serve so to speak. 282 00:14:50,279 --> 00:14:53,080 Speaker 2: Right, that was their job and they see their job 283 00:14:53,200 --> 00:14:53,960 is being moored. 284 00:14:54,399 --> 00:14:57,039 Speaker 3: They little bit, I'm out of here. But then they 285 00:14:57,559 --> 00:15:00,279 I was gonna say it is a started a new. 286 00:15:00,279 --> 00:15:03,720 Speaker 2: Company and see, unbelievable. 287 00:15:04,279 --> 00:15:06,360 Speaker 3: I probably have this wrong, but if I remember, I 288 00:15:06,440 --> 00:15:08,840 was thirty billion dollars be raised. 289 00:15:08,519 --> 00:15:10,759 Speaker 1: To start a start up. Oh my, I mean it 290 00:15:10,919 --> 00:15:15,039 used to be thirty thousand. I'm right, Yeah, it's unheard of. 291 00:15:15,360 --> 00:15:20,159 Speaker 3: Yeah again, because the economic incentive is off the charts. 292 00:15:20,000 --> 00:15:21,279 Speaker 2: It's a super weapon. 293 00:15:21,399 --> 00:15:25,080 Speaker 3: It's way more deadly than an atomic bombed ever being. 294 00:15:26,039 --> 00:15:29,960 Speaker 1: It certainly could be. And also it could be basically 295 00:15:30,039 --> 00:15:32,639 in a destroyer of economy in a way, because if 296 00:15:32,639 --> 00:15:34,879 you take away all the jobs, who's going to be 297 00:15:35,039 --> 00:15:36,840 you know, there's going to be the rich getting richer 298 00:15:36,879 --> 00:15:40,120 and a very small percentage of people making all the 299 00:15:40,159 --> 00:15:43,159 money and they're going to have to rely on someone 300 00:15:43,240 --> 00:15:45,519 for other things. Yeah, I mean it could be a 301 00:15:45,600 --> 00:15:46,600 destroyer in that way. 302 00:15:47,679 --> 00:15:50,200 Speaker 2: I mean, I intend it's easy for me to kind 303 00:15:50,240 --> 00:15:52,759 of turn you turn. 304 00:15:52,600 --> 00:15:54,919 Speaker 3: Off the time scales on and just see how things 305 00:15:54,960 --> 00:15:58,039 are going to pan out, you know, and. 306 00:15:57,440 --> 00:16:01,399 Speaker 1: It's really clear that all these well, let. 307 00:16:01,279 --> 00:16:03,080 Speaker 3: Me just back up a second. It used to be 308 00:16:03,200 --> 00:16:05,519 people say robots are going to take over, you know, 309 00:16:05,559 --> 00:16:09,960 like further flippers. But back in even twenty eighteen, they 310 00:16:09,960 --> 00:16:14,279 were saying AI is better at reading X rays than 311 00:16:14,639 --> 00:16:17,840 radiologists down on the top of the chart for doctors, 312 00:16:18,000 --> 00:16:20,080 high speed and then most schooled. 313 00:16:20,519 --> 00:16:22,679 Speaker 2: So what you have to do telling to go when 314 00:16:22,799 --> 00:16:23,919 the code, I. 315 00:16:23,879 --> 00:16:26,639 Speaker 1: Mean, you have a real problem there. Yeah, and wow, 316 00:16:27,200 --> 00:16:29,679 And so it's real clear to need. 317 00:16:29,639 --> 00:16:32,080 Speaker 3: To see how that's gonna pan out. And also we're 318 00:16:32,120 --> 00:16:34,759 going up to spurt where the you know, next year 319 00:16:34,799 --> 00:16:36,720 with the AI is going to be is way more 320 00:16:36,799 --> 00:16:39,559 than this year was compared to last year. It's going 321 00:16:39,639 --> 00:16:42,440 to keep it's you know, increasing like that. I mean, 322 00:16:42,440 --> 00:16:45,039 they're running into this temporary roadblock, but they're running out 323 00:16:45,080 --> 00:16:45,399 of data. 324 00:16:45,799 --> 00:16:47,000 Speaker 1: Yeah, but they're going to solve that. 325 00:16:47,279 --> 00:16:50,039 Speaker 2: We you know, I've been in tech technology long enough, 326 00:16:50,120 --> 00:16:53,320 you know. Yeah, the paper stuffed up five thousand papers 327 00:16:53,360 --> 00:16:54,799 a month, one of them is gonna work. 328 00:16:55,159 --> 00:16:59,519 Speaker 1: Yeah. So so the reasoning part that's don't they always 329 00:16:59,600 --> 00:17:03,120 kind of s say that's when it starts reasoning. That's 330 00:17:03,120 --> 00:17:05,759 when it's going to get scary, you know, I mean, 331 00:17:05,839 --> 00:17:07,599 do you understand what I mean by that way I do? 332 00:17:08,079 --> 00:17:10,559 And what do you what's your opinion on that? 333 00:17:10,759 --> 00:17:14,720 Speaker 3: Well, in my talk I talk about this because every 334 00:17:14,880 --> 00:17:18,359 starting in twenty twelve, when I started actually work, you know, 335 00:17:18,480 --> 00:17:22,119 sort of do things, there's been a just a list 336 00:17:22,200 --> 00:17:29,400 of big surprises. So these designers, you know, where do 337 00:17:29,480 --> 00:17:31,519 I start? But I'm just going to throw in here. 338 00:17:32,039 --> 00:17:33,400 You don't program it. 339 00:17:33,759 --> 00:17:34,480 Speaker 2: That's not what you do. 340 00:17:34,559 --> 00:17:37,119 Speaker 3: It's not programmers telling it what to do or how 341 00:17:37,119 --> 00:17:39,759 to think. That's not how it works. There's a there's 342 00:17:39,799 --> 00:17:43,440 an artificial neuron that you do locate a billion times 343 00:17:43,720 --> 00:17:48,119 and literally billions of neurons, and all this offtware is 344 00:17:48,160 --> 00:17:51,160 doing is simulating them, but inputs the neuron and outputs 345 00:17:51,160 --> 00:17:53,240 and it doesn't a billion times for each gigeration. 346 00:17:53,960 --> 00:17:57,559 Speaker 2: So there's it's not like anybody's programming to think. What 347 00:17:57,599 --> 00:17:59,119 they're doing is architecting it. 348 00:17:59,640 --> 00:18:02,160 Speaker 3: So they they have found out that certain arrangements of 349 00:18:03,319 --> 00:18:05,400 you know, neurons have. 350 00:18:05,480 --> 00:18:08,119 Speaker 6: A certain you know function very you know, it's deliberate, 351 00:18:08,200 --> 00:18:11,160 and they're working on it in writing papers, but they 352 00:18:11,200 --> 00:18:14,039 still don't know exactly even how it works, but I 353 00:18:14,079 --> 00:18:16,519 was gonna say, like these emerging features that come out. 354 00:18:16,640 --> 00:18:20,680 Speaker 2: For example, in twenty twelve to twelve, Google. 355 00:18:20,440 --> 00:18:23,400 Speaker 3: Was trying to just make something that would complete a 356 00:18:23,480 --> 00:18:27,200 video frame, so they wanted it to show an incomplete 357 00:18:27,200 --> 00:18:28,759 image and have it fill in the gaps. 358 00:18:29,359 --> 00:18:31,440 Speaker 2: And the thing ended up learning what a cat was 359 00:18:31,960 --> 00:18:34,400 and what a human face was. And I wouldn't even 360 00:18:34,400 --> 00:18:36,039 tell them anything about that. 361 00:18:36,599 --> 00:18:37,640 Speaker 1: Somehow I figured that out. 362 00:18:37,640 --> 00:18:42,039 Speaker 3: O my god, and you fast forward, they have a 363 00:18:42,160 --> 00:18:46,079 language model where they were in fact, language models were. 364 00:18:45,920 --> 00:18:47,519 Speaker 2: Basically invented for translation. 365 00:18:48,119 --> 00:18:50,319 Speaker 3: They found out that this is one of the events 366 00:18:50,319 --> 00:18:53,079 that occurred where they they taught it. You know, your 367 00:18:53,279 --> 00:18:57,799 English to Spanish right now, you're Spanish to French. They 368 00:18:57,799 --> 00:19:00,799 found out it would actually do English Diffrench even though 369 00:19:00,839 --> 00:19:03,119 they never trained it to do that, so it had 370 00:19:03,160 --> 00:19:04,880 some kind of intermediate. 371 00:19:04,359 --> 00:19:07,400 Speaker 2: Language it was working with if they didn't even know them. 372 00:19:07,720 --> 00:19:11,640 Speaker 5: So when you talk about reasoning, these AI models have 373 00:19:11,720 --> 00:19:16,279 been surprising their developers, and they didn't they don't even 374 00:19:16,400 --> 00:19:18,839 know how to design that in They don't. 375 00:19:19,440 --> 00:19:22,240 Speaker 3: They're just they just learned that it could put me 376 00:19:22,359 --> 00:19:25,559 more data, it does more stuff, so you. 377 00:19:25,519 --> 00:19:28,799 Speaker 2: Know, their ego language models actually convert poetry. They can 378 00:19:28,839 --> 00:19:31,200 be humorous, they can I. 379 00:19:31,160 --> 00:19:33,839 Speaker 3: Mean, nobody expected that. Nobody knew that was going to happen. 380 00:19:34,319 --> 00:19:37,599 And then when we get to the next stage, they're reasoning. 381 00:19:37,720 --> 00:19:41,039 They're actually really reasoning. So they figured out how to 382 00:19:41,079 --> 00:19:43,480 like optimize that. So they think they can go back 383 00:19:43,519 --> 00:19:44,440 and improve. 384 00:19:44,119 --> 00:19:46,240 Speaker 2: Things, but nobody really knows. 385 00:19:46,680 --> 00:19:49,319 Speaker 3: You know, five years ago, they would have never known 386 00:19:49,359 --> 00:19:52,319 how to design something that could reason. Not here we are, 387 00:19:52,480 --> 00:19:55,119 we had stuff that reasons and it's and you know 388 00:19:55,160 --> 00:19:56,720 how you're training of thought. 389 00:19:57,599 --> 00:19:59,680 Speaker 2: Yeah, yeah, so they'll actually tell you what they're thinking. 390 00:20:00,279 --> 00:20:03,200 So you know, ask a question and you can push 391 00:20:03,200 --> 00:20:04,880 a button and it will show you all the intermediate 392 00:20:04,920 --> 00:20:05,559 steps that had. 393 00:20:06,000 --> 00:20:10,839 Speaker 3: I see, it's believe fast not all you know, and 394 00:20:10,960 --> 00:20:13,960 it knows everything will go through and you know, tell 395 00:20:14,000 --> 00:20:14,759 you what I was thinking. 396 00:20:16,000 --> 00:20:16,119 Speaker 1: You know. 397 00:20:16,240 --> 00:20:20,440 Speaker 3: So they're they're figuring out kind of retrospectively optimizing some 398 00:20:20,480 --> 00:20:22,519 of the stuff it does, but then it always gives 399 00:20:22,559 --> 00:20:23,440 them a new surprise. 400 00:20:24,039 --> 00:20:27,079 Speaker 2: Well consciousness top of that. I don't have any help 401 00:20:27,119 --> 00:20:30,079 about that personal really, a lot of people you're probably do. 402 00:20:30,839 --> 00:20:31,359 Speaker 1: Yeah, but. 403 00:20:32,920 --> 00:20:35,160 Speaker 2: I think this is a processor, and I think that's 404 00:20:35,200 --> 00:20:40,079 a processor. Yeah, and nobody knows how this one works works. 405 00:20:40,160 --> 00:20:44,640 Speaker 1: Yeah, so just to just maybe maybe it's jumping to 406 00:20:44,720 --> 00:20:48,759 a different topic. I'm not sure, but supercomputers there's some 407 00:20:48,880 --> 00:20:51,519 type of relation, you could say, And is it Sycamore? 408 00:20:51,640 --> 00:20:53,680 What is what am I thinking? You're not familiar with that. 409 00:20:54,839 --> 00:20:58,759 There's a new chip that's supposed to be scary for 410 00:20:59,319 --> 00:21:04,240 what it can do. Possibly, yes, it's quantum computing, but 411 00:21:04,319 --> 00:21:06,839 there's a I think it was called Sycamore by Google, 412 00:21:07,039 --> 00:21:10,640 but I'm not really sure. But anyway, supposedly it was 413 00:21:10,680 --> 00:21:15,519 able to figure out something that would take thousands of 414 00:21:15,640 --> 00:21:19,559 years for a supercomputer to do, did it instantly, like 415 00:21:19,759 --> 00:21:23,920 within within ten minutes. So I mean that's that's scary 416 00:21:24,000 --> 00:21:27,119 because well, for one thing, again this isn't exactly what 417 00:21:27,559 --> 00:21:31,319 AI or anything, but for one thing, these type of 418 00:21:31,440 --> 00:21:34,839 situations could get through any type of security. You know, 419 00:21:34,920 --> 00:21:37,799 there's no if there's no code that it couldn't get 420 00:21:37,839 --> 00:21:41,480 around to get through into any bank account or anything 421 00:21:41,599 --> 00:21:43,880 like that. And that is, you know, that is scary. 422 00:21:44,200 --> 00:21:47,359 But getting back to AI in general, you know, I 423 00:21:47,519 --> 00:21:54,680 heard something about ninety megawatts or something like that to 424 00:21:55,079 --> 00:21:58,079 run these things, And why does it take so much power. 425 00:21:59,519 --> 00:22:01,200 Speaker 2: Well, talk about supercomputers. 426 00:22:01,839 --> 00:22:04,960 Speaker 3: So you know, they figured out, oh look if that's something, 427 00:22:05,000 --> 00:22:07,519 but if we have twice of many neurons, it doesn't better. 428 00:22:08,279 --> 00:22:10,160 Speaker 2: So it's the end. They just turned up the scale. 429 00:22:10,240 --> 00:22:12,799 How many neurons did we want? Let's keep going. 430 00:22:13,279 --> 00:22:15,079 Speaker 1: So I mean, right now, I think rock. 431 00:22:15,039 --> 00:22:15,480 Speaker 2: Is that people. 432 00:22:15,559 --> 00:22:19,599 Speaker 3: We'd set a trillion shit trillion neurons in it. So 433 00:22:20,279 --> 00:22:23,519 for a trillion parameters I'm sorry, but I mean parameters. 434 00:22:24,039 --> 00:22:27,519 You know, let's say a neuron has fifty parameters just 435 00:22:27,599 --> 00:22:29,200 to throw anu there, it's not it's. 436 00:22:29,119 --> 00:22:30,559 Speaker 1: Not far off you have so. 437 00:22:32,119 --> 00:22:34,839 Speaker 2: Anyway, it takes more power to compood. And then they 438 00:22:34,920 --> 00:22:36,880 wanted to, you know, want to so you give it 439 00:22:37,319 --> 00:22:39,640 an inquot you want an answer right away. So speed 440 00:22:39,799 --> 00:22:42,440 is important and there's no limit to that. Yeah, let's 441 00:22:42,440 --> 00:22:46,079 say or whatever, one hundred milliseconds nicely, fifty million seconds, right, 442 00:22:46,240 --> 00:22:49,480 So that and the economics to it all. 443 00:22:49,839 --> 00:22:53,960 Speaker 3: So that's Pastorward's going. But there's no really, there's no 444 00:22:54,799 --> 00:22:57,920 Oh we've got enough. Yeah, there's never I mean, if 445 00:22:58,119 --> 00:23:02,720 if it took only lots whatever, they say. 446 00:23:02,599 --> 00:23:08,799 Speaker 1: Okay, neurons, Yeah, we're gonna go use whatever to get 447 00:23:08,799 --> 00:23:15,640 available yeah, yes it is. I guess the the scary 448 00:23:15,720 --> 00:23:19,599 part of how this could go is, you know, like 449 00:23:19,880 --> 00:23:24,400 you always see these dystopian you know, movies like what 450 00:23:24,599 --> 00:23:28,079 was that with Arnold Schwarzenegger, you know in this sky net? Yeah, 451 00:23:28,359 --> 00:23:32,559 things like that sort of sort of taking over. But 452 00:23:33,759 --> 00:23:37,480 I guess the rationale would be like for to be 453 00:23:37,640 --> 00:23:40,200 programmed into this thing. I know you mentioned there's there's 454 00:23:40,200 --> 00:23:42,480 no real programming to do, but if it had and 455 00:23:42,559 --> 00:23:47,279 it's set, it's I want to say mindset, like to 456 00:23:47,400 --> 00:23:51,720 do it to do a specific thing, and that would 457 00:23:51,720 --> 00:23:54,720 actually wipe out people to do that specific thing. 458 00:23:54,880 --> 00:23:56,799 Speaker 2: You were talking about the utility function. 459 00:23:57,119 --> 00:24:00,839 Speaker 1: So yeah, I haven't all along doesn't really have utility. 460 00:24:02,519 --> 00:24:03,319 Speaker 2: The best answer. 461 00:24:05,240 --> 00:24:10,200 Speaker 1: Yeah, okay, so let's see. So I'm just going to 462 00:24:10,279 --> 00:24:16,799 go back to that. Yeah, so if you could just 463 00:24:16,839 --> 00:24:25,559 started out with the utility functions while we're on this 464 00:24:25,640 --> 00:24:28,160 little break, is there any question that I can ask 465 00:24:28,200 --> 00:24:30,599 you that you would like to talk about, like a 466 00:24:30,759 --> 00:24:34,720 specific area of what you've flipked into or is it 467 00:24:34,799 --> 00:24:37,599 going well enough? Is it going okay? 468 00:24:38,440 --> 00:24:38,720 Speaker 2: Okay? 469 00:24:39,279 --> 00:24:46,279 Speaker 1: Good? Let me just my son's been tet he's going 470 00:24:46,319 --> 00:24:48,920 to be excited to hear the show he hasn't listen 471 00:24:48,960 --> 00:24:54,960 to any of the OA. 472 00:24:58,880 --> 00:25:18,119 Speaker 7: Okay, I'll do that. One guy n't like this? Good, 473 00:25:18,440 --> 00:25:19,799 all right, and. 474 00:25:19,839 --> 00:25:21,000 Speaker 2: You're gonna do your plot plot? 475 00:25:21,079 --> 00:25:23,720 Speaker 1: Two? What's on to your plot plot? You know? Ship, 476 00:25:24,000 --> 00:25:26,960 I will be just happy that. I mean, you're I'm 477 00:25:27,119 --> 00:25:29,079 telling you is you're good? Okay, thank you? 478 00:25:36,200 --> 00:25:39,039 Speaker 2: Net tricking product fifteen. 479 00:25:39,839 --> 00:25:44,039 Speaker 1: Yeah, come on, it's in the media. Where do you 480 00:25:44,119 --> 00:25:47,079 happen to? Where than this question on? It's through the 481 00:25:47,160 --> 00:25:48,519 vendors on the bottom floor. 482 00:25:48,799 --> 00:25:51,720 Speaker 2: We'll see crystal rocks. 483 00:25:51,519 --> 00:25:54,559 Speaker 7: And things like that at the end of it, and 484 00:25:54,920 --> 00:25:57,400 just jig around and it's there to meet you there 485 00:25:57,799 --> 00:25:58,720 at five fifteen. 486 00:26:01,039 --> 00:26:06,240 Speaker 1: All right, all right, So that you start out saying utility. 487 00:26:06,279 --> 00:26:08,960 Speaker 2: Oh yeah, yeah, yeah, I'm talking. You're talking about the 488 00:26:09,079 --> 00:26:12,160 utility function. So the utility function is where you tell 489 00:26:12,720 --> 00:26:14,599 the system what it's gold is. 490 00:26:15,279 --> 00:26:17,839 Speaker 3: You have a utility function and I'm probably like reproducing, 491 00:26:18,519 --> 00:26:21,839 keeping alive. So and that that's unchangeable. It's something like 492 00:26:21,880 --> 00:26:24,799 you learn a different one later. It's built into you 493 00:26:24,960 --> 00:26:28,359 and you follow that guidance right and then and then 494 00:26:28,480 --> 00:26:30,480 you know, somebody's got to set that. 495 00:26:30,640 --> 00:26:33,279 Speaker 2: Up what that utility function is. In other words, you 496 00:26:33,319 --> 00:26:37,359 know from a taxi driver getting to the customer to 497 00:26:37,519 --> 00:26:40,279 the goal. It makes my score goal up. You know. 498 00:26:40,480 --> 00:26:42,400 Speaker 3: If I don't get into the goal, my score goes down. 499 00:26:42,440 --> 00:26:44,119 So I'm trying to get into the goal, you know. 500 00:26:44,599 --> 00:26:45,839 So that's the utility. 501 00:26:45,599 --> 00:26:48,039 Speaker 2: Function, and then what you're referring to is called the 502 00:26:48,119 --> 00:26:50,240 alignment of company utility. 503 00:26:50,240 --> 00:26:54,240 Speaker 3: You know, be the best taxi driver, and maybe maybe 504 00:26:54,279 --> 00:26:55,759 it comes up with a different way of being a 505 00:26:55,799 --> 00:26:57,799 taxi driver, you know. 506 00:26:57,880 --> 00:27:00,640 Speaker 2: Maybe whatever it does. You know, there's a there's a 507 00:27:00,680 --> 00:27:02,480 classic example. I don't know how deep I want to 508 00:27:02,480 --> 00:27:04,559 know it's good, but there's it's called the stamp Collector. 509 00:27:04,599 --> 00:27:07,079 Speaker 3: You can find a YouTube video from twenty seventeen where 510 00:27:07,119 --> 00:27:10,480 he talks about it. But the stamp collector, you know, 511 00:27:11,400 --> 00:27:14,759 the idea is like more stamps. So then it said, 512 00:27:14,839 --> 00:27:18,640 you know, oh, it's an eBay, and I mean some 513 00:27:18,799 --> 00:27:20,359 stamps show up and that's good, you know. 514 00:27:20,759 --> 00:27:22,599 Speaker 2: But then it figures out, hey. 515 00:27:22,480 --> 00:27:25,680 Speaker 3: You know, I don't invite all of the standpoint flectors 516 00:27:25,680 --> 00:27:28,480 around the world. Just send them, send me stamps and 517 00:27:28,519 --> 00:27:30,319 I'll put them in the museum. And then it makes 518 00:27:30,359 --> 00:27:33,039 a fake website for a museum, you know, or maybe 519 00:27:33,079 --> 00:27:35,319 it builds a virus that turns. 520 00:27:35,039 --> 00:27:37,000 Speaker 2: All the printers in the world on and they start 521 00:27:37,200 --> 00:27:38,640 turning stamps and sending. 522 00:27:38,400 --> 00:27:41,160 Speaker 1: It to them or something like that. So it can 523 00:27:41,359 --> 00:27:43,279 you can go off the rails pretty quick. And that's 524 00:27:43,319 --> 00:27:45,960 the alignment problem. How do you how do you make 525 00:27:46,039 --> 00:27:46,440 sure that. 526 00:27:48,200 --> 00:27:49,799 Speaker 2: I'm not going to remember this quote? 527 00:27:49,880 --> 00:27:52,680 Speaker 3: It's something like you haven't really proven that it that 528 00:27:52,799 --> 00:27:55,400 it's but it's dangerous. You've just proven that everything I 529 00:27:55,440 --> 00:27:57,640 can think about it, I can't think the way it 530 00:27:57,799 --> 00:27:58,480 is dangerous. 531 00:28:00,079 --> 00:28:01,039 Speaker 2: Thing is not dangerous. 532 00:28:01,440 --> 00:28:01,640 Speaker 1: Yeah. 533 00:28:01,680 --> 00:28:05,480 Speaker 3: So the trick is to know one of your problem 534 00:28:05,559 --> 00:28:08,079 is the whole thing you have, the super intelligent thing. 535 00:28:08,240 --> 00:28:11,200 You give it a goal and might decide killing all 536 00:28:11,279 --> 00:28:13,240 humans would actually achieve it all better. 537 00:28:13,720 --> 00:28:16,119 Speaker 1: Yeah, and that's where you get into problem. Yeah, I 538 00:28:16,160 --> 00:28:20,039 guess you would. Yeah. Now, there was a thought experiment 539 00:28:20,279 --> 00:28:24,079 and it had something to do with a say, there 540 00:28:24,160 --> 00:28:26,119 was a cat inside of a box. Have you ever 541 00:28:26,160 --> 00:28:29,880 heard of this one? Whether the cat was alive or short? Yeah, 542 00:28:29,960 --> 00:28:34,039 that's it. Yeah. I mean what if you asked AI 543 00:28:34,200 --> 00:28:36,960 to tackle something like that? I mean, would it just 544 00:28:38,160 --> 00:28:41,119 I mean does it would it just randomize on an answer? 545 00:28:43,559 --> 00:28:46,000 Speaker 3: I mean, you know, move here's what it would do 546 00:28:46,160 --> 00:28:48,119 right now, right, you're asking a couple of years what 547 00:28:48,240 --> 00:28:50,279 it's going to do. I mean again, we want before 548 00:28:50,279 --> 00:28:52,839 it's going to do stuff we can't do. Yeah, And 549 00:28:53,240 --> 00:28:57,039 the shortage is a classic physics philosophical question where they 550 00:28:57,079 --> 00:29:00,599 were dating fifty years whatever, seventy years ago and we're. 551 00:29:00,440 --> 00:29:02,440 Speaker 2: Still debating that because we don't really know what the 552 00:29:02,519 --> 00:29:02,920 answer is. 553 00:29:03,559 --> 00:29:06,640 Speaker 3: But how long will it be before you know, AI 554 00:29:06,839 --> 00:29:10,160 can be you know, work on physics problems and come 555 00:29:10,240 --> 00:29:11,559 up with a brand unified theory. 556 00:29:12,200 --> 00:29:16,079 Speaker 1: Yeah, well we don't know. Yeah, but again, like so they. 557 00:29:16,039 --> 00:29:18,920 Speaker 3: Have you know, right's artificial intelligence, and then when it 558 00:29:19,000 --> 00:29:21,079 becomes general. 559 00:29:20,359 --> 00:29:23,599 Speaker 2: And can attack new problems kind in general. 560 00:29:23,440 --> 00:29:25,960 Speaker 3: Sense and his body smart as us as a GI 561 00:29:26,240 --> 00:29:30,720 are general intelligence. And then there's artificial superintelligence just you know, 562 00:29:30,920 --> 00:29:35,319 just beyond where we are. They say that the difference 563 00:29:35,359 --> 00:29:38,279 between those two is one month. Do you tell a 564 00:29:38,440 --> 00:29:42,359 g I it's making the anside. So that's that exponential curve. 565 00:29:42,440 --> 00:29:45,680 You know, you can start basically making it build itself better, 566 00:29:46,359 --> 00:29:47,200 self improvement. 567 00:29:47,319 --> 00:29:49,599 Speaker 2: That's part of this called the singularity. 568 00:29:50,720 --> 00:29:54,960 Speaker 1: Now, who is a Kurt that was just singularity years ago? 569 00:29:55,039 --> 00:29:59,000 We're talking about Kurt Curswell, that's curse while I'm thinking 570 00:29:59,039 --> 00:30:01,640 of yeah, yeah, uh so, yeah, I mean was it? 571 00:30:01,640 --> 00:30:03,759 Would you consider him ahead of his time? And on 572 00:30:03,920 --> 00:30:05,880 those thoughts, sure, yeah, you know. 573 00:30:06,240 --> 00:30:07,599 Speaker 8: I don't want to see anything away from him. But 574 00:30:07,680 --> 00:30:10,440 a lot of this is kind of self evident. So oh, 575 00:30:10,519 --> 00:30:12,559 I see a lot of these ideas are or what 576 00:30:12,640 --> 00:30:15,079 you would expect. And it's not like somebody is now 577 00:30:15,640 --> 00:30:17,240 reading his book and following along. 578 00:30:17,839 --> 00:30:19,759 Speaker 1: Yeah, this is just the way it's going to roll out. 579 00:30:19,799 --> 00:30:21,640 Speaker 3: And of course you're going to use your best tools 580 00:30:22,039 --> 00:30:24,279 to make it to make itself better, you know, and 581 00:30:24,400 --> 00:30:25,240 that's gonna. 582 00:30:24,960 --> 00:30:28,200 Speaker 1: Be natural things. So yeah, he nailed it. 583 00:30:28,759 --> 00:30:30,759 Speaker 2: Yeah, of course he's a visionary. 584 00:30:30,559 --> 00:30:35,640 Speaker 1: Right, yeah, tell but yeah, yeah, so I guess I 585 00:30:35,720 --> 00:30:38,839 want to ask you, how scared do we think we 586 00:30:38,960 --> 00:30:43,960 have to be of where AI is going? I don't 587 00:30:44,480 --> 00:30:48,759 really weird answer that's okay because to me, it's like 588 00:30:49,319 --> 00:30:54,559 I gid up, really, And here's why there's no. 589 00:30:54,599 --> 00:30:57,599 Speaker 3: Way Congress can keep ahead to disturb or right right, 590 00:30:58,160 --> 00:31:05,119 you know AI regulations. You have capitalism which you cannot control. Yeah, 591 00:31:05,279 --> 00:31:11,440 if anybody hasn't noticed control capitalism and the profit for this, 592 00:31:11,839 --> 00:31:13,319 you know, the economic. 593 00:31:12,920 --> 00:31:18,319 Speaker 2: Incentive for this is off the charts, and you know, 594 00:31:18,519 --> 00:31:19,480 I don't know, just. 595 00:31:19,559 --> 00:31:20,680 Speaker 1: Let's see what's gonna happen. 596 00:31:20,759 --> 00:31:23,160 Speaker 2: But yeah, I don't know you could ever do anything 597 00:31:23,359 --> 00:31:24,039 with rests. 598 00:31:24,680 --> 00:31:27,960 Speaker 1: All right. As I was saying earlier, not too much earlier, 599 00:31:28,000 --> 00:31:32,599 but I said, it seems to me that in this 600 00:31:32,839 --> 00:31:35,880 type of situation, people with the money are going to 601 00:31:35,960 --> 00:31:39,599 get richer and richard at the top. But anytime that happens, 602 00:31:39,640 --> 00:31:42,960 you think of historically, you know, there's been the revolution, 603 00:31:43,160 --> 00:31:47,680 the French Revolution in Russia nineteen seventeen. Whenever that was, 604 00:31:49,319 --> 00:31:53,440 they toppled because you know, people are hurting, people are starving, 605 00:31:53,519 --> 00:31:56,640 and you know, I mean this, I don't see anyone 606 00:31:56,720 --> 00:32:00,519 writing AI to help people that are hungry or you know, 607 00:32:01,319 --> 00:32:05,079 so I wonder, you know, if that's if that's what's 608 00:32:05,319 --> 00:32:09,200 going to happen is because eventually, if it goes in 609 00:32:09,279 --> 00:32:11,920 that direction, it's going. 610 00:32:11,839 --> 00:32:12,400 Speaker 2: To top them. 611 00:32:12,480 --> 00:32:13,720 Speaker 1: It's going to have to top them. 612 00:32:13,759 --> 00:32:16,880 Speaker 3: There's a very valid argument to say, well, this is 613 00:32:16,960 --> 00:32:18,640 what happened in the past, this is what's going to 614 00:32:18,680 --> 00:32:21,759 happen now. But there is some point in the history 615 00:32:21,920 --> 00:32:24,759 of a civilization or a planner or something where it. 616 00:32:24,839 --> 00:32:28,279 Speaker 2: Actually doesn't you know like this is the end, and 617 00:32:28,319 --> 00:32:31,960 it actually does end, so it's something you can red 618 00:32:32,440 --> 00:32:34,039 and yeah, and that doesn't work anymore. 619 00:32:34,119 --> 00:32:35,319 Speaker 1: But here I think. 620 00:32:37,480 --> 00:32:39,480 Speaker 3: You know you have this ability, like I said, where 621 00:32:39,640 --> 00:32:43,240 an agent can be following, a super intelligent agent can 622 00:32:43,279 --> 00:32:46,160 be following you and finding out what your being finding 623 00:32:46,200 --> 00:32:48,680 out if you're trying to start a counter revolution, you. 624 00:32:48,680 --> 00:32:50,640 Speaker 1: Know, and we know, and you wouldn't even know what 625 00:32:50,759 --> 00:32:53,480 was following you, right, I mean that's a possibility. 626 00:32:53,599 --> 00:32:54,039 Speaker 2: Excuse me. 627 00:32:54,079 --> 00:32:57,640 Speaker 3: It's been following since twenty twelve, since social media started, 628 00:32:58,240 --> 00:33:01,000 You've had an a agent's follow Oh. 629 00:33:01,759 --> 00:33:03,599 Speaker 1: The story there's a book that came out of years ago, 630 00:33:03,920 --> 00:33:07,000 in twenty sixteen or something like that, where this. 631 00:33:07,400 --> 00:33:10,440 Speaker 2: Lady one of the examples in the book but I 632 00:33:10,519 --> 00:33:12,359 think she called it. The name of the book is 633 00:33:14,400 --> 00:33:16,880 Surveillance Capitalism what the book is called. 634 00:33:16,880 --> 00:33:20,119 Speaker 3: But she says, there's somebody, an example of her book, 635 00:33:20,599 --> 00:33:24,279 who's getting ads now and she finds out she's springing 636 00:33:24,319 --> 00:33:28,039 it because the ads are like baby stuff. Oh my god, 637 00:33:28,240 --> 00:33:30,119 she didn't know she was bringing it when she was 638 00:33:30,160 --> 00:33:34,440 getting baby ads. Yes, it saw her patterns and understood 639 00:33:34,440 --> 00:33:36,720 what well she, you know, was going on with her 640 00:33:37,319 --> 00:33:39,119 yeap whatever morning sickness or what. 641 00:33:39,200 --> 00:33:41,160 Speaker 2: I don't even know whether it was looking at it 642 00:33:41,519 --> 00:33:43,480 to decide this person is probably pregnant. 643 00:33:44,079 --> 00:33:46,039 Speaker 1: Well, I have to tell you something. This This happened 644 00:33:46,079 --> 00:33:50,279 to me recently. I thought about getting an electric bicycle, okay, 645 00:33:50,680 --> 00:33:53,200 and an e bike. And I thought about it. And 646 00:33:53,400 --> 00:33:56,960 I didn't search anything at all, no searchers. I talked 647 00:33:57,000 --> 00:33:59,599 to my son on the phone about it. Yeah, I 648 00:33:59,720 --> 00:34:02,839 opened up Facebook. What's the first thing I see electric bikes? 649 00:34:03,640 --> 00:34:06,640 E bikes? And I'm still getting the ads okay, And 650 00:34:07,599 --> 00:34:10,000 but I did after I did look, I did, like 651 00:34:10,159 --> 00:34:13,239 look at one of marketplace or something like that. After 652 00:34:13,519 --> 00:34:17,199 But how did that happen when I made it? When 653 00:34:17,199 --> 00:34:18,039 I was on the phone. 654 00:34:18,119 --> 00:34:20,320 Speaker 3: I mean, of course I don't have the answer because 655 00:34:20,320 --> 00:34:24,159 I'm smart, But I mean they're watching everything you're doing. 656 00:34:24,440 --> 00:34:26,280 I don't know if it's smart enough to figure out 657 00:34:26,280 --> 00:34:29,199 you're just starting to the interesting in like, yeah, maybe 658 00:34:29,239 --> 00:34:30,360 you didn't look at something. 659 00:34:30,400 --> 00:34:32,880 Speaker 1: I don't, who knows, But yeah I might have. I 660 00:34:32,960 --> 00:34:35,440 might have looked at sometimes forgot I did. That's possible. 661 00:34:35,639 --> 00:34:39,400 Speaker 2: Yeah, But a lot of people say it's listening to calls. 662 00:34:40,000 --> 00:34:41,320 Speaker 1: Yeah, we know that's not true. 663 00:34:41,480 --> 00:34:42,760 Speaker 2: I don't think you've anything god of that. 664 00:34:43,199 --> 00:34:44,159 Speaker 1: Yeah, you know what I mean? 665 00:34:46,039 --> 00:34:49,760 Speaker 2: Right now in such a valuable tool and if you 666 00:34:50,039 --> 00:34:51,360 if you have an opening eye. 667 00:34:51,559 --> 00:34:54,239 Speaker 9: For example, when you use the voice, yes, yeah, the 668 00:34:54,320 --> 00:34:57,159 voice thing, so you ask the questions and conversation with it, 669 00:34:58,920 --> 00:35:01,800 it's all going to the it's all going over they 670 00:35:02,559 --> 00:35:03,519 what are they doing with it? 671 00:35:04,360 --> 00:35:09,880 Speaker 1: Yeah, well let's talk about just a couple more questions. 672 00:35:10,440 --> 00:35:13,599 Thank you for your time. So I'm trying to remember 673 00:35:15,440 --> 00:35:19,199 Walking Phoenix was in the movie called Her I believe 674 00:35:19,239 --> 00:35:22,320 it was, and it was about his he fell in 675 00:35:22,400 --> 00:35:28,400 love with Ai. That and then I heard someone h 676 00:35:28,920 --> 00:35:31,159 a friend of mine, of the family whatever I was 677 00:35:31,239 --> 00:35:34,119 talking about, said that she has a companion. 678 00:35:35,000 --> 00:35:37,119 Speaker 2: Oh yeah, and it's an. 679 00:35:37,079 --> 00:35:40,599 Speaker 1: AI companion and it's kind of like it keeps me 680 00:35:40,679 --> 00:35:43,960 from feeling lonely and all that. Is this where we're going? 681 00:35:44,239 --> 00:35:48,599 Do you think we're going to have relationships with with AI? 682 00:35:48,800 --> 00:35:52,280 Do you think that's the possible future. I don't think 683 00:35:52,320 --> 00:35:53,039 it's so far off. 684 00:35:53,199 --> 00:35:56,679 Speaker 2: It's already happening. I know that it's actually pretty good 685 00:35:56,719 --> 00:35:59,920 at being a psychologists. Yeah, I think it's pretty easy 686 00:36:00,039 --> 00:36:02,880 a human. I don't think we're not complicated. 687 00:36:02,960 --> 00:36:05,480 Speaker 3: And nothing, you know, has the ability to live at 688 00:36:05,800 --> 00:36:09,800 multiple variables, which we normally would call it intuition. Right, 689 00:36:09,840 --> 00:36:12,239 you meet somebody who's not that ability to kind of 690 00:36:12,280 --> 00:36:16,280 figure things out, here's a AI probably ten times the 691 00:36:16,360 --> 00:36:18,280 ability to do that for all I know. 692 00:36:19,159 --> 00:36:21,079 Speaker 1: And you know, just. 693 00:36:22,840 --> 00:36:26,320 Speaker 2: There's there's stories like someone just told me their good 694 00:36:26,360 --> 00:36:30,920 friend not his girlfriend, because a I had told him. 695 00:36:32,840 --> 00:36:37,480 Speaker 1: My god, it's happening. Yeah, it's happening. There are stories about. 696 00:36:37,320 --> 00:36:41,280 Speaker 3: You know, millennials whatever, they're there, you know, they have companions, 697 00:36:41,400 --> 00:36:43,880 and that's that's happening. And these things are only getting 698 00:36:43,880 --> 00:36:46,719 get smarter, only to get more intuitive, only going to 699 00:36:46,760 --> 00:36:50,440 get more I mean, I've my wife inside know that's 700 00:36:50,440 --> 00:36:53,760 when I say this, But I mean, they're gonna have feelings, They're. 701 00:36:53,599 --> 00:36:57,039 Speaker 2: Gonna have emotions, especially when you have that untility function. 702 00:36:58,159 --> 00:36:59,280 Uh you know, here's my. 703 00:36:59,360 --> 00:37:03,199 Speaker 3: Example, a robot that's not the utility function. You plug 704 00:37:03,239 --> 00:37:07,800 yourself in every night to recharge. We'll try standing in 705 00:37:07,880 --> 00:37:12,199 front of the plug. You know, it's at least looser. No, 706 00:37:12,800 --> 00:37:16,039 I don't want to move why, I just don't want to. 707 00:37:16,639 --> 00:37:19,000 Speaker 1: Just see what happens. I think you can see robot, 708 00:37:19,639 --> 00:37:21,119 that's what that would be wow. 709 00:37:21,360 --> 00:37:25,599 Speaker 10: And I think, you know, essentially, I think our our 710 00:37:26,039 --> 00:37:27,800 feelings are that same thing. I mean, I'm not going 711 00:37:27,880 --> 00:37:30,840 to attend you know, what feelings are, but I can 712 00:37:30,960 --> 00:37:35,239 see how that would start with a robot. And again, 713 00:37:35,599 --> 00:37:38,440 stuff is going to emerge that no one ever expected. 714 00:37:38,840 --> 00:37:41,960 Speaker 1: Yes, and I'm the design it to do A and 715 00:37:42,039 --> 00:37:45,159 it's going to do A plus B plus see plus see. Yeah. 716 00:37:46,000 --> 00:37:49,159 So we we've already I think we've already covered probably 717 00:37:49,280 --> 00:37:53,039 like the worst things that could happen. So why don't 718 00:37:53,039 --> 00:37:55,119 we end on like a higher note and just say, 719 00:37:55,440 --> 00:37:57,840 what do you think is the best that kind of 720 00:37:57,920 --> 00:38:00,480 I've already said medical, Maybe that's it. No, do you 721 00:38:00,760 --> 00:38:03,960 have any other ideas? Do you think it will can 722 00:38:04,880 --> 00:38:07,760 like we can grow in some type of way with 723 00:38:09,079 --> 00:38:14,599 like off of artificial intelligence. I'm sorry, I'm so that's okay. 724 00:38:14,719 --> 00:38:17,440 So that's all right. I wanted I would rather matter 725 00:38:18,400 --> 00:38:22,800 pure for cancer. I mean, I don't know, Yeah, I 726 00:38:23,159 --> 00:38:24,920 really think I just it's. 727 00:38:24,800 --> 00:38:28,199 Speaker 2: Hard to imagine it not being a dystopian. That's that's one. 728 00:38:28,360 --> 00:38:31,239 Speaker 1: Well, that's what all the all the movies portray anything 729 00:38:31,320 --> 00:38:34,079 in the future just about is always dystopian. 730 00:38:34,280 --> 00:38:36,079 Speaker 2: Well, I mean we're. 731 00:38:36,079 --> 00:38:38,880 Speaker 3: Kind of in a dystopian now, and so you throw 732 00:38:38,960 --> 00:38:42,320 in there at super intelligence in the next year, certainly 733 00:38:42,400 --> 00:38:44,559 before the end of this administration. 734 00:38:44,920 --> 00:38:47,719 Speaker 2: Okay, you got a big mess. 735 00:38:48,400 --> 00:38:52,320 Speaker 1: Yeah. Yeah, it's a superweapon. It's in the hands, right 736 00:38:52,400 --> 00:38:56,960 it's only in the hands right now, this ultra ultra rich. Yeah, 737 00:38:57,320 --> 00:38:59,719 they're the one to own it and can control it. Yeah. 738 00:39:00,039 --> 00:39:02,559 What are they gonna do with it? Yeah? I mean 739 00:39:02,599 --> 00:39:06,760 there's no type of grassroots situation that I can think 740 00:39:06,800 --> 00:39:10,280 of in any type of way to stop that from happening. 741 00:39:10,840 --> 00:39:13,599 But anyway, I guess we're not going to you know, 742 00:39:14,360 --> 00:39:20,679 that's okay. It's been a real pleasure. Yeah, yeah, oh no, 743 00:39:20,840 --> 00:39:23,880 it is fascinating. And I love what you can find 744 00:39:23,920 --> 00:39:28,719 out just by asking questions. You know, I did trivia. 745 00:39:29,239 --> 00:39:31,920 My son's pretty good in astronomy, and we did like 746 00:39:32,039 --> 00:39:34,719 in trivia and ask us you know, easy question to 747 00:39:34,719 --> 00:39:37,920 ask us hard questions, and boy can ask some tough questions. 748 00:39:38,000 --> 00:39:40,320 And yeah, it's in that way it's entertaining. 749 00:39:40,559 --> 00:39:42,960 Speaker 2: My friend is just telling me, you know, you have 750 00:39:43,079 --> 00:39:43,679 a voice thing. 751 00:39:44,280 --> 00:39:47,039 Speaker 3: Let me saying, in a few minutes, he resolved this 752 00:39:47,599 --> 00:39:51,280 medical issue that he's chronically had for ten years just 753 00:39:51,360 --> 00:39:54,280 because he's able to have this open conversation that it's 754 00:39:54,639 --> 00:39:58,199 very objective open responses which you can. 755 00:39:58,440 --> 00:39:59,679 Speaker 1: Do with the doctor about Yeah. 756 00:40:00,119 --> 00:40:02,920 Speaker 2: Right, I'm not putting down doctors. What I'm just saying, Yeah, 757 00:40:03,440 --> 00:40:05,599 you know, I mean that's such fantastic. 758 00:40:05,719 --> 00:40:08,280 Speaker 1: Yeah. One more thing. I'm just thinking of the old 759 00:40:08,360 --> 00:40:11,000 saying garbage and garbage out. There's got to be a 760 00:40:11,079 --> 00:40:14,920 component of that. Yeah, I mean we kind of start 761 00:40:14,960 --> 00:40:16,079 a conspiracy theory. 762 00:40:16,599 --> 00:40:18,480 Speaker 3: I mean, we are if you call it garbage, we're 763 00:40:18,480 --> 00:40:23,199 putting in all all the tax humans have ever made. Yeah, 764 00:40:23,360 --> 00:40:27,320 but I mean you get from that you I kind 765 00:40:27,320 --> 00:40:29,239 of think we're beyond the garbage in the verage out 766 00:40:29,280 --> 00:40:32,000 thing because I just think that there's so much knowledge 767 00:40:32,079 --> 00:40:35,320 that in self consistency that starts to come out and 768 00:40:35,760 --> 00:40:38,639 that's part of what makes uh, you know, when treatment training, 769 00:40:38,639 --> 00:40:40,519 AI zeros. 770 00:40:40,280 --> 00:40:43,239 Speaker 2: In on patterns, right, and the patterns and speak the truth, 771 00:40:44,000 --> 00:40:47,159 you know, the stuff that's logically consistent. So I think 772 00:40:47,159 --> 00:40:47,920 we're okay there. 773 00:40:48,000 --> 00:40:50,719 Speaker 1: But on the other hand, you know, if you ask 774 00:40:51,239 --> 00:40:51,880 an AI. 775 00:40:52,079 --> 00:40:54,800 Speaker 3: Argue e goo is real, it also kind of gives 776 00:40:54,960 --> 00:40:57,320 like the average answer and the average answer is no. 777 00:40:58,280 --> 00:41:01,559 Speaker 11: Yeah, So you know you have to fight through that sometimes, Yeah, 778 00:41:03,000 --> 00:41:07,480 you find out it has established an answer amount of times, 779 00:41:07,960 --> 00:41:08,480 so that's. 780 00:41:08,360 --> 00:41:09,280 Speaker 2: A little bit of page. 781 00:41:10,320 --> 00:41:13,039 Speaker 1: Yeah. I think the most amazing one of the most 782 00:41:13,079 --> 00:41:15,920 amazing parts of this conversation that I got out of 783 00:41:15,960 --> 00:41:18,519 it is that it told someone to break up with 784 00:41:18,639 --> 00:41:22,480 the girlfriend. That to me is a mind blower. Is 785 00:41:22,559 --> 00:41:25,599 a mind blower. Yeah, yeah, how about that. Well, it's 786 00:41:25,599 --> 00:41:27,920 been a real pleasure. I really enjoyed that, enjoyed to 787 00:41:27,960 --> 00:41:30,280 talk very much. I'm so glad that you were willing 788 00:41:30,400 --> 00:41:33,119 to talk about this and that we had this conversation. 789 00:41:33,400 --> 00:41:58,519 Thank you. M m mmm mm hmmm.