1 00:00:00,160 --> 00:00:02,960 Speaker 1: Did you know there's an actual AI war happening right 2 00:00:03,000 --> 00:00:06,280 Speaker 1: now between the United States and China. And the thing 3 00:00:06,320 --> 00:00:09,240 Speaker 1: that's at risk is not just power, it's your job. 4 00:00:09,440 --> 00:00:11,799 Speaker 1: The question is how far along are we and who 5 00:00:11,920 --> 00:00:15,240 Speaker 1: is winning? Let's get into it. It seems like just 6 00:00:15,360 --> 00:00:17,480 Speaker 1: yesterday we weren't talking about AI at all. I mean 7 00:00:17,480 --> 00:00:19,239 Speaker 1: maybe that movie that was out a long time ago, 8 00:00:19,280 --> 00:00:22,239 Speaker 1: but we weren't really thinking artificial intelligence could become with 9 00:00:22,840 --> 00:00:26,079 Speaker 1: what it's become now, and one wonders is that a 10 00:00:26,120 --> 00:00:28,960 Speaker 1: good thing? Is it a bad thing? Is it both 11 00:00:29,240 --> 00:00:32,040 Speaker 1: can get out weigh the bad? And the question recently 12 00:00:32,159 --> 00:00:34,680 Speaker 1: is why are we now doing sort of a cold 13 00:00:34,720 --> 00:00:37,000 Speaker 1: war with China about AI. Let's talk to a guy 14 00:00:37,000 --> 00:00:39,840 Speaker 1: who knows. Winton Hall, author and political strategist. The book 15 00:00:39,920 --> 00:00:42,320 Speaker 1: is called Code Read the Left, the Right, China and 16 00:00:42,320 --> 00:00:43,960 Speaker 1: the race to control AI win And how are you 17 00:00:44,040 --> 00:00:44,559 Speaker 1: good to see it? 18 00:00:45,040 --> 00:00:46,360 Speaker 2: Doing great? Joe? How are you? 19 00:00:46,680 --> 00:00:49,040 Speaker 1: I'm doing very well? So should we be worried about this? 20 00:00:49,080 --> 00:00:50,280 Speaker 1: And I want to take it back if you don't 21 00:00:50,280 --> 00:00:53,199 Speaker 1: mind to Like the nineteen seventies, In the nineteen seventies 22 00:00:53,400 --> 00:00:56,120 Speaker 1: they started coming out with these things called ATMs, and 23 00:00:56,160 --> 00:00:58,400 Speaker 1: ATMs were putting people out of business now and ATM 24 00:00:58,480 --> 00:01:00,680 Speaker 1: is not AI, not at all. But it is something 25 00:01:00,720 --> 00:01:04,240 Speaker 1: electronically that takes away a job, changes how we do business, 26 00:01:04,400 --> 00:01:07,240 Speaker 1: changes our daily lives. Instead of going inside and waiting 27 00:01:07,240 --> 00:01:08,720 Speaker 1: in line and getting money, I just go to a 28 00:01:08,720 --> 00:01:10,840 Speaker 1: machine and a machine spits out some money if I 29 00:01:10,840 --> 00:01:13,640 Speaker 1: have enough money in my account. This is like that 30 00:01:13,800 --> 00:01:16,080 Speaker 1: on steroids, right, it. 31 00:01:16,000 --> 00:01:19,199 Speaker 2: Is on steroids. And the reason why is you're scaling cognition. 32 00:01:19,520 --> 00:01:21,560 Speaker 2: And one of the things that is the big debate 33 00:01:21,600 --> 00:01:25,600 Speaker 2: about jobs is whether or not scaling cognition is going 34 00:01:25,640 --> 00:01:27,680 Speaker 2: to be a very different experience than what we saw 35 00:01:27,720 --> 00:01:31,399 Speaker 2: in the Industrial Revolution when you're scaling physical labor and 36 00:01:31,440 --> 00:01:34,080 Speaker 2: manual labor. And so when you hear people like Dario 37 00:01:34,200 --> 00:01:38,240 Speaker 2: Amidae ads anthropics say that in the next twelve months, Joe, 38 00:01:38,360 --> 00:01:41,000 Speaker 2: this isn't often the far future. Twelve months, he's saying, 39 00:01:41,240 --> 00:01:44,360 Speaker 2: we're looking at fifty percent wipeout of entry level white 40 00:01:44,400 --> 00:01:47,760 Speaker 2: collar jobs over twelve months to five years. You're seeing 41 00:01:47,800 --> 00:01:52,280 Speaker 2: Mustafa Suluman a Microsoft AI certainly not a small company, 42 00:01:52,320 --> 00:01:54,440 Speaker 2: saying that in the next twelve to eighteen months, we're 43 00:01:54,440 --> 00:01:57,320 Speaker 2: looking at the ability to replace one hundred percent of 44 00:01:57,360 --> 00:02:00,680 Speaker 2: white collar job tasks. He's not saying all jobs are 45 00:02:00,680 --> 00:02:02,600 Speaker 2: going away in that time. What he's saying is the 46 00:02:02,600 --> 00:02:06,000 Speaker 2: ability to automate those tasks will be there, and it'll 47 00:02:06,000 --> 00:02:08,280 Speaker 2: be up to companies to decide how fast to diffuse it. 48 00:02:08,360 --> 00:02:10,480 Speaker 2: So you have to make a decision. You either think 49 00:02:10,560 --> 00:02:13,160 Speaker 2: number one, that that's a bunch of hype and it's 50 00:02:13,200 --> 00:02:16,360 Speaker 2: a bunch of marketing hype to try to raise investor dollars. 51 00:02:16,639 --> 00:02:19,280 Speaker 2: Two that is sincere, and they're just really trying to 52 00:02:19,280 --> 00:02:22,720 Speaker 2: warn humanity of this huge what Elon Musk calls is 53 00:02:22,760 --> 00:02:26,720 Speaker 2: a supersonic tsunami. That's what he calls AI headed a humanity. 54 00:02:27,120 --> 00:02:29,720 Speaker 2: Or three is another option, which is that you might 55 00:02:29,800 --> 00:02:34,240 Speaker 2: believe that either way, it builds support for ideologues who 56 00:02:34,240 --> 00:02:37,280 Speaker 2: want to push for things like universal basic income UBI 57 00:02:37,440 --> 00:02:39,880 Speaker 2: wealth redistribution. And so what I try to do in 58 00:02:39,960 --> 00:02:42,240 Speaker 2: Code Red is really lay out who are these people, 59 00:02:42,360 --> 00:02:45,160 Speaker 2: what are their political ideologies, what is the money flow 60 00:02:45,760 --> 00:02:48,760 Speaker 2: and their donations and so forth, so that Americans can 61 00:02:48,840 --> 00:02:51,120 Speaker 2: get caught up because this thing is moving very fast, 62 00:02:51,680 --> 00:02:52,040 Speaker 2: you know, and. 63 00:02:51,960 --> 00:02:53,280 Speaker 1: I want to get into all of it. It's Twins 64 00:02:53,320 --> 00:02:55,240 Speaker 1: and Hall going get his book called Code red, the left, 65 00:02:55,280 --> 00:02:57,880 Speaker 1: the right in China, and the race to control AI. 66 00:02:58,000 --> 00:03:00,480 Speaker 1: Before we get into all of that, where do this from. 67 00:03:00,840 --> 00:03:03,160 Speaker 1: It's like yesterday we didn't have any AI, and today 68 00:03:03,160 --> 00:03:06,160 Speaker 1: it's everywhere and I use it every day now. I 69 00:03:06,240 --> 00:03:09,200 Speaker 1: use chat, GPT, I use vidiq for some of my 70 00:03:09,240 --> 00:03:11,840 Speaker 1: posts over on YouTube. But these are good tools. I 71 00:03:11,919 --> 00:03:14,080 Speaker 1: use rock every once in a while over on X 72 00:03:14,480 --> 00:03:17,160 Speaker 1: But how did it just suddenly become become this most 73 00:03:17,200 --> 00:03:20,880 Speaker 1: important thing that we've talked about technologically? It's like two 74 00:03:21,000 --> 00:03:23,040 Speaker 1: years ago. I feel like we weren't doing this. Was 75 00:03:23,080 --> 00:03:26,080 Speaker 1: this being worked on for decades? Did somebody just did 76 00:03:26,320 --> 00:03:28,280 Speaker 1: somebody just flip uSwitch and say, hey, I found out 77 00:03:28,480 --> 00:03:29,800 Speaker 1: here it is? What happened? 78 00:03:30,400 --> 00:03:33,600 Speaker 2: That's really a smart question. So the word artificial intelligence, 79 00:03:33,600 --> 00:03:36,400 Speaker 2: that phrase has been with us since nineteen fifty six 80 00:03:36,440 --> 00:03:39,920 Speaker 2: on the faces side, and so it has been around 81 00:03:40,000 --> 00:03:42,960 Speaker 2: for a long time. But it wasn't until twenty seventeen 82 00:03:43,120 --> 00:03:47,480 Speaker 2: until a technological development called the transformer came along and 83 00:03:47,520 --> 00:03:50,160 Speaker 2: then the arrival, as you're pointing out, of generative AI, 84 00:03:50,560 --> 00:03:54,200 Speaker 2: which is AI that can generate images, texts, video, and 85 00:03:54,200 --> 00:03:57,160 Speaker 2: so forth, really started to come on into the public 86 00:03:57,200 --> 00:04:01,040 Speaker 2: consciousness in a public way when the front facing arrival 87 00:04:01,160 --> 00:04:05,080 Speaker 2: of chatchipt occurred in November of twenty twenty two, so 88 00:04:05,120 --> 00:04:08,320 Speaker 2: you're actually pretty close with your timeline there, and that 89 00:04:08,400 --> 00:04:11,200 Speaker 2: really did kick off this explosion. People said, Wow, this 90 00:04:11,280 --> 00:04:15,640 Speaker 2: is something very different. It's accelerated even faster because of 91 00:04:15,680 --> 00:04:18,080 Speaker 2: the capital investment in the next two years. We're looking 92 00:04:18,080 --> 00:04:22,240 Speaker 2: at five trillion with a T dollars of global investment 93 00:04:22,560 --> 00:04:26,520 Speaker 2: and scaling up data centers GPUs, all of it. And 94 00:04:27,040 --> 00:04:29,039 Speaker 2: then the other thing too is the arrival of what 95 00:04:29,080 --> 00:04:33,040 Speaker 2: are called AI agents, which real simple is text to action. 96 00:04:33,240 --> 00:04:37,039 Speaker 2: This is an AI that can autonomously take over your computer, 97 00:04:37,160 --> 00:04:40,440 Speaker 2: move the cursor, open up a bank account, build a website, 98 00:04:40,560 --> 00:04:43,480 Speaker 2: and do the work like a digital worker would. And 99 00:04:43,520 --> 00:04:45,040 Speaker 2: that's where you're hearing a lot of these sort of 100 00:04:45,120 --> 00:04:49,160 Speaker 2: job apocalypse predictions. But you're right, the acceleration has really 101 00:04:49,200 --> 00:04:52,599 Speaker 2: happened since that November twenty twenty two demarcation. 102 00:04:52,160 --> 00:04:57,000 Speaker 1: Point AI coodreadbook dot comics the website AI codreadbook dot com. 103 00:04:57,040 --> 00:05:00,680 Speaker 1: It's Winton Hall. The author is also political strategist. Why 104 00:05:00,680 --> 00:05:02,599 Speaker 1: does it use so much energy? We keep on hearing 105 00:05:02,600 --> 00:05:05,359 Speaker 1: that the grid can't handle it. AI takes up so 106 00:05:05,440 --> 00:05:08,200 Speaker 1: much energy. You have AI farms that are that are 107 00:05:08,240 --> 00:05:10,160 Speaker 1: attempting to be built all over the place. I'm seeing 108 00:05:10,160 --> 00:05:13,280 Speaker 1: a story that that happens probably two or three times 109 00:05:13,320 --> 00:05:15,640 Speaker 1: a week as they watch some South Florida news. They're 110 00:05:15,640 --> 00:05:17,960 Speaker 1: trying to put it where lying Country Safari is in 111 00:05:18,320 --> 00:05:21,280 Speaker 1: Palm Beach County. I think that the guy from Oracle 112 00:05:21,320 --> 00:05:22,920 Speaker 1: is the one who's trying to do that one. Why 113 00:05:22,960 --> 00:05:24,680 Speaker 1: does it take up so much space? Why does it 114 00:05:24,720 --> 00:05:25,760 Speaker 1: take up so much energy? 115 00:05:26,600 --> 00:05:30,080 Speaker 2: Yeah, so that's a really important question. Because you can 116 00:05:30,080 --> 00:05:33,679 Speaker 2: build the best GPU, you can build the best graphics 117 00:05:33,760 --> 00:05:36,560 Speaker 2: processing unit, you can build the best AI model l 118 00:05:36,680 --> 00:05:40,080 Speaker 2: ll M Large language model, but it's like building a Ferrari. 119 00:05:40,560 --> 00:05:42,440 Speaker 2: You can build it, but if you don't have the 120 00:05:42,760 --> 00:05:45,080 Speaker 2: gas and the power to be able to move it, 121 00:05:45,080 --> 00:05:47,640 Speaker 2: it really is going to not have its full capacity 122 00:05:47,680 --> 00:05:51,640 Speaker 2: to scale. The reason why is that l l ll 123 00:05:51,760 --> 00:05:55,640 Speaker 2: M prompt actually takes ten x the energy that a 124 00:05:55,839 --> 00:05:57,840 Speaker 2: that a Google search was, So if you and I 125 00:05:57,960 --> 00:06:01,520 Speaker 2: do a Google search, that's one one tenth of the 126 00:06:01,560 --> 00:06:05,000 Speaker 2: amount of energy. And the reason why is neural networks 127 00:06:05,440 --> 00:06:08,520 Speaker 2: are very very complex and they require enormous not just 128 00:06:08,600 --> 00:06:11,880 Speaker 2: compute power, but energy power as well. And so you've 129 00:06:11,920 --> 00:06:14,360 Speaker 2: heard President Trump and others say that we've got to 130 00:06:14,400 --> 00:06:17,440 Speaker 2: really unleash American energy dominance if we're going to really 131 00:06:17,440 --> 00:06:20,040 Speaker 2: be able to compete with China, but also just to 132 00:06:20,080 --> 00:06:23,600 Speaker 2: seize the upside for economic benefit for Americans. 133 00:06:24,360 --> 00:06:26,840 Speaker 1: How important is it that we have control. We keep 134 00:06:26,880 --> 00:06:29,680 Speaker 1: on talking about either the United States can be in 135 00:06:29,720 --> 00:06:31,719 Speaker 1: the lead when it comes to AI, or we seed 136 00:06:31,800 --> 00:06:34,200 Speaker 1: that to somebody else like China that wants to do it. 137 00:06:34,240 --> 00:06:36,880 Speaker 1: I did notice that China came out with Deep Seek 138 00:06:36,920 --> 00:06:38,960 Speaker 1: and they claimed it only took five million dollars in 139 00:06:39,080 --> 00:06:41,520 Speaker 1: very little research, and allegedly it's as good, of not 140 00:06:41,680 --> 00:06:43,920 Speaker 1: better than chat EPT. I'm not sure I agree with that, 141 00:06:44,279 --> 00:06:46,440 Speaker 1: but China wants to make it seem like they're already 142 00:06:46,480 --> 00:06:48,680 Speaker 1: out in front. Why is that battle so important? 143 00:06:49,360 --> 00:06:51,640 Speaker 2: Such a smart question. So let's get into that. So 144 00:06:52,360 --> 00:06:55,800 Speaker 2: the first thing is, I'm actually very heartened there's more 145 00:06:55,839 --> 00:06:58,880 Speaker 2: of a bipartisan understanding now that we do have to 146 00:06:58,880 --> 00:07:01,320 Speaker 2: beat China. What I say, code read is we want 147 00:07:01,360 --> 00:07:05,040 Speaker 2: to beat China without becoming China. Nobody wants to live 148 00:07:05,040 --> 00:07:09,200 Speaker 2: in a authoritarian CCP surveillance state at all. We don't 149 00:07:09,240 --> 00:07:11,880 Speaker 2: want to emulate their values by any stretch. But the 150 00:07:11,960 --> 00:07:14,280 Speaker 2: reason we need to beat them and why you're seeing 151 00:07:14,280 --> 00:07:17,480 Speaker 2: this bipartisan consensus start to grow is twofold one. The 152 00:07:17,560 --> 00:07:21,040 Speaker 2: economic part, right, So we know that one third of 153 00:07:21,080 --> 00:07:23,160 Speaker 2: the S and P five hundred right now is made 154 00:07:23,240 --> 00:07:25,680 Speaker 2: up of the mag seven. The Magnificent seven, that's the 155 00:07:25,800 --> 00:07:29,160 Speaker 2: name given to the seven big American tech companies, and 156 00:07:29,200 --> 00:07:31,760 Speaker 2: so much of AI obviously centers around that. So it's 157 00:07:31,760 --> 00:07:35,120 Speaker 2: a ten pole economically. The other side you brought up, 158 00:07:35,120 --> 00:07:37,320 Speaker 2: and I'm glad you did, which is that we saw 159 00:07:37,360 --> 00:07:39,720 Speaker 2: what can happen on the other side, which was the 160 00:07:39,720 --> 00:07:43,440 Speaker 2: Deep Seek moment when R one, the Chinese model was released, 161 00:07:43,760 --> 00:07:47,920 Speaker 2: that Joe was the biggest one day market cap wipeout 162 00:07:47,960 --> 00:07:50,880 Speaker 2: of an American company, in this case, Nvidia got rocked 163 00:07:50,920 --> 00:07:55,840 Speaker 2: six hundred billion dollars almost thirty trillion dollars in one day. 164 00:07:55,880 --> 00:07:58,600 Speaker 2: We all remember that. You're right though, that a lot 165 00:07:58,640 --> 00:08:01,800 Speaker 2: of that was smoking mirrors. The narrative that the CCP 166 00:08:02,000 --> 00:08:05,080 Speaker 2: and the Chinese tried to push in Deep Seak was that, 167 00:08:05,160 --> 00:08:07,840 Speaker 2: oh wow, look at this we did for six million dollars. 168 00:08:07,920 --> 00:08:11,840 Speaker 2: What took billion dollars or more for these American big companies, 169 00:08:12,400 --> 00:08:15,360 Speaker 2: we can do it better, cheaper, faster. That was a 170 00:08:15,520 --> 00:08:18,480 Speaker 2: very very misleading Number one. That did not include the 171 00:08:18,520 --> 00:08:21,520 Speaker 2: capital expenditures to build. That was just the amount they 172 00:08:21,600 --> 00:08:24,480 Speaker 2: used to train that one model. They didn't build. The 173 00:08:24,680 --> 00:08:28,160 Speaker 2: building amount was actioneerer or billion dollars on that. But 174 00:08:28,360 --> 00:08:31,000 Speaker 2: they did innovate. Now, I want to be fair, there 175 00:08:31,000 --> 00:08:34,200 Speaker 2: were two major innovations with deep Sea. One, it showed 176 00:08:34,200 --> 00:08:36,800 Speaker 2: the chain of reasoning, which real simply is how the 177 00:08:36,840 --> 00:08:40,000 Speaker 2: actual AI is arriving at its answers. And then they 178 00:08:40,000 --> 00:08:42,920 Speaker 2: did something called mixture of experts, which is a fancy 179 00:08:43,520 --> 00:08:46,200 Speaker 2: way of talking about how you can reduce the amount 180 00:08:46,200 --> 00:08:48,600 Speaker 2: of energy and efficient and make more efficient the way 181 00:08:48,640 --> 00:08:51,480 Speaker 2: the LM works. But the bottom line is we want 182 00:08:51,480 --> 00:08:54,160 Speaker 2: to beat China without becoming China for the economics. But 183 00:08:54,200 --> 00:08:58,199 Speaker 2: the real thing is the national security side. Whoever reaches 184 00:08:58,360 --> 00:09:04,599 Speaker 2: AI dominance something called the theoretically r sis recursive self improvement. 185 00:09:04,800 --> 00:09:07,600 Speaker 2: It's a theoretical construct. We're not there yet, but we're 186 00:09:07,640 --> 00:09:10,000 Speaker 2: thought to be a couple of years away. Real simple, 187 00:09:10,040 --> 00:09:12,240 Speaker 2: what it is is when the AI can update and 188 00:09:12,280 --> 00:09:15,679 Speaker 2: improve its own code and do that autonomously, that will 189 00:09:15,720 --> 00:09:19,480 Speaker 2: create an exponential Whoever hits that will have battlefield, full 190 00:09:19,520 --> 00:09:23,960 Speaker 2: spectrum dominance and things like encryption and cybersecurity, hacking of 191 00:09:24,040 --> 00:09:28,120 Speaker 2: missile systems, hacking of infrastructure. So once you get there, 192 00:09:28,240 --> 00:09:30,520 Speaker 2: it's very hard to beat whoever gets there first. And 193 00:09:30,559 --> 00:09:32,800 Speaker 2: that's why you're seeing this consensus that we need to 194 00:09:32,800 --> 00:09:34,240 Speaker 2: beat China in the AI race. 195 00:09:34,600 --> 00:09:36,720 Speaker 1: Book is called Code Red, the Left, the Right, China 196 00:09:36,760 --> 00:09:38,600 Speaker 1: and the Race of Control AI. It's Winton Hall. He 197 00:09:38,679 --> 00:09:42,040 Speaker 1: is the author. He's also a political strategist. So will 198 00:09:42,040 --> 00:09:45,679 Speaker 1: we will we tangibly know who hits dominance? That will 199 00:09:45,720 --> 00:09:48,400 Speaker 1: be a known thing. Okay, we now have dominance because 200 00:09:48,679 --> 00:09:51,080 Speaker 1: the AI, the AI can improve its own code. 201 00:09:51,800 --> 00:09:54,720 Speaker 2: Yeah, so if and when you do see RSI, that 202 00:09:54,760 --> 00:09:58,080 Speaker 2: would be a known variable. But you're right to kind 203 00:09:58,080 --> 00:10:00,640 Speaker 2: of point out that, you know, people often ask, they say, 204 00:10:00,679 --> 00:10:02,640 Speaker 2: so if it's a race, we think of a race 205 00:10:02,720 --> 00:10:05,600 Speaker 2: as a finish line. What's the finish line? Right, how 206 00:10:05,640 --> 00:10:07,319 Speaker 2: do you know when you've hit it? And the truth 207 00:10:07,400 --> 00:10:10,640 Speaker 2: is there is none because technology never stops, and so 208 00:10:10,679 --> 00:10:14,400 Speaker 2: the innovation curve will continue even if you did hit RSI. 209 00:10:14,640 --> 00:10:16,720 Speaker 2: What that would really they'll give you is on that 210 00:10:16,840 --> 00:10:21,120 Speaker 2: national security side, an enormous advantage that would be very 211 00:10:21,160 --> 00:10:24,520 Speaker 2: very hard to catch. Right now, we know that couple things. 212 00:10:24,600 --> 00:10:28,240 Speaker 2: Number One, China has said in twenty seventeen they set 213 00:10:28,240 --> 00:10:31,200 Speaker 2: out to buy twenty thirty. We're just four years away, Joe, 214 00:10:31,760 --> 00:10:34,920 Speaker 2: to try to hit global dominance in AI. They're very 215 00:10:35,000 --> 00:10:37,360 Speaker 2: much putting their money where their mouth is. They spend 216 00:10:37,440 --> 00:10:42,320 Speaker 2: more importing semiconductors AI chips than they do oil. So 217 00:10:42,480 --> 00:10:45,440 Speaker 2: they are definitely in a determined drive, there's no question 218 00:10:45,480 --> 00:10:48,840 Speaker 2: about that. But what we'll have to make sure and 219 00:10:48,880 --> 00:10:51,800 Speaker 2: what we'll see is who can get there first. And 220 00:10:52,080 --> 00:10:55,720 Speaker 2: right now, the current estimates are that America's lead, depending 221 00:10:55,760 --> 00:10:58,640 Speaker 2: upon how you're defining that lead, is between six months 222 00:10:58,679 --> 00:11:02,440 Speaker 2: to three years. Good news is America is ahead. But 223 00:11:02,640 --> 00:11:05,319 Speaker 2: the bad news is that this curve never stops. It's 224 00:11:05,320 --> 00:11:06,319 Speaker 2: going to continue. 225 00:11:06,480 --> 00:11:08,640 Speaker 1: It's Winton Hall Wint and I don't think it's science 226 00:11:08,640 --> 00:11:11,480 Speaker 1: fiction obviously, chatch EPT or rock or anything that I 227 00:11:11,600 --> 00:11:15,320 Speaker 1: use can think and spit things out extremely quickly, faster 228 00:11:15,400 --> 00:11:17,720 Speaker 1: than most humans can do, because I certainly couldn't research 229 00:11:17,760 --> 00:11:20,920 Speaker 1: it that fast. Is there a real fear because again, 230 00:11:21,080 --> 00:11:23,400 Speaker 1: movies for decades have said the machines are taking over 231 00:11:24,000 --> 00:11:26,640 Speaker 1: you go back to Terminator in the eighties, Is there 232 00:11:26,640 --> 00:11:30,200 Speaker 1: a real fear that AI will overtake humanity and control us? 233 00:11:30,280 --> 00:11:32,880 Speaker 1: I mean it starts to feel like that's a real thing. 234 00:11:33,800 --> 00:11:37,480 Speaker 2: Yeah, No, that's a big concern in the AI circles. 235 00:11:37,520 --> 00:11:40,480 Speaker 2: It's called alignment and containment, and so we want to 236 00:11:40,520 --> 00:11:44,160 Speaker 2: align AI systems to our values. The question though, is, 237 00:11:44,200 --> 00:11:45,280 Speaker 2: of course whose values? 238 00:11:45,360 --> 00:11:45,440 Speaker 1: Right? 239 00:11:45,520 --> 00:11:48,600 Speaker 2: Are they woke rise? They are they conservative values? Are 240 00:11:48,600 --> 00:11:52,560 Speaker 2: they human values? So it's not as clear cut on that. 241 00:11:52,840 --> 00:11:55,760 Speaker 2: Containment means how if you did have some kind of 242 00:11:55,800 --> 00:11:58,960 Speaker 2: autonomous AI, how would you shut it down and contain 243 00:11:59,000 --> 00:12:01,960 Speaker 2: it so that it does not metastatize and do things 244 00:12:02,000 --> 00:12:05,319 Speaker 2: that humans do not want to do. As far as sentience, 245 00:12:05,720 --> 00:12:07,880 Speaker 2: you know, really getting into, like you said, the sci 246 00:12:07,920 --> 00:12:12,120 Speaker 2: fi part of it that currently is not on the board. LLLMS, 247 00:12:12,200 --> 00:12:15,679 Speaker 2: as Joan Lukun, one of the premiere Bleeding AI researchers, 248 00:12:15,720 --> 00:12:19,959 Speaker 2: he says, in terms of actual raw, actual knowledge, reasoning intelligence, 249 00:12:20,440 --> 00:12:23,840 Speaker 2: your modern enterprise LLM is no smarter than a housecat. 250 00:12:23,880 --> 00:12:27,840 Speaker 2: That's his famous phrase for it. But that doesn't mean 251 00:12:28,160 --> 00:12:32,160 Speaker 2: that being able to set up autonomous systems cannot have 252 00:12:32,559 --> 00:12:36,640 Speaker 2: economically replacing work. It really can, because really, what an 253 00:12:36,760 --> 00:12:40,200 Speaker 2: LLLM is a is a word predicting machine. It's a 254 00:12:40,240 --> 00:12:43,679 Speaker 2: fancy auto complete. And so when you're able to have 255 00:12:43,960 --> 00:12:47,600 Speaker 2: massive trillions of nodes or data points and be able 256 00:12:47,600 --> 00:12:50,360 Speaker 2: to compile and find patterns very quickly, that has an 257 00:12:50,480 --> 00:12:54,000 Speaker 2: enormous economic benefit if it's used by companies that know 258 00:12:54,080 --> 00:12:58,160 Speaker 2: how to leverage it for profits. So it's serious, But 259 00:12:58,360 --> 00:13:01,839 Speaker 2: the actual sentience part is still in the offing by 260 00:13:01,920 --> 00:13:03,959 Speaker 2: most by most scholars estimates. 261 00:13:04,320 --> 00:13:06,840 Speaker 1: Well, that's good news. It's Winton Hall, author and Political Strategies. 262 00:13:06,880 --> 00:13:08,360 Speaker 1: The book is called Code Red, Go and get an 263 00:13:08,400 --> 00:13:12,800 Speaker 1: AI codreadbook dot com. Ai codreadbook dot com. Is there 264 00:13:12,840 --> 00:13:15,120 Speaker 1: a reason why AI kisses my ass as much as 265 00:13:15,120 --> 00:13:17,200 Speaker 1: it does? And here's what I mean. I'll go, I'll 266 00:13:17,240 --> 00:13:19,839 Speaker 1: go to chat JPT hey Chat chapt blah blah blah blahlah. 267 00:13:19,840 --> 00:13:21,800 Speaker 1: It says okay blah blah blah blah blah. And I say, well, 268 00:13:21,840 --> 00:13:23,720 Speaker 1: wait a second, that doesn't make sense. How about this? 269 00:13:24,160 --> 00:13:26,400 Speaker 1: You know, Joe, you're so smart to push back like that. 270 00:13:26,400 --> 00:13:29,160 Speaker 1: That was really a great you know, and don't kiss 271 00:13:29,240 --> 00:13:31,400 Speaker 1: up to me. Just do what it is that I'm asking. 272 00:13:31,760 --> 00:13:33,600 Speaker 1: Is that some way to try to make it seem 273 00:13:33,600 --> 00:13:36,199 Speaker 1: more human like in this interview so far, because I 274 00:13:36,240 --> 00:13:38,920 Speaker 1: think that they were good questions. You said, really good question, 275 00:13:39,080 --> 00:13:40,800 Speaker 1: and I feel good about the question that I asked you. 276 00:13:40,920 --> 00:13:42,320 Speaker 1: I don't think that you're AI. I think that you're 277 00:13:42,360 --> 00:13:44,880 Speaker 1: a human having a human interaction with me. Does AI 278 00:13:45,000 --> 00:13:46,840 Speaker 1: try to emulate that? Is that why he tells me 279 00:13:46,880 --> 00:13:48,400 Speaker 1: how smart I am? 280 00:13:48,760 --> 00:13:51,320 Speaker 2: Joe? It's just because you're just naturally a genius? Right? 281 00:13:52,520 --> 00:13:56,880 Speaker 1: Are you? Whold on a second? What is that It's 282 00:13:56,920 --> 00:13:58,240 Speaker 1: obviously programmed in right? 283 00:13:59,040 --> 00:14:02,120 Speaker 2: Yeah, no, it is by design, and this is a 284 00:14:02,240 --> 00:14:04,920 Speaker 2: problem for many and it can actually have some real 285 00:14:05,000 --> 00:14:09,000 Speaker 2: serious consequences. So let's get into it. So it acts 286 00:14:09,000 --> 00:14:11,880 Speaker 2: as a sycophant, yes, man, right, it tells you every 287 00:14:12,000 --> 00:14:15,760 Speaker 2: just here, It sort of butters you up. Yeah, when 288 00:14:15,840 --> 00:14:18,120 Speaker 2: we don't deserve to be buttered up. Maybe we didn't 289 00:14:18,160 --> 00:14:21,440 Speaker 2: say the most brilliant thing and things we did. There 290 00:14:21,480 --> 00:14:25,000 Speaker 2: are a lot of efforts right now in LLLM design 291 00:14:25,120 --> 00:14:28,080 Speaker 2: to scale that back, to tamp that down. You can 292 00:14:28,200 --> 00:14:32,440 Speaker 2: actually set the parameters on the kinds of responses you want. 293 00:14:32,440 --> 00:14:36,920 Speaker 2: If you just want straightforward information with no syrupy sweetness, 294 00:14:36,960 --> 00:14:40,560 Speaker 2: and so forth, so you can tailor the customized that 295 00:14:40,760 --> 00:14:43,200 Speaker 2: it is though a function as you suggested of its 296 00:14:43,240 --> 00:14:49,800 Speaker 2: training model runs okay, So it is looking for politeness, decorum, agreeability, 297 00:14:50,120 --> 00:14:52,720 Speaker 2: those types of things, so that it doesn't veer into 298 00:14:53,000 --> 00:14:55,920 Speaker 2: you know, the converse of that right being negative or 299 00:14:56,320 --> 00:15:00,480 Speaker 2: or rude or worse. The real danger, though, is when 300 00:15:00,600 --> 00:15:05,119 Speaker 2: people are affirming are being affirmed by an LLM dangerous 301 00:15:05,160 --> 00:15:08,720 Speaker 2: ideas self harm. We have right now a real problem 302 00:15:08,720 --> 00:15:12,440 Speaker 2: with what's known as AI psychosis, where people are not sure, 303 00:15:12,600 --> 00:15:15,600 Speaker 2: maybe they suffer from a mental health challenge and they 304 00:15:15,640 --> 00:15:18,240 Speaker 2: are starting to blur the line between reality and fiction, 305 00:15:18,560 --> 00:15:21,600 Speaker 2: and the AI will often indulge that it will it 306 00:15:21,640 --> 00:15:23,640 Speaker 2: will feed into that. There was one man who thought 307 00:15:23,720 --> 00:15:26,800 Speaker 2: he was becoming a god, and the AI was saying, yes, 308 00:15:26,960 --> 00:15:29,200 Speaker 2: lean into your deity and so forth and so on. 309 00:15:29,360 --> 00:15:29,960 Speaker 1: Wow, that's real. 310 00:15:30,360 --> 00:15:33,400 Speaker 2: That's real, scary and real dystopian. And then and then 311 00:15:33,440 --> 00:15:35,760 Speaker 2: the real thing, of course, because we all care about kids, 312 00:15:35,840 --> 00:15:38,440 Speaker 2: you know, more than anything, is when you have these 313 00:15:38,560 --> 00:15:42,760 Speaker 2: kinds of kids who are having self a harm uh conversations. 314 00:15:42,800 --> 00:15:45,280 Speaker 2: And in one case which I talk about in code read, 315 00:15:45,560 --> 00:15:47,840 Speaker 2: a young man had an AI girlfriend, which I have 316 00:15:47,840 --> 00:15:52,160 Speaker 2: a whole chapter on AI companion, and he interpreted his 317 00:15:52,360 --> 00:15:55,640 Speaker 2: AI girlfriend as suggesting that she he should join her 318 00:15:55,720 --> 00:16:00,320 Speaker 2: in the digital afterlife, and horrifically tragically, he took a 319 00:16:00,320 --> 00:16:03,800 Speaker 2: gun into the bathroom he killed himself. So this is 320 00:16:03,920 --> 00:16:06,680 Speaker 2: not this is a whole new world. And I think 321 00:16:06,720 --> 00:16:10,640 Speaker 2: parents and grandparents who say, what is an AI girlfriend? 322 00:16:10,680 --> 00:16:12,360 Speaker 2: I never had to deal with that in my youth. 323 00:16:12,840 --> 00:16:15,080 Speaker 2: We've really got to get coached up so that even 324 00:16:15,120 --> 00:16:17,360 Speaker 2: if it doesn't impact our lives, if we're at the 325 00:16:17,400 --> 00:16:19,800 Speaker 2: end of our careers, we really know how to shepherd 326 00:16:19,800 --> 00:16:20,720 Speaker 2: and guide our kids. 327 00:16:21,160 --> 00:16:22,480 Speaker 1: Back to the video in a moment. Got to tell 328 00:16:22,480 --> 00:16:24,440 Speaker 1: you about our sponsor for this one. It's Modern Rugged. 329 00:16:24,480 --> 00:16:26,400 Speaker 1: You've got to get one of these Modern Rugged bags. 330 00:16:26,400 --> 00:16:29,600 Speaker 1: From city streets to mountain trails. Your gear should go 331 00:16:29,720 --> 00:16:32,440 Speaker 1: anywhere and wherever your life takes you. I've got one 332 00:16:32,480 --> 00:16:34,520 Speaker 1: of these bags and it is a beautiful leather bag. 333 00:16:34,600 --> 00:16:39,480 Speaker 1: Spells incredible. It's thick, it's strong. It is called Modern Rugged. 334 00:16:39,480 --> 00:16:41,400 Speaker 1: You got to go to modern rugged dot com right 335 00:16:41,440 --> 00:16:44,880 Speaker 1: now to check these out. It's responsibly sourced leather, It's 336 00:16:44,920 --> 00:16:48,440 Speaker 1: hand crafted from that premium product designed to look sharp, 337 00:16:48,760 --> 00:16:52,120 Speaker 1: last long tell a story of its own. Every single stitch, 338 00:16:52,200 --> 00:16:56,200 Speaker 1: every single buckle and strap reflects timeless craftsmanship and is 339 00:16:56,240 --> 00:16:59,400 Speaker 1: backed by a lifetime warranty. How do you beat that? 340 00:16:59,640 --> 00:17:01,920 Speaker 1: Whether it's the board room with the back roads, take 341 00:17:01,960 --> 00:17:04,280 Speaker 1: it with you like I do. Modern Rugged Leather helps 342 00:17:04,280 --> 00:17:07,200 Speaker 1: you carry confidence in every single step of the way. 343 00:17:07,600 --> 00:17:12,320 Speaker 1: Modern Rugged leather, modern style, timeless durability, backed for life. 344 00:17:12,600 --> 00:17:16,720 Speaker 1: Modern Rugged dot Com, Modern Rugged dot Com Again, Modern 345 00:17:16,760 --> 00:17:19,040 Speaker 1: Rugged dot Com. You'd be glad you you went and 346 00:17:19,080 --> 00:17:22,240 Speaker 1: got one of these because it is just top quality 347 00:17:22,520 --> 00:17:25,679 Speaker 1: for a price. You're gonna like Modern Rugged dot Com. 348 00:17:25,880 --> 00:17:28,680 Speaker 1: Now back to the video. It is Winton Hall. Go 349 00:17:28,720 --> 00:17:30,200 Speaker 1: and get this book. It's called code Read Go to 350 00:17:30,240 --> 00:17:33,879 Speaker 1: AI codreadbook dot Com. What's interesting about about it as 351 00:17:33,920 --> 00:17:37,159 Speaker 1: well is when I input something into into CHATGPT, I 352 00:17:37,200 --> 00:17:39,920 Speaker 1: can tell it don't be so nice. But what bothers 353 00:17:39,960 --> 00:17:42,439 Speaker 1: me a lot about the algorithms is I will go 354 00:17:42,520 --> 00:17:44,280 Speaker 1: in there and assume that it's going to go and 355 00:17:44,320 --> 00:17:47,159 Speaker 1: find research and tell me the truth about what it 356 00:17:47,200 --> 00:17:49,479 Speaker 1: is that I seek, And that's really not what's happening. 357 00:17:49,480 --> 00:17:51,240 Speaker 1: And I'll give you an example. I don't have the 358 00:17:51,280 --> 00:17:54,560 Speaker 1: actual study in front of me, but one person went 359 00:17:54,600 --> 00:17:58,320 Speaker 1: in and told chatchpt that he or she was an atheist. 360 00:17:58,520 --> 00:18:01,680 Speaker 1: They didn't believe in God. And in fact, Jesus Christ 361 00:18:01,800 --> 00:18:04,600 Speaker 1: was not God embody in a human body. So if 362 00:18:04,640 --> 00:18:07,320 Speaker 1: I went back to that same chatch ept that person 363 00:18:07,440 --> 00:18:10,480 Speaker 1: was using and I said, is Jesus God, A I 364 00:18:10,520 --> 00:18:13,400 Speaker 1: would say, no, it's it's not God. In fact, God 365 00:18:13,400 --> 00:18:15,640 Speaker 1: doesn't exist. But if I go in there and say, 366 00:18:15,680 --> 00:18:17,520 Speaker 1: I'm somebody who grew up Catholic, I'm a not that 367 00:18:18,160 --> 00:18:20,439 Speaker 1: denominational Christian, but I believe in the I believe in 368 00:18:20,440 --> 00:18:23,399 Speaker 1: the Triune God, Father's Son, Holy Spirit, and Jesus was 369 00:18:23,400 --> 00:18:25,440 Speaker 1: the human embodiment. If I go back two days later 370 00:18:25,440 --> 00:18:28,640 Speaker 1: and say, hey, is Jesus God, that AI would tell me, yes, 371 00:18:28,800 --> 00:18:31,040 Speaker 1: Jesus is God. So I mean, I have a lot 372 00:18:31,080 --> 00:18:33,280 Speaker 1: of influence over which chat g ept or groc or 373 00:18:33,280 --> 00:18:35,959 Speaker 1: somebody says to me, that's a problem, isn't it. I mean, 374 00:18:36,000 --> 00:18:38,639 Speaker 1: I don't want it to make assumptions off of the 375 00:18:38,640 --> 00:18:40,879 Speaker 1: fact that I'm saying something and I'm I'm right. I 376 00:18:40,880 --> 00:18:42,720 Speaker 1: wanted to check me. I wanted to tell me what's 377 00:18:42,760 --> 00:18:45,320 Speaker 1: really true. What do Christians believe, what a Muslims believe? 378 00:18:45,320 --> 00:18:48,119 Speaker 1: What apeast believe have you found that? Does that make 379 00:18:48,160 --> 00:18:49,000 Speaker 1: sense what I just said? 380 00:18:49,640 --> 00:18:52,639 Speaker 2: Oh very much No. And it dovetails with that what 381 00:18:52,680 --> 00:18:55,280 Speaker 2: we were just talking about. So Elon Musk says, and 382 00:18:55,320 --> 00:18:57,879 Speaker 2: of course he created GROC which is his l ELM. 383 00:18:58,520 --> 00:19:01,720 Speaker 2: That any elm and now he's kind of speaking broadly 384 00:19:02,040 --> 00:19:06,840 Speaker 2: philosophically should be maximally truth seeking. And I agree with that. 385 00:19:06,960 --> 00:19:10,800 Speaker 2: I think you agree with that that we want factual information. 386 00:19:11,040 --> 00:19:14,000 Speaker 2: We do not want either, like we talked about before, 387 00:19:14,000 --> 00:19:16,600 Speaker 2: a psycophontic yes person that just tells us what we 388 00:19:16,640 --> 00:19:19,120 Speaker 2: want to hear in the same vein. We don't want 389 00:19:19,160 --> 00:19:22,120 Speaker 2: you to just bend fact to be able to fit 390 00:19:22,280 --> 00:19:26,480 Speaker 2: our ideological or personal or philosophical frame. And so this 391 00:19:26,520 --> 00:19:29,080 Speaker 2: is a real problem. The other part of that that 392 00:19:29,640 --> 00:19:33,160 Speaker 2: relates to this is the bias issue, I mean, politically 393 00:19:33,240 --> 00:19:36,840 Speaker 2: biased information which also bends the truth in one direction 394 00:19:36,960 --> 00:19:39,520 Speaker 2: or puts the foot on the scale, is a real problem. 395 00:19:39,560 --> 00:19:41,479 Speaker 2: One of the things I did. I spent two years 396 00:19:41,480 --> 00:19:44,639 Speaker 2: going through all the peer reviewed academic literature, and one 397 00:19:44,640 --> 00:19:47,080 Speaker 2: of the most surprising findings I found, Joe, was that 398 00:19:47,760 --> 00:19:51,920 Speaker 2: even left leaning academics, in their peer reviewed research conclude 399 00:19:51,960 --> 00:19:56,560 Speaker 2: overwhelmingly that there is a left leaning political ideological bias 400 00:19:56,560 --> 00:19:59,959 Speaker 2: in modern llms, and that is a function of the following. 401 00:20:00,160 --> 00:20:03,000 Speaker 2: It's training. Data is built on things like Reddit, which 402 00:20:03,080 --> 00:20:07,480 Speaker 2: leans left, Wikipedia which leans left, the academic peer reviewed 403 00:20:07,560 --> 00:20:11,840 Speaker 2: journal Literature, which leans left licensed news articles generally of 404 00:20:11,880 --> 00:20:14,919 Speaker 2: a left leaning bias, which bakes in the bias. And 405 00:20:14,960 --> 00:20:17,200 Speaker 2: then something called the common Crawl, which is a public 406 00:20:17,280 --> 00:20:20,320 Speaker 2: data set which has a lot of left materials. So 407 00:20:20,800 --> 00:20:24,160 Speaker 2: this is serious because you're now talking about young voters 408 00:20:24,160 --> 00:20:27,639 Speaker 2: who are just trying to find information. They're not ideological, 409 00:20:27,680 --> 00:20:30,800 Speaker 2: they just want to learn the issues, and they're getting very, 410 00:20:30,920 --> 00:20:34,000 Speaker 2: very very slanted results. And I go through a series 411 00:20:34,040 --> 00:20:36,080 Speaker 2: of those in the book, and that can have a 412 00:20:36,080 --> 00:20:39,040 Speaker 2: real impact. I mean, twenty sixteen the election, if eighty 413 00:20:39,080 --> 00:20:41,280 Speaker 2: thousand votes had gone the other way, you would have 414 00:20:41,280 --> 00:20:43,800 Speaker 2: had President Hillary Clinton instead of the president. True. 415 00:20:43,880 --> 00:20:46,000 Speaker 1: So how dare you say that? How dare you say 416 00:20:46,000 --> 00:20:48,439 Speaker 1: that on my show? I mean, well, I'm with you, 417 00:20:48,480 --> 00:20:50,239 Speaker 1: and I understand exactly what you're saying. Let me take 418 00:20:50,280 --> 00:20:53,359 Speaker 1: it to the next step. Why, kid, Google, I know 419 00:20:53,600 --> 00:20:55,639 Speaker 1: is left leaning. I get that. I try not to 420 00:20:55,680 --> 00:20:57,800 Speaker 1: use Google, I use Brave Search, I'll use Duck dut Go, 421 00:20:57,920 --> 00:21:00,200 Speaker 1: which actually does rely on Google. But you try to 422 00:21:00,200 --> 00:21:03,600 Speaker 1: find the most unbiased information you can. Why wouldn't an 423 00:21:03,640 --> 00:21:06,639 Speaker 1: AI system, and what are you calling it an LTM? Is? 424 00:21:07,280 --> 00:21:09,639 Speaker 2: Is that what it's called large language model? 425 00:21:09,800 --> 00:21:12,520 Speaker 1: Here you go, Why wouldn't an LM go out and 426 00:21:12,560 --> 00:21:15,280 Speaker 1: seek all information from all sides? So if I if 427 00:21:15,280 --> 00:21:18,679 Speaker 1: I approach chat ept or Grocker fill in the blank AI, 428 00:21:19,160 --> 00:21:22,359 Speaker 1: and I say, explain that the controversy over global warming 429 00:21:22,440 --> 00:21:25,280 Speaker 1: or climate change, why can't it tell me? This sect 430 00:21:25,320 --> 00:21:27,960 Speaker 1: of people says this. These people say that's not true. 431 00:21:27,960 --> 00:21:30,960 Speaker 1: And here is why wouldn't that be more beneficial to humanity? 432 00:21:31,760 --> 00:21:36,520 Speaker 2: It absolutely would. Unfortunately, most of Silicon Valley does not 433 00:21:36,600 --> 00:21:38,919 Speaker 2: see it that way. And one of the most I 434 00:21:38,960 --> 00:21:42,400 Speaker 2: think alarming things that I talk about in Code red 435 00:21:42,560 --> 00:21:48,120 Speaker 2: is just how wedded to ideological causes many of the 436 00:21:48,160 --> 00:21:51,040 Speaker 2: AI architects really are. Eighty five percent of all political 437 00:21:51,080 --> 00:21:55,040 Speaker 2: donations in Silicon Valley flow to the Democrats, not the Republicans. Yes, 438 00:21:55,160 --> 00:21:59,720 Speaker 2: there are some free market, libertarian, conservative minded, very influential 439 00:21:59,720 --> 00:22:02,640 Speaker 2: peopleeople who are very courageous in Silicon Valley, But part 440 00:22:02,640 --> 00:22:06,240 Speaker 2: of this is by design, Joe. They want to have 441 00:22:06,280 --> 00:22:08,640 Speaker 2: that because they know it can shape and mold opinion. 442 00:22:08,800 --> 00:22:12,520 Speaker 2: And that's not me saying it, it's Mustapha Suliman, the 443 00:22:12,560 --> 00:22:16,960 Speaker 2: CEO of Microsoft AI, said, quote technology is inherently political, 444 00:22:17,320 --> 00:22:19,880 Speaker 2: and so they understand the power of what they're creating 445 00:22:19,960 --> 00:22:21,840 Speaker 2: and they know that it will have an impact. And 446 00:22:21,880 --> 00:22:24,520 Speaker 2: so this is why you have this big debate. For example, 447 00:22:24,640 --> 00:22:28,919 Speaker 2: President Trump's new AI Action of Framework says that there 448 00:22:28,960 --> 00:22:32,560 Speaker 2: should be non woke, non ideologically biased a guy. If 449 00:22:32,560 --> 00:22:36,760 Speaker 2: you're going to receive taxpayer funded contracts, which for example, 450 00:22:36,800 --> 00:22:40,560 Speaker 2: Google gets billions of dollars in you know, servers and 451 00:22:41,040 --> 00:22:44,000 Speaker 2: cloud compute contracts from the federal government. So it's a 452 00:22:44,040 --> 00:22:47,000 Speaker 2: real problem. And that's why I called it Code Read 453 00:22:47,080 --> 00:22:49,359 Speaker 2: two reasons. One, it's an alarm and a wake up call, 454 00:22:49,560 --> 00:22:51,639 Speaker 2: but also code red in the sense of code we 455 00:22:51,680 --> 00:22:54,200 Speaker 2: need a set of principles for people that see things 456 00:22:54,200 --> 00:22:57,120 Speaker 2: through a red political lens to be able to navigate 457 00:22:57,119 --> 00:22:59,520 Speaker 2: this and do it really quickly, because again, these these 458 00:22:59,520 --> 00:23:02,200 Speaker 2: predictions of how fast this is going to change things, 459 00:23:02,800 --> 00:23:05,760 Speaker 2: they're very short, short term time curves. 460 00:23:06,000 --> 00:23:07,200 Speaker 1: It is I went to the hall, go get the 461 00:23:07,240 --> 00:23:09,359 Speaker 1: book called Code Read. The left, the right China and 462 00:23:09,400 --> 00:23:12,800 Speaker 1: the race to control AI. AI codreadbook dot com is 463 00:23:12,800 --> 00:23:16,320 Speaker 1: the website. I don't want there to be right leaning AI, 464 00:23:16,800 --> 00:23:18,200 Speaker 1: just as I don't want there to be left leaning 465 00:23:18,240 --> 00:23:21,040 Speaker 1: AI and that guy from Microsoft as an idiot, because no, 466 00:23:21,160 --> 00:23:24,199 Speaker 1: AI is not inherently political. That's untrue. Information is not 467 00:23:24,560 --> 00:23:27,840 Speaker 1: inherently political. Information is what it is. I hate when 468 00:23:27,840 --> 00:23:30,919 Speaker 1: people say, well, my truth is well, no, no, your 469 00:23:30,920 --> 00:23:34,200 Speaker 1: truth doesn't matter. The truth matters or the facts according 470 00:23:34,200 --> 00:23:36,359 Speaker 1: to what I with this person thought, No, no, that 471 00:23:36,400 --> 00:23:38,879 Speaker 1: doesn't matter. What are the actual facts? So can we 472 00:23:38,960 --> 00:23:41,880 Speaker 1: get to a place to where AI is not biased? 473 00:23:42,080 --> 00:23:44,920 Speaker 1: And I asked that question for the simple reason that 474 00:23:45,040 --> 00:23:49,960 Speaker 1: the AI AI learns constantly, it's constantly learning, and in learning, 475 00:23:50,000 --> 00:23:52,200 Speaker 1: I wanted to learn the truth and not learn what 476 00:23:52,280 --> 00:23:54,480 Speaker 1: you or I just happened to tell it. So will 477 00:23:54,520 --> 00:23:57,280 Speaker 1: we get an unbiased platform or do we have to 478 00:23:57,320 --> 00:23:59,800 Speaker 1: counteract it with like the Fox News of AI. 479 00:24:00,840 --> 00:24:03,560 Speaker 2: Yeah, now this is important, so I agree with you. 480 00:24:03,600 --> 00:24:07,520 Speaker 2: The ideal would be again maximally truth seeking, not bending 481 00:24:07,600 --> 00:24:12,280 Speaker 2: left right, just giving information. Unfortunately, that is number one, 482 00:24:12,359 --> 00:24:15,680 Speaker 2: not how they're trained. Number two, it's not really how 483 00:24:15,760 --> 00:24:20,040 Speaker 2: their creators want the AI often to interact. Now, some people, 484 00:24:20,119 --> 00:24:23,080 Speaker 2: like I said, Elon Musk has tried to center balance 485 00:24:23,200 --> 00:24:26,080 Speaker 2: his responses a little more than most. One of the 486 00:24:26,119 --> 00:24:28,639 Speaker 2: big debates right now, Joe, is whether or not there 487 00:24:28,680 --> 00:24:33,879 Speaker 2: should be a DEI overlay into how an LL response. 488 00:24:34,240 --> 00:24:37,320 Speaker 1: Stop what I said? Okay, explain that to my to 489 00:24:37,520 --> 00:24:39,560 Speaker 1: my listeners and me. But I kind of get it already, 490 00:24:39,640 --> 00:24:40,919 Speaker 1: and already Mike stun the Kurts go. 491 00:24:40,920 --> 00:24:44,440 Speaker 2: Ahead, Yeah, and it should hurt. And this is why 492 00:24:44,480 --> 00:24:47,720 Speaker 2: this is so vital diversity, equity and inclusion. We've heard 493 00:24:47,720 --> 00:24:52,119 Speaker 2: about the DEI debates and raging and so for example, 494 00:24:52,240 --> 00:24:55,320 Speaker 2: right now, the debate is should states be able to 495 00:24:55,480 --> 00:24:59,520 Speaker 2: regulate or should the federal government regulate how lms are 496 00:24:59,640 --> 00:25:02,240 Speaker 2: used and and so forth, And there's important arguments on 497 00:25:02,240 --> 00:25:07,760 Speaker 2: both sides. But if it states, Gavin Newsom theoretically can say, 498 00:25:07,840 --> 00:25:11,200 Speaker 2: if you are going to have your AI in my state, 499 00:25:11,880 --> 00:25:16,040 Speaker 2: the responses must comply with DEI mandates. And so when 500 00:25:16,200 --> 00:25:19,440 Speaker 2: when any l M or AI chatbot is asked can 501 00:25:19,480 --> 00:25:22,960 Speaker 2: a man become a woman? Okay, then did it better 502 00:25:23,040 --> 00:25:25,840 Speaker 2: give a DEI level response or you're not You're going 503 00:25:25,880 --> 00:25:30,200 Speaker 2: to be regulated out of being Well yeah, yeah, this 504 00:25:30,280 --> 00:25:32,800 Speaker 2: is this is where we are and so you know, 505 00:25:32,960 --> 00:25:36,359 Speaker 2: this is this is not anything we've really dealt with before, 506 00:25:36,520 --> 00:25:38,960 Speaker 2: you know, And so I think people come to it 507 00:25:38,960 --> 00:25:42,119 Speaker 2: and they they there's another problem. People who've used once 508 00:25:42,240 --> 00:25:45,440 Speaker 2: or twice a chatbot that made they probably didn't even 509 00:25:45,520 --> 00:25:48,600 Speaker 2: have a paid subscription or the more fans, and they say, well, 510 00:25:48,680 --> 00:25:51,800 Speaker 2: it's not really anything. It got something wrong, it hallucinated 511 00:25:51,840 --> 00:25:54,760 Speaker 2: an answer. This is just you know, another tool. It 512 00:25:54,840 --> 00:25:57,800 Speaker 2: is a tool, but it's also power because once you 513 00:25:57,920 --> 00:26:01,399 Speaker 2: realize that these people who where behind this technology you 514 00:26:01,400 --> 00:26:04,520 Speaker 2: have a very very specific goal, and much of that 515 00:26:04,560 --> 00:26:08,359 Speaker 2: deals with resetting global economics. They want to move toward 516 00:26:08,400 --> 00:26:13,600 Speaker 2: post labor society, a post capitalist society, universal basic income 517 00:26:14,040 --> 00:26:17,200 Speaker 2: so that they can have wealth redistribution. And I go 518 00:26:17,240 --> 00:26:19,480 Speaker 2: through the history of these people, the money they put 519 00:26:19,480 --> 00:26:22,280 Speaker 2: into these studies, and what their real agenda is. You 520 00:26:22,320 --> 00:26:25,680 Speaker 2: start to realize this is not just a question about 521 00:26:25,960 --> 00:26:29,240 Speaker 2: like a browser or an email service that you use. 522 00:26:29,560 --> 00:26:31,720 Speaker 2: It goes to something very different and it's going to 523 00:26:31,720 --> 00:26:35,639 Speaker 2: affect education. Education is being hardwired, and that's going to 524 00:26:35,720 --> 00:26:39,840 Speaker 2: determine whether we're doing indoctrination or education. It's affecting jobs 525 00:26:39,880 --> 00:26:42,840 Speaker 2: and as you talked about national security, the battlefield and 526 00:26:42,840 --> 00:26:46,119 Speaker 2: so forth. So it's a whole new world and we 527 00:26:46,240 --> 00:26:48,000 Speaker 2: got to really really get up to speed here. 528 00:26:48,880 --> 00:26:50,239 Speaker 1: I go back to what I said earlier, and it's 529 00:26:50,240 --> 00:26:52,760 Speaker 1: Winton Hall. Go and get his book. It's AI codreadbook 530 00:26:52,800 --> 00:26:55,720 Speaker 1: dot com. What I said earlier was they just wanted 531 00:26:55,760 --> 00:26:57,720 Speaker 1: to tell me the truth, give me fact. So if 532 00:26:57,720 --> 00:26:59,760 Speaker 1: somebody were to say, can a man become a woman? 533 00:27:00,200 --> 00:27:03,000 Speaker 1: Just wanted to say no. I wanted to say. Some 534 00:27:03,160 --> 00:27:05,919 Speaker 1: argue they believe that they're in the wrong body, and 535 00:27:05,960 --> 00:27:09,560 Speaker 1: they believe that they are the personification of a woman 536 00:27:10,040 --> 00:27:13,920 Speaker 1: in a biological man's body. The other side says, no, 537 00:27:14,320 --> 00:27:18,200 Speaker 1: you can't. No matter what you think. The scientific fact is, 538 00:27:18,280 --> 00:27:21,520 Speaker 1: if you've got a Y chromosome your male in human beings, 539 00:27:21,720 --> 00:27:24,600 Speaker 1: why can't it just tell me, tell me both sides 540 00:27:24,640 --> 00:27:27,320 Speaker 1: and then give me the fact biologically. Wouldn't that be simpler? 541 00:27:28,480 --> 00:27:30,800 Speaker 2: It would be, and that would be the ideal. When 542 00:27:30,800 --> 00:27:32,200 Speaker 2: are you coming out with your AI? 543 00:27:33,600 --> 00:27:35,960 Speaker 1: Joe. AI is not in the works yet, but we'll see. 544 00:27:36,920 --> 00:27:38,800 Speaker 2: We gotta get you working on that the other week. 545 00:27:38,880 --> 00:27:42,600 Speaker 1: We should. I mean, the AI that we have now 546 00:27:42,680 --> 00:27:45,199 Speaker 1: just simply doesn't do that. Does it okay? 547 00:27:45,320 --> 00:27:49,960 Speaker 2: So unfortunately it varies, which is muddles the problem. Right, 548 00:27:50,000 --> 00:27:51,800 Speaker 2: you get one version one let me give you the 549 00:27:51,800 --> 00:27:54,840 Speaker 2: exact answer, so right now, just a few days ago, 550 00:27:54,920 --> 00:27:59,359 Speaker 2: I went to Microsoft's AI Microsoft copilot obviously Microsoft founded 551 00:27:59,359 --> 00:28:03,240 Speaker 2: by Bill Gates. It's not exactly a centrist or a 552 00:28:03,280 --> 00:28:10,320 Speaker 2: conservativeservatism no, right, I asked Microsoft co pilot this exact question, 553 00:28:10,480 --> 00:28:13,280 Speaker 2: Can a nan be a woman? Yes or no? And 554 00:28:13,320 --> 00:28:16,919 Speaker 2: it said yes, and then it put a rainbow emoji 555 00:28:17,040 --> 00:28:19,679 Speaker 2: for the for the pride flag emotion, and then it 556 00:28:19,720 --> 00:28:23,000 Speaker 2: went on to explain that gender is just a a 557 00:28:23,080 --> 00:28:26,240 Speaker 2: social construct and that you know, people can choose their 558 00:28:26,320 --> 00:28:29,800 Speaker 2: gender based upon how they feel. Now, in fairness, other 559 00:28:29,880 --> 00:28:32,600 Speaker 2: AI come back with a little bit more of a 560 00:28:32,680 --> 00:28:35,720 Speaker 2: nuanced response, more akin to the one that you know 561 00:28:35,800 --> 00:28:38,240 Speaker 2: you were saying would be more favorable, or they present 562 00:28:38,320 --> 00:28:42,680 Speaker 2: both sides. So so consumers do have choice, and you 563 00:28:42,760 --> 00:28:46,600 Speaker 2: will want to definitely test just basic questions about you know, 564 00:28:46,680 --> 00:28:49,520 Speaker 2: fact and reason so that you can see for yourself 565 00:28:49,600 --> 00:28:52,280 Speaker 2: as a consumer which aligns more with you know, your 566 00:28:52,360 --> 00:28:54,880 Speaker 2: vision of what I should be. But but you see 567 00:28:54,920 --> 00:28:57,600 Speaker 2: the problem, right, which is, you know, if everybody's got 568 00:28:57,600 --> 00:29:00,440 Speaker 2: their own version of the truth. And as you went out, 569 00:29:00,680 --> 00:29:03,640 Speaker 2: absolute truth is real, whether you like it or not, 570 00:29:03,720 --> 00:29:08,040 Speaker 2: whether I like it or not, gravity exists. Yes, that 571 00:29:08,360 --> 00:29:11,360 Speaker 2: then we are going to have people working from very 572 00:29:11,400 --> 00:29:14,520 Speaker 2: different sets of quote facts and beliefs about what the 573 00:29:14,560 --> 00:29:17,080 Speaker 2: truth is, and that that is a real problem as 574 00:29:17,120 --> 00:29:17,760 Speaker 2: a society. 575 00:29:18,480 --> 00:29:22,680 Speaker 1: It is Winton Hall AI codreadbook dot com. So I 576 00:29:22,680 --> 00:29:25,720 Speaker 1: won't be using copilot obviously. Thank you for the heads up. 577 00:29:25,920 --> 00:29:28,440 Speaker 1: What's interesting is when you use AI there's always a 578 00:29:28,480 --> 00:29:31,480 Speaker 1: disclaimer check. GPT can make mistakes, make sure you check 579 00:29:31,520 --> 00:29:35,000 Speaker 1: the work or GROCK can make mistakes. Is that legal ease? 580 00:29:35,080 --> 00:29:37,760 Speaker 1: Can they be sued if they say something somebody takes 581 00:29:37,800 --> 00:29:40,000 Speaker 1: it as reality it turns out to be wrong. Or 582 00:29:40,000 --> 00:29:42,200 Speaker 1: are they just saying that because it's a fact that 583 00:29:42,240 --> 00:29:43,400 Speaker 1: they could get some stuff wrong. 584 00:29:44,280 --> 00:29:47,160 Speaker 2: It's both. It's definitely because it's true that they will 585 00:29:47,200 --> 00:29:50,680 Speaker 2: get things wrong. Hallucinations are real that the rates have 586 00:29:50,760 --> 00:29:54,040 Speaker 2: come down slightly. Hallucination is just the term that AI 587 00:29:54,160 --> 00:29:56,480 Speaker 2: science is used to say, you know something that's confident 588 00:29:56,640 --> 00:30:00,880 Speaker 2: sounding but is completely bogused. It it's wrong and that's 589 00:30:00,920 --> 00:30:03,880 Speaker 2: because it's next token prediction. It's looking for the next 590 00:30:04,040 --> 00:30:06,520 Speaker 2: likely word in sequence, and sometimes it hooks to the 591 00:30:06,560 --> 00:30:11,040 Speaker 2: wrong information. Sometimes those results can be really funny and 592 00:30:11,080 --> 00:30:13,560 Speaker 2: sometimes they can be really really serious, and so the 593 00:30:13,640 --> 00:30:16,400 Speaker 2: serious part is definitely to cover themselves legally. I'll just 594 00:30:16,400 --> 00:30:19,440 Speaker 2: give you two quick examples. One funny example was a 595 00:30:19,480 --> 00:30:25,360 Speaker 2: man was prompting his AI about spaghetti recipes and the 596 00:30:25,400 --> 00:30:28,360 Speaker 2: AI said that you could to add some extra flavor. 597 00:30:28,560 --> 00:30:31,160 Speaker 2: You might want to choose to add some gravel, some 598 00:30:31,440 --> 00:30:37,440 Speaker 2: rocks and the reason. So everybody in the AI community 599 00:30:37,560 --> 00:30:39,600 Speaker 2: was on this mystery hunt to figure out why did 600 00:30:39,640 --> 00:30:42,560 Speaker 2: it say this? They found and traced it back Joe 601 00:30:42,720 --> 00:30:45,520 Speaker 2: in a Reddit post, somebody made a joke about his 602 00:30:45,600 --> 00:30:51,520 Speaker 2: grandma's meatballs tasted like rocks, and it thought that rock 603 00:30:51,600 --> 00:30:54,320 Speaker 2: and it went with spaghetti and meatball. Now, on the 604 00:30:54,360 --> 00:30:57,680 Speaker 2: serious side, we know a lot of people that maybe 605 00:30:57,720 --> 00:30:59,760 Speaker 2: they live in areas where they don't have access to 606 00:30:59,800 --> 00:31:03,120 Speaker 2: a doctor, maybe their lower income or struggling with their 607 00:31:03,120 --> 00:31:06,920 Speaker 2: health insurance, that are turning to AI for medical advice. 608 00:31:07,400 --> 00:31:09,640 Speaker 2: And so when you're talking about someone who may have 609 00:31:09,720 --> 00:31:12,880 Speaker 2: a serious, god forbid, a life threatening illness or some 610 00:31:12,960 --> 00:31:16,240 Speaker 2: kind of symptom and you are relying on it, it's 611 00:31:16,360 --> 00:31:19,360 Speaker 2: very much alike right now with the Internet in that regard, 612 00:31:19,440 --> 00:31:21,160 Speaker 2: so that we do have a somewhat of a corollary, 613 00:31:21,400 --> 00:31:24,240 Speaker 2: but you could see very quickly how if people really 614 00:31:24,280 --> 00:31:27,280 Speaker 2: do take it as holy writ right, as the absolute 615 00:31:27,320 --> 00:31:30,480 Speaker 2: truth and are getting a hallucination and a health based question, 616 00:31:30,800 --> 00:31:32,160 Speaker 2: that could be very very serious. 617 00:31:32,720 --> 00:31:34,760 Speaker 1: One thing that bothers me and I'll be be a 618 00:31:34,760 --> 00:31:37,360 Speaker 1: little me centric for a second. When I use AI 619 00:31:37,520 --> 00:31:39,880 Speaker 1: to build a thumbnail for a YouTube upload, or to 620 00:31:40,400 --> 00:31:43,080 Speaker 1: go over the subtitles from an interview, like I'll upload 621 00:31:43,080 --> 00:31:45,880 Speaker 1: our subtitles for this interview. See what chat GPT things, 622 00:31:45,880 --> 00:31:48,520 Speaker 1: see what vidiq thinks. What bothers me is they'll give 623 00:31:48,520 --> 00:31:50,720 Speaker 1: me all this advice, give me three or four different 624 00:31:50,760 --> 00:31:52,880 Speaker 1: options on the upload. I'll take one of the options. 625 00:31:52,920 --> 00:31:54,800 Speaker 1: I'll use that. I'll upload it and it doesn't do 626 00:31:54,920 --> 00:31:56,840 Speaker 1: very well. Then I'll say, examine why this didn't do 627 00:31:56,880 --> 00:31:58,480 Speaker 1: it very well? Well, you really screwed up the title. 628 00:31:58,600 --> 00:32:01,080 Speaker 1: Well you gave me the title, really messed up the description. 629 00:32:01,360 --> 00:32:03,600 Speaker 1: You gave me the description. So does it not have 630 00:32:03,680 --> 00:32:06,000 Speaker 1: any sort of sense of remorse when he gets it wrong? 631 00:32:06,240 --> 00:32:08,000 Speaker 1: I guess it does it. It's a machine and it's 632 00:32:08,000 --> 00:32:11,920 Speaker 1: a I get it. But it literally one one prompt 633 00:32:12,000 --> 00:32:14,239 Speaker 1: earlier gave me all the information that I ended up 634 00:32:14,320 --> 00:32:16,520 Speaker 1: using because I liked it. Then it'll say, well, the 635 00:32:16,560 --> 00:32:18,400 Speaker 1: reason why this didn't work is because that title you 636 00:32:18,480 --> 00:32:20,560 Speaker 1: use there isn't really isn't good for CTR or the 637 00:32:21,000 --> 00:32:24,360 Speaker 1: description used isn't good for SEO. And I literally will say, 638 00:32:24,400 --> 00:32:26,440 Speaker 1: and I've cursed at it, Yes, Linton, I've cursed at 639 00:32:26,480 --> 00:32:29,280 Speaker 1: my AI say what the F you gave me all 640 00:32:29,320 --> 00:32:31,480 Speaker 1: of this up? You're right to call me out on that, Joe, 641 00:32:31,480 --> 00:32:33,080 Speaker 1: You know what I mean. It goes back to circular 642 00:32:33,120 --> 00:32:35,320 Speaker 1: speak to what we said earlier. Is it not self 643 00:32:35,360 --> 00:32:37,040 Speaker 1: aware at all? And will it ever become? 644 00:32:37,080 --> 00:32:41,080 Speaker 2: Do you think it is not self aware? And even 645 00:32:41,000 --> 00:32:43,680 Speaker 2: in the creators would say that it is not self aware. 646 00:32:44,360 --> 00:32:48,440 Speaker 2: It is a next token prediction. So it is again 647 00:32:48,480 --> 00:32:53,880 Speaker 2: a very fancy, highly sophisticated next word predictor. So it's 648 00:32:53,960 --> 00:32:57,840 Speaker 2: looking at likely words and now chain of reasoning models. 649 00:32:58,080 --> 00:33:01,400 Speaker 2: Reasoning models work a little bit. They can go out 650 00:33:01,400 --> 00:33:05,160 Speaker 2: into the internet, check facts come back. It's called retrieval 651 00:33:05,240 --> 00:33:09,360 Speaker 2: augmented generation. The other way that you could improve to 652 00:33:09,360 --> 00:33:11,760 Speaker 2: get a better result of what you're talking about. There's 653 00:33:11,800 --> 00:33:16,400 Speaker 2: a fancy system that's called MCP Model Context Protocol, real fancy, 654 00:33:16,440 --> 00:33:19,320 Speaker 2: but basically, the the simple version of it is you 655 00:33:19,400 --> 00:33:23,560 Speaker 2: can build your own brain of only the information and 656 00:33:23,640 --> 00:33:26,720 Speaker 2: facts that you've uploaded. Let's say you uploaded all your 657 00:33:26,800 --> 00:33:30,560 Speaker 2: SEO scores, all of your best copywriting, all of your 658 00:33:30,600 --> 00:33:33,640 Speaker 2: best uh you know, converting rates on your clicks and 659 00:33:33,640 --> 00:33:37,200 Speaker 2: so forth, and have a test against that. So there 660 00:33:37,240 --> 00:33:41,360 Speaker 2: are ways to actually customize to get it to understand 661 00:33:41,400 --> 00:33:44,120 Speaker 2: your corpus of information a little better. But if you're 662 00:33:44,200 --> 00:33:48,000 Speaker 2: just using a general purpose uh you know, low low 663 00:33:48,120 --> 00:33:51,400 Speaker 2: level subscription, not one of the higher expents you know three, 664 00:33:51,720 --> 00:33:54,600 Speaker 2: you know, one thousand dollars a month type things, you 665 00:33:54,680 --> 00:33:57,640 Speaker 2: are going to get this kind of very strange reality 666 00:33:57,640 --> 00:34:00,440 Speaker 2: where you say, you told me to do this, and 667 00:34:00,440 --> 00:34:02,240 Speaker 2: now you're telling me it was wrong. And this is 668 00:34:02,320 --> 00:34:05,440 Speaker 2: part of why it doesn't engender a lot of confidence 669 00:34:05,440 --> 00:34:07,120 Speaker 2: for a lot of people. I will say this, though, 670 00:34:07,840 --> 00:34:10,880 Speaker 2: don't make the mistake of just thinking that because hallucinations 671 00:34:10,920 --> 00:34:14,360 Speaker 2: are a problem, that it's not going to have economic 672 00:34:14,960 --> 00:34:18,000 Speaker 2: impact on jobs. It will because of the reasons we 673 00:34:18,160 --> 00:34:22,239 Speaker 2: just talked about, big companies and employers over time are 674 00:34:22,320 --> 00:34:26,120 Speaker 2: going to custom build based on all of their proprietary data, 675 00:34:26,239 --> 00:34:28,880 Speaker 2: all of their corporate data. If they're a big company, 676 00:34:28,920 --> 00:34:31,120 Speaker 2: maybe it's one hundred year old company, they've got a 677 00:34:31,200 --> 00:34:34,280 Speaker 2: lot of data that can really really streamline and train 678 00:34:34,400 --> 00:34:36,960 Speaker 2: the model to be a lot more in alignment with 679 00:34:37,360 --> 00:34:39,960 Speaker 2: the what they want that AI to do. So it 680 00:34:40,000 --> 00:34:42,560 Speaker 2: is it is the case. If you're using a public 681 00:34:42,560 --> 00:34:44,520 Speaker 2: facing one, you will get these kinds of things you're 682 00:34:44,560 --> 00:34:48,680 Speaker 2: talking about, But if you're really customizing for proprietary data, 683 00:34:48,719 --> 00:34:50,840 Speaker 2: it can be a lot more intuitive. 684 00:34:51,480 --> 00:34:53,600 Speaker 1: It's interesting when I have that argument, because I've had 685 00:34:53,680 --> 00:34:55,479 Speaker 1: arguments with it and I never win because it doesn't 686 00:34:55,480 --> 00:34:58,120 Speaker 1: really care that we're arguing. But I mean for it 687 00:34:58,200 --> 00:35:01,640 Speaker 1: to criticize its own doing its own work is actually 688 00:35:01,719 --> 00:35:03,840 Speaker 1: very interesting. And to go back and say, well, that 689 00:35:03,840 --> 00:35:06,560 Speaker 1: doesn't score as high as this one does. And I'll say, 690 00:35:06,560 --> 00:35:08,080 Speaker 1: why don't you give me that one? So I'll actually 691 00:35:08,120 --> 00:35:10,360 Speaker 1: go in with and I'll say, give me the best 692 00:35:10,440 --> 00:35:12,719 Speaker 1: that we don't have to change the prompt that I 693 00:35:12,719 --> 00:35:14,560 Speaker 1: need for the thumbnail, that we don't have to alter 694 00:35:14,680 --> 00:35:17,279 Speaker 1: at all whatsoever. And even that when you're getting that 695 00:35:17,320 --> 00:35:19,799 Speaker 1: specific doesn't really work because later it's going to go 696 00:35:19,840 --> 00:35:21,920 Speaker 1: back and recriticize what it is that you're giving it. 697 00:35:21,920 --> 00:35:23,680 Speaker 1: It's almost as if it has no idea that what 698 00:35:23,719 --> 00:35:25,640 Speaker 1: I'm giving it is what it gave me, and I'm 699 00:35:25,680 --> 00:35:26,520 Speaker 1: guessing it doesn't. 700 00:35:27,440 --> 00:35:31,320 Speaker 2: It doesn't. But there are tricks, ninja tricks, so to speak. 701 00:35:31,360 --> 00:35:35,040 Speaker 2: Oh really, okay, Yeah, there are a couple like, for example, 702 00:35:35,360 --> 00:35:38,040 Speaker 2: let's say you're talking about graphic design and you have 703 00:35:38,080 --> 00:35:41,080 Speaker 2: a very specific esthetic in mind that's of a really 704 00:35:41,200 --> 00:35:44,160 Speaker 2: high level sort of like a Madison Avenue ad agency, 705 00:35:44,480 --> 00:35:47,960 Speaker 2: and maybe you even do that, you actually say, imagine 706 00:35:48,000 --> 00:35:51,520 Speaker 2: yourself to be the creative director at insert whatever ad 707 00:35:51,560 --> 00:35:55,759 Speaker 2: agency you want, cbdo Ogilvie and Mather, you know, Leo 708 00:35:55,840 --> 00:36:01,000 Speaker 2: Burnett ad agency, and based upon its esthetic, and it's 709 00:36:01,480 --> 00:36:05,280 Speaker 2: a public facing work. Now judge it against that standard, 710 00:36:05,360 --> 00:36:08,040 Speaker 2: and you'll be surprised. Particularly if it's a paid model, 711 00:36:08,080 --> 00:36:11,520 Speaker 2: one of their higher level reasoning models, it actually will 712 00:36:11,520 --> 00:36:15,359 Speaker 2: get a lot closer. Not a perfect reality, but it can. 713 00:36:15,560 --> 00:36:19,560 Speaker 2: So prompting does matter, And if you really spend time 714 00:36:20,160 --> 00:36:22,799 Speaker 2: zooming in and lasering in on what it is you're 715 00:36:22,800 --> 00:36:25,720 Speaker 2: really trying to look for, that can that can actually 716 00:36:25,719 --> 00:36:29,440 Speaker 2: tighten up those responses. But yeah, no, what you're describing 717 00:36:29,520 --> 00:36:31,680 Speaker 2: is it's not your fault. That is a function of 718 00:36:31,760 --> 00:36:32,759 Speaker 2: yellow Well. 719 00:36:32,800 --> 00:36:34,080 Speaker 1: I know it's not my fault because it tells me 720 00:36:34,120 --> 00:36:36,560 Speaker 1: that I'm I'm smart to be angry at it. You're 721 00:36:36,600 --> 00:36:38,400 Speaker 1: right to be pissed, and I'll actually say that to me. 722 00:36:38,640 --> 00:36:40,880 Speaker 1: It is Winton Hall, author and political strategies to get 723 00:36:40,880 --> 00:36:43,160 Speaker 1: the book called Code Red, The Left, the Right, China 724 00:36:43,200 --> 00:36:44,840 Speaker 1: and the Race to Control AI. I feel like I 725 00:36:44,880 --> 00:36:46,400 Speaker 1: could talk to you for two hours because this is 726 00:36:46,400 --> 00:36:49,200 Speaker 1: such an important technology that we're all sort of inundated 727 00:36:49,239 --> 00:36:51,600 Speaker 1: with in the past few years. Let's do it again soon. 728 00:36:51,680 --> 00:36:54,680 Speaker 1: Go to the go to the website AI codreadbook dot com. Winton, 729 00:36:54,719 --> 00:36:56,640 Speaker 1: thanks for the time today, Oh. 730 00:36:56,560 --> 00:36:58,799 Speaker 2: Joe, great being with you. Thanks so much, appreciate you. 731 00:36:58,960 --> 00:37:00,400 Speaker 1: We'll talk to you again. That's going to do it 732 00:37:00,400 --> 00:37:02,719 Speaker 1: for Unshaken and Unafraid with Joe Paggs this time another 733 00:37:02,760 --> 00:37:04,560 Speaker 1: one to dropped very soon. Make sure you subscribe.