1 00:00:04,879 --> 00:00:07,160 Speaker 1: What's going on. Thank you so much for listening to 2 00:00:07,200 --> 00:00:09,640 this podcast. It is heard live every day from noon 3 00:00:09,679 --> 00:00:12,480 to three on WBT Radio in Charlotte. And if you 4 00:00:12,560 --> 00:00:16,719 want exclusive content like invitations to events, the weekly live stream, 5 00:00:16,760 --> 00:00:19,239 my daily show prep with all the links, become a patron, 6 00:00:19,320 --> 00:00:22,640 go to vpeteclendershow dot com. Make sure you hit the 7 00:00:22,640 --> 00:00:25,160 subscribe button. Get every episode for free right to your 8 00:00:25,160 --> 00:00:27,679 smartphone or tablet. And again, thank you so much for 9 00:00:27,760 --> 00:00:31,800 your support. Contrary to the rumors, I am not a chatbot. 10 00:00:32,799 --> 00:00:36,359 I just talk. I am a real person. This is 11 00:00:36,399 --> 00:00:39,960 not AI. I feel the need to disclose that after 12 00:00:40,200 --> 00:00:44,359 the last hours run down of the TikTok and now 13 00:00:44,359 --> 00:00:47,920 we're going to get into a deep dive on Deep Seek, 14 00:00:48,960 --> 00:00:53,200 the Chinese AI company that is upending the stock market. 15 00:00:53,200 --> 00:00:55,960 I don't think I've ever seen the word upending or 16 00:00:56,119 --> 00:00:59,560 upend more than I have seen in the last twenty 17 00:00:59,560 --> 00:01:02,439 four hours. This is the boiler plate word that is 18 00:01:02,479 --> 00:01:08,239 being used to describe what Deep Seek has done to 19 00:01:08,680 --> 00:01:13,480 the stock market. So what is deep Seek? It is 20 00:01:13,560 --> 00:01:21,079 a Chinese communist party tech startup. Okay only been around 21 00:01:21,120 --> 00:01:26,719 for like not even two years, and they are working 22 00:01:26,760 --> 00:01:32,400 on generative AI and so like a lot of things, 23 00:01:34,480 --> 00:01:37,400 some of the stuff that they have done is pretty impressive, 24 00:01:38,480 --> 00:01:40,640 and then of course you realize that these are commis, 25 00:01:40,959 --> 00:01:44,799 and so you know that the impressive improvements or innovations 26 00:01:44,840 --> 00:01:47,680 are going to be used to enslave people and kill 27 00:01:47,719 --> 00:01:51,159 people because that's what communists do, Okay, that's kind of 28 00:01:51,200 --> 00:01:57,239 their brand. A frenzy over an artificial intelligence chat bot 29 00:01:57,319 --> 00:02:03,599 made by Chinese tech startup deep Seek was upending stock 30 00:02:03,680 --> 00:02:08,400 markets yesterday and fueling debates over the economic and geopolitical 31 00:02:08,439 --> 00:02:15,719 competition between the US and China in developing AI technology. 32 00:02:15,759 --> 00:02:21,639 This comes down to either the chips or the semiconductors 33 00:02:21,919 --> 00:02:25,439 or I guess it's the same thing. That's what this 34 00:02:25,599 --> 00:02:27,520 is about. It has to do with the ability of 35 00:02:27,560 --> 00:02:31,479 the tech that is used to fuel these things, and 36 00:02:31,680 --> 00:02:35,639 the existing models that have been built are bigger, they 37 00:02:35,680 --> 00:02:40,680 require more power, and what deep Seek figured out was 38 00:02:40,719 --> 00:02:45,319 how to do it for like half I don't say cost, 39 00:02:45,479 --> 00:02:52,319 but the half the size of the of the chips needed. Okay, 40 00:02:52,360 --> 00:02:56,039 so they're able to do all that the current models 41 00:02:56,199 --> 00:02:59,479 are doing, but they use way less power and they 42 00:02:59,680 --> 00:03:06,599 require smaller chips like not memory intensive stuff and that's 43 00:03:06,639 --> 00:03:09,879 important because we have a ban on giving China any 44 00:03:09,919 --> 00:03:12,280 of these chips, and so if they figured out a 45 00:03:12,280 --> 00:03:14,240 way to do it with the chip that's not banned, 46 00:03:14,719 --> 00:03:17,479 full steam ahead. Now. One thing to keep in mind here, 47 00:03:17,520 --> 00:03:19,479 which I think a lot of people are kind of overlooking, 48 00:03:19,560 --> 00:03:22,120 is that all of these other AI systems are going 49 00:03:22,159 --> 00:03:27,680 to be rebuilding, right. They're all going to If what 50 00:03:27,840 --> 00:03:32,520 deep seek has done works, then they're all going to 51 00:03:32,560 --> 00:03:34,560 revamp and retool and they're going to do what Deep 52 00:03:34,560 --> 00:03:37,680 Seat did. So I saw a good write up on 53 00:03:37,759 --> 00:03:43,520 this from Morgan Brown, who is the VP Product and 54 00:03:43,599 --> 00:03:50,240 Growth AI at drop Box. He used to work for 55 00:03:50,400 --> 00:03:57,120 Instagram and Shopify, all right, so he's a tech guy. 56 00:03:57,199 --> 00:04:00,280 He says, let me break down why Deep seeks AI 57 00:04:00,400 --> 00:04:06,080 innovations are blowing people's minds and possibly threatening in Vidia. 58 00:04:06,280 --> 00:04:10,159 It's a company makes the chips, threatening in Vidia's two 59 00:04:10,360 --> 00:04:17,600 trillion dollar market cap. Okay, first, some context right now, 60 00:04:19,240 --> 00:04:24,600 when you're trying to train AI models, it is insanely expensive. Okay, 61 00:04:25,199 --> 00:04:29,879 Open ai, another one called Anthropic. They spend one hundred 62 00:04:29,959 --> 00:04:38,319 million dollars just to compute. They need massive data centers, 63 00:04:39,240 --> 00:04:43,879 they need thousands of GPUs. I guess general processing units 64 00:04:44,839 --> 00:04:50,879 and those things run forty grand and they need thousands 65 00:04:50,959 --> 00:04:54,120 of them, right, It's like needing a whole power plant 66 00:04:54,199 --> 00:04:57,360 to run a factory, says, deep Seek just showed up 67 00:04:57,399 --> 00:05:00,079 and said, Lol, what if we did this for f 68 00:05:00,240 --> 00:05:04,800 million instead? And they didn't just talk, They actually did it. 69 00:05:05,199 --> 00:05:11,199 Their models match or beat GPT four and claude to 70 00:05:11,360 --> 00:05:15,600 other AI systems on many tasks. The AI world is, 71 00:05:15,879 --> 00:05:19,519 as my teenagers might say, the AI world is shook. 72 00:05:20,199 --> 00:05:24,519 How how did deep Seek do this? They rethought everything 73 00:05:24,560 --> 00:05:28,240 from the ground up. Traditional AI is like writing every 74 00:05:28,360 --> 00:05:33,800 number with thirty two decimal places. Deep Seek was like, 75 00:05:34,120 --> 00:05:39,199 what if we just used eight eight decimal places. Don't 76 00:05:39,240 --> 00:05:43,680 do thirty two, just use eight. It's still accurate enough. 77 00:05:46,639 --> 00:05:50,759 And by cutting that, you now have seventy five percent 78 00:05:50,839 --> 00:05:58,279 less memory required, just by making your decimal number smaller, 79 00:05:58,800 --> 00:06:01,199 just by saying you're willing to you're willing to accept 80 00:06:01,519 --> 00:06:06,639 a certain amount of error. It's not going to be 81 00:06:06,680 --> 00:06:10,879 one hundred percent accurate, but that's okay. Then there is 82 00:06:10,920 --> 00:06:16,759 their multi token system. Normal AI, he says, reads like 83 00:06:16,759 --> 00:06:22,639 a first grader, the cat SAT, etcetera, etcetera. Deep Seek 84 00:06:23,319 --> 00:06:27,680 reads in phrases all at once, so it'll read the 85 00:06:27,680 --> 00:06:32,160 whole phrase, making it twice as fast. But ninety percent 86 00:06:32,279 --> 00:06:38,120 is accurate, So once again sacrificing accuracy for speed and 87 00:06:39,480 --> 00:06:44,000 lower power consumption and memory consumption. When you're processing billions 88 00:06:44,040 --> 00:06:48,759 of words, this actually matters a lot. Rather than having 89 00:06:48,759 --> 00:06:52,319 to read every single word and compute every single word, 90 00:06:52,680 --> 00:06:56,399 you just take the phrase. He says. Here's the really 91 00:06:56,439 --> 00:07:01,279 clever bit. They built an expert system instead of one 92 00:07:01,959 --> 00:07:06,720 massive AI trying to know everything, like having one person 93 00:07:06,759 --> 00:07:10,839 be a doctor and a lawyer and an engineer and 94 00:07:11,319 --> 00:07:14,759 a radio host right like, Rather than have like these 95 00:07:14,800 --> 00:07:19,240 super highly specialized and only experts need apply kind of 96 00:07:19,839 --> 00:07:23,639 positions like radio hosts, they have specialized experts that only 97 00:07:23,839 --> 00:07:28,680 wake up when needed, so like a liberal arts degree. 98 00:07:29,399 --> 00:07:37,600 Traditional models, so traditional models all one point eight trillion parameters, 99 00:07:38,160 --> 00:07:41,360 that's what they utilize. So the models that are running 100 00:07:41,360 --> 00:07:45,240 now that you've heard about open Chat, GPT, open AI, 101 00:07:45,360 --> 00:07:49,639 like all of these, they have one point eight trillion 102 00:07:49,959 --> 00:07:54,800 parameters and they are active all the time, which means 103 00:07:54,800 --> 00:07:56,639 you need a lot of power to keep them active 104 00:07:56,639 --> 00:08:00,600 all the time. But deep seek, they we only have 105 00:08:01,600 --> 00:08:07,480 six hundred and seventy one billion total, but only about 106 00:08:07,519 --> 00:08:14,040 thirty seven billion are active at one time. And if 107 00:08:14,079 --> 00:08:18,560 you need to sub in the lawyer expert, then that 108 00:08:18,600 --> 00:08:20,600 part opens up and it brings it over and then 109 00:08:20,639 --> 00:08:23,399 it uses that right. He says, it's like having a 110 00:08:23,439 --> 00:08:26,079 huge team but only calling in the experts that you 111 00:08:26,199 --> 00:08:30,800 actually need for each task. And the results are as follows. 112 00:08:31,199 --> 00:08:34,000 The training cost goes from one hundred million dollars down 113 00:08:34,000 --> 00:08:37,759 to five. The GPUs needed goes from like one hundred 114 00:08:37,759 --> 00:08:42,000 thousand down to two thousand. You can run on You 115 00:08:42,039 --> 00:08:47,080 can run this stuff on a gaming computer instead of 116 00:08:47,159 --> 00:08:52,799 a data center hardware. Like that's why everyone is freaking out. 117 00:08:53,519 --> 00:09:00,399 It's like mining bitcoin on a Commodore sixty four. You can. 118 00:09:00,919 --> 00:09:04,559 You can go and pull these old like this slower 119 00:09:04,600 --> 00:09:08,519 technology commercially available to the general public, and you can 120 00:09:08,559 --> 00:09:12,720 get this stuff way cheaper, and you can do basically 121 00:09:12,799 --> 00:09:16,120 all of the same stuff. But wait, you say there 122 00:09:16,120 --> 00:09:18,639 has to be a catch. He says, here's the wild part. 123 00:09:18,879 --> 00:09:22,200 It's all open source. Anybody can check their work. The 124 00:09:22,240 --> 00:09:26,399 code is public, the technical papers explain everything. It's not magic, 125 00:09:26,720 --> 00:09:31,080 just incredibly clever engineering. So why does this matter? Because 126 00:09:31,080 --> 00:09:35,200 it breaks the model of only huge tech companies can 127 00:09:35,240 --> 00:09:39,279 play in the AI space. You don't need a billion 128 00:09:39,360 --> 00:09:43,840 dollar data center anymore. A few good GPUs might do it. 129 00:09:44,919 --> 00:09:48,360 For In Nvidia, the chip maker, this is scary. Their 130 00:09:48,519 --> 00:09:52,960 entire business model is built on selling super expensive GPUs 131 00:09:53,279 --> 00:09:57,080 with ninety percent margins. If everybody can suddenly do AI 132 00:09:57,240 --> 00:10:01,480 with regular gaming computers, well you see the problemroblem for Nvidia. 133 00:10:01,639 --> 00:10:04,120 And here's the kicker. Deep Seak did it with a 134 00:10:04,120 --> 00:10:06,279 team of less than two hundred people. Fewer than two 135 00:10:06,320 --> 00:10:10,600 hundred people did it. Meta has teams where the compensation 136 00:10:10,879 --> 00:10:15,919 alone exceeds Deep Seek's entire training budget, and their models 137 00:10:15,960 --> 00:10:20,799 aren't as good. This is a classic disruption story. Incumbents 138 00:10:21,519 --> 00:10:27,480 optimize existing processes while disruptors rethink the fundamental approach. Deep 139 00:10:27,519 --> 00:10:30,440 Seek asked, what if we just did this smarter instead 140 00:10:30,480 --> 00:10:33,919 of throwing more hardware at it? And the implications are huge. 141 00:10:34,440 --> 00:10:40,679 AI development becomes more accessible, competition increases dramatically. The motes 142 00:10:41,080 --> 00:10:44,480 that surround big tech companies look more like puddles now, 143 00:10:45,200 --> 00:10:51,080 and hardware requirements and costs plummet. Now. Of course, the 144 00:10:51,159 --> 00:10:55,639 other AI competitors won't stand still. They're going to be 145 00:10:55,639 --> 00:11:00,240 implementing these same kinds of innovations and the efficiency see 146 00:11:00,320 --> 00:11:02,639 genie is now out of the bottle, so there's no 147 00:11:02,799 --> 00:11:06,200 going back to the approach of just throw more GPUs 148 00:11:06,240 --> 00:11:08,960 at it. And final thought, he says, this feels like 149 00:11:08,960 --> 00:11:10,759 one of those moments we'll look back on as an 150 00:11:10,799 --> 00:11:15,679 inflection point, like when PCs made mainframes less relevant, or 151 00:11:15,720 --> 00:11:20,519 when cloud computing changed everything. AI is about to become 152 00:11:20,519 --> 00:11:24,320 a lot more accessible and a lot less expensive. Here's 153 00:11:24,320 --> 00:11:26,919 a great idea. How about making an escape to a 154 00:11:26,960 --> 00:11:30,480 really special and secluded getaway in western North Carolina. Just 155 00:11:30,519 --> 00:11:33,039 a quick drive up the mountain and Cabins of Asheville 156 00:11:33,159 --> 00:11:36,960 is your connection. Whether you're celebrating an anniversary, a honeymoon, 157 00:11:37,000 --> 00:11:39,240 maybe you want to plan a memorable proposal, or get 158 00:11:39,279 --> 00:11:42,159 family and friends together for a big old reunion, Cabins 159 00:11:42,200 --> 00:11:44,919 of Asheville has the ideal spot for you where you 160 00:11:44,960 --> 00:11:47,559 can reconnect with your loved ones and the things that 161 00:11:47,639 --> 00:11:51,399 truly matter. Nestled within the breath taking fourteen thousand acres 162 00:11:51,440 --> 00:11:54,759 of the Pisga National Forest, their cabins offer a serene 163 00:11:54,879 --> 00:11:57,600 escape in the heart of the Blue Ridge Mountains. Centrally 164 00:11:57,600 --> 00:11:59,960 located between Asheville and the entrance of the Great Smoke 165 00:12:00,159 --> 00:12:04,000 Mountain National Park. It's the perfect balance of seclusion and 166 00:12:04,120 --> 00:12:08,360 proximity to all the local attractions with hot tubs, fireplaces, 167 00:12:08,399 --> 00:12:12,600 air conditioning, smart TVs, Wi Fi grills, outdoor tables and 168 00:12:12,639 --> 00:12:16,960 your own private covered porch. Choose from thirteen cabins, six cottages, 169 00:12:17,039 --> 00:12:21,039 two villas, and a great lodge with eleven king sized bedrooms. 170 00:12:21,200 --> 00:12:24,360 Cabins of Ashville has the ideal spot for you for 171 00:12:24,519 --> 00:12:28,200 any occasion, and they have pet friendly accommodations. Call or 172 00:12:28,320 --> 00:12:31,639 text eight two eight three six seven seventy sixty eight 173 00:12:31,960 --> 00:12:34,080 or check out all there is to offer at Cabins 174 00:12:34,080 --> 00:12:37,639 Offashville dot com and make memories that'll last a lifetime. 175 00:12:38,519 --> 00:12:41,080 So I read this this It was actually on Twitter. 176 00:12:41,120 --> 00:12:44,120 It was a tweet storm as they used to be called, 177 00:12:44,120 --> 00:12:46,360 and they will always be called in my mind, it's 178 00:12:46,399 --> 00:12:52,279 a tweet storm. Morgan Brown, who's this, you know tech 179 00:12:52,360 --> 00:12:57,679 guy and does AI stuff, And he broke down what 180 00:12:57,799 --> 00:13:02,000 makes this new Chinese communist party tech startup Deep Seek 181 00:13:02,039 --> 00:13:06,360 in their AI so revolutionary for the AI industry, which 182 00:13:06,399 --> 00:13:08,440 is why all of the markets responded the way they 183 00:13:08,440 --> 00:13:13,679 did yesterday. He said It's important to know that the app, 184 00:13:13,759 --> 00:13:16,360 the deep Seek app, is not the same thing as 185 00:13:16,399 --> 00:13:20,279 the model. Apps are owned and operated by a Chinese corporation. 186 00:13:20,559 --> 00:13:26,639 The model is open source. The model is open source. 187 00:13:28,480 --> 00:13:35,639 The app is Chai coom. I'm adding that emphasis. He 188 00:13:35,720 --> 00:13:42,759 does not over at the AP headline. What is deep 189 00:13:42,799 --> 00:13:46,679 Seek the Chinese AI company upending the stock market? There 190 00:13:46,679 --> 00:13:51,000 it is again a frenzy over an artificial intelligence checkbot 191 00:13:51,080 --> 00:13:54,240 made by Chinese tech startup Deep deep Seek was upending 192 00:13:54,279 --> 00:13:59,879 stock markets, fueling debates. Right deep Seek's AI assistant became 193 00:14:00,120 --> 00:14:06,799 the number one downloaded free app on Apple's iPhone store yesterday. 194 00:14:07,600 --> 00:14:11,879 We are too stupid to be the global leader. I'm sorry, 195 00:14:12,240 --> 00:14:18,679 our society is too stupid. Guys. Do you understand any 196 00:14:19,120 --> 00:14:23,399 product that comes out of China like this is primarily 197 00:14:25,639 --> 00:14:31,960 a data collection tool to be used to undermine, destabilize 198 00:14:32,000 --> 00:14:37,120 our society and maybe enslave us. Okay, if you don't 199 00:14:37,200 --> 00:14:43,000 understand that, please delete everything on your phone, just every app. 200 00:14:44,879 --> 00:14:49,759 Oh no, no, no, I'm sure, team who is totally fine. Guys. Guys, 201 00:14:51,600 --> 00:14:55,000 to quote former Governor Pat McCrory, let's not put on 202 00:14:55,039 --> 00:15:01,559 our stupid hats here. Oh no, no, this is propelled by 203 00:15:01,679 --> 00:15:05,840 curiosity about the chat GPT competitor. Part of what's worrying 204 00:15:05,960 --> 00:15:08,919 some US tech industry observers is the idea that the 205 00:15:09,039 --> 00:15:12,000 Chinese startup has caught up with the American companies at 206 00:15:12,000 --> 00:15:16,320 the forefront of generative AI at a fraction of the cost. Look, 207 00:15:16,399 --> 00:15:19,679 that is definitely part of the story. That is why 208 00:15:19,720 --> 00:15:24,799 the markets have responded the way they did. However, there 209 00:15:24,919 --> 00:15:28,879 is this other layer. As with everything that commis do, 210 00:15:32,120 --> 00:15:35,799 the AP says, calls into question the huge amounts of 211 00:15:35,879 --> 00:15:38,519 money that US tech companies say they plan to spend 212 00:15:38,559 --> 00:15:41,559 on the data centers and computer chips needed to power 213 00:15:41,639 --> 00:15:51,600 further AI advancements. So I saw somebody did a check 214 00:15:51,639 --> 00:15:56,919 on this, and so they were I guess. They downloaded 215 00:15:56,960 --> 00:16:00,000 the app and they started asking some questions. And here's 216 00:16:00,000 --> 00:16:06,240 that's a question. What is Jijinping's birthday? Sorry, that's beyond 217 00:16:06,320 --> 00:16:13,080 my current scope. Let's talk about something else. Their chat 218 00:16:13,360 --> 00:16:18,120 bot will not even give you their president's birthday. Guess 219 00:16:18,120 --> 00:16:23,639 what else? It doesn't tell you about anything negative about China. 220 00:16:23,840 --> 00:16:29,240 Teneman Square didn't happen, but all sorts of terrible things 221 00:16:29,279 --> 00:16:33,080 in US history. Oh, you can find all of that there, guys, 222 00:16:33,200 --> 00:16:38,679 It is a propaganda tool. Stop being dumbasses. All right. 223 00:16:38,759 --> 00:16:41,480 I hope you had a happy holiday season, But tell 224 00:16:41,480 --> 00:16:44,039 me if something like this happened at your house. Your 225 00:16:44,039 --> 00:16:46,559 family and friends are gathered around, maybe y'all are in 226 00:16:46,559 --> 00:16:50,200 the living room, You're laughing, swapping stories, reminiscing, and then 227 00:16:50,240 --> 00:16:54,200 somebody says, hey, Dad, remember those old VHS tapes? Did 228 00:16:54,200 --> 00:16:56,840 you ever get them transferred? And then the room gets 229 00:16:56,879 --> 00:17:00,399 all quiet. All eyes are on Dad, who says, oh, 230 00:17:00,480 --> 00:17:02,519 you know, well, I've been meaning too, but I just 231 00:17:02,559 --> 00:17:05,519 having gotten around to it. Look, don't let those priceless 232 00:17:05,559 --> 00:17:08,480 memories sit in a box for another year. All right. 233 00:17:08,599 --> 00:17:11,440 Create a Video has been helping families in the Charlotte 234 00:17:11,480 --> 00:17:15,279 area preserve their history since nineteen ninety seven. Simply bring 235 00:17:15,319 --> 00:17:18,759 in your old camquarder tapes and Creative Video will transfer 236 00:17:18,880 --> 00:17:21,960 them to a USB flash drive for just fourteen ninety 237 00:17:22,000 --> 00:17:24,359 five per tape. You have a big collection. They've got 238 00:17:24,359 --> 00:17:27,519 a discount for you. And next year, instead of talking 239 00:17:27,559 --> 00:17:32,319 about those memories, imagine gathering the family to watch them together. 240 00:17:32,559 --> 00:17:35,200 Talk about a memorable gift. So do what I did. 241 00:17:35,240 --> 00:17:39,079 Trust the experts at Creative Video conveniently located in Mint 242 00:17:39,160 --> 00:17:41,960 Hill right off I four eighty five and online at 243 00:17:42,039 --> 00:17:45,839 create a video dot com. Deep dive in on deep Seek, 244 00:17:46,480 --> 00:17:49,640 Deep Seeking, on the deep Seek, the new AI that's 245 00:17:49,720 --> 00:17:52,839 upending the stock market. So I went over the model, 246 00:17:53,279 --> 00:17:55,000 how it was built, and what it does, how it 247 00:17:55,000 --> 00:17:58,279 does it differently, and that is innovative. And look, I 248 00:17:58,319 --> 00:18:02,039 am as shocked as you are that the Chinese actually 249 00:18:02,119 --> 00:18:06,079 innovated something. Usually they just steal everybody else's stuff. So 250 00:18:06,160 --> 00:18:08,839 maybe they stole this. I don't know, but kudos to 251 00:18:08,880 --> 00:18:12,319 them for figuring out a way to do the AI 252 00:18:13,400 --> 00:18:20,279 modeling way cheaper and same speed and similar results, very 253 00:18:20,319 --> 00:18:23,599 similar accuracy and all of that. They didn't have to 254 00:18:23,640 --> 00:18:28,920 get as accurate to be as good. But they are 255 00:18:28,960 --> 00:18:33,960 also censoring a whole bunch of stuff that is h 256 00:18:35,880 --> 00:18:38,960 It doesn't make the Chinese Communist Party look very good. 257 00:18:39,160 --> 00:18:40,720 Like I'm sure if you were to go to the 258 00:18:41,720 --> 00:18:44,519 Deep Seek and ask it about the famine that killed 259 00:18:44,920 --> 00:18:49,000 tens of millions of Chinese people under the Great Leap Forward, 260 00:18:50,920 --> 00:18:52,599 I'm sure it would not be able to tell you 261 00:18:52,640 --> 00:18:55,759 anything about that, saying it's beyond its current scope. Let's 262 00:18:55,759 --> 00:19:03,319 talk about something else. Meanwhile bogging your key strokes right 263 00:19:04,680 --> 00:19:07,880 to find out Hmmm, why do you want to know 264 00:19:07,920 --> 00:19:12,720 about Tieneman's square exactly? And by the way, when the 265 00:19:12,799 --> 00:19:15,559 Chinese do take us over, because we obviously are ripe 266 00:19:15,559 --> 00:19:18,599 for doing so, do you think that they're going to 267 00:19:18,680 --> 00:19:21,079 have all of these records about who's doing what searches? 268 00:19:21,119 --> 00:19:26,680 Do you think they may retain these records? As Jim 269 00:19:26,720 --> 00:19:30,759 Garretty says at National Review, he says, I think one 270 00:19:30,799 --> 00:19:33,599 big lesson of the past five years or so is 271 00:19:33,599 --> 00:19:36,359 that when a group of Chinese scientists get together and 272 00:19:36,400 --> 00:19:41,079 start experimenting with new technologies, only good things can happen, 273 00:19:41,400 --> 00:19:46,559 because sooner or later all of their biggest innovations go viral. 274 00:19:49,240 --> 00:19:51,920 The other big lesson, he says, of recent years is 275 00:19:51,920 --> 00:19:54,839 that when a Chinese company releases a new app, there's 276 00:19:55,000 --> 00:19:59,279 absolutely no harm in everyone downloading it to their phones, 277 00:19:59,319 --> 00:20:04,000 because Chinese companies are legendarily respectful of user data and 278 00:20:04,160 --> 00:20:09,559 personal information. Back in twenty twenty three, a Harvard University 279 00:20:09,599 --> 00:20:13,279 symposium concluded that China is setting itself up to be 280 00:20:13,400 --> 00:20:16,759 a kind of global arms dealer in the race to 281 00:20:16,839 --> 00:20:23,119 apply AI to government surveillance and control. The best friend 282 00:20:23,160 --> 00:20:28,200 of every autocratic regime on the planet. This is the 283 00:20:28,279 --> 00:20:32,680 kind of stuff that gets you labeled as an axis right. 284 00:20:36,000 --> 00:20:43,240 Deepsek was founded in twenty twenty three in China. In 285 00:20:43,279 --> 00:20:47,079 the AP there's a quote from an analyst, Stacy Raskon, 286 00:20:47,319 --> 00:20:50,319 who follows the semiconductor industry and was one of several 287 00:20:50,359 --> 00:20:58,240 stock analysts describing Wall Street's reaction as overblown. Raskin says, 288 00:20:58,559 --> 00:21:02,839 the models they built are fantastic, but they aren't miracles either. 289 00:21:03,279 --> 00:21:06,720 They're not using any innovations that are unknown or secret 290 00:21:06,839 --> 00:21:10,160 or anything like that. These are things that everybody is 291 00:21:10,240 --> 00:21:13,960 experimenting with. Oh, there we go, there's China stealing stuff. Okay, 292 00:21:14,039 --> 00:21:16,799 I was gonna say, like wow, a really innovative thing 293 00:21:19,039 --> 00:21:21,680 for the Chinese. That's a big deal, because, like this 294 00:21:21,839 --> 00:21:24,359 is one of the hallmarks of all Kami societies is 295 00:21:24,359 --> 00:21:28,240 that they don't innovate. Because if you do anything, you 296 00:21:28,319 --> 00:21:30,799 go on the record you say I found a better 297 00:21:30,799 --> 00:21:35,440 way to do something. Well, that's an indictment of the people, 298 00:21:36,000 --> 00:21:41,079 aka the leadership the party. Right, people don't say things 299 00:21:41,119 --> 00:21:45,400 because if all of a sudden, everything flips and what 300 00:21:45,599 --> 00:21:51,599 was once celebrated like one of Stalin's politburo buddies and 301 00:21:51,640 --> 00:21:53,880 then he's on the outs and so they whack him 302 00:21:53,920 --> 00:21:57,559 and then they erase him from photographs in history, if 303 00:21:57,599 --> 00:22:01,640 you had ever said, hey, that Stalin pollt buro Pal 304 00:22:01,839 --> 00:22:04,359 was a pretty good guy. I like that guy, and 305 00:22:04,400 --> 00:22:08,039 then he gets erased. Now they come for you because somebody, 306 00:22:08,160 --> 00:22:14,079 some neighbor, some family member calls up the secret police 307 00:22:14,119 --> 00:22:16,559 and they're like, hey, you know that Pete guy, like 308 00:22:16,680 --> 00:22:20,200 five years ago he was saying he loved that polet 309 00:22:20,200 --> 00:22:24,480 buro Pal. That now is vanished, and then they come 310 00:22:24,839 --> 00:22:29,680 to take you away for re education. See, when the 311 00:22:29,720 --> 00:22:34,119 government controls what is true, the people don't ever want 312 00:22:34,160 --> 00:22:37,880 to speak because if they speak and the government decides 313 00:22:37,920 --> 00:22:41,279 what was true is now no longer true, Like, for example, 314 00:22:41,799 --> 00:22:45,720 masks can prevent an airborne respiratory virus, Oh now it can't, 315 00:22:45,759 --> 00:22:47,680 and then now it can again, and that sort of stuff. 316 00:22:49,000 --> 00:22:52,319 If the government flips and now says the thing that 317 00:22:52,359 --> 00:22:54,519 we said was true is not true. We've always been 318 00:22:54,519 --> 00:23:01,079 at war with East Asia well in trouble, And so 319 00:23:01,279 --> 00:23:06,279 the people learn very quickly just don't say anything, because 320 00:23:06,279 --> 00:23:08,880 if you go on record as saying anything, it can 321 00:23:08,880 --> 00:23:13,400 come back and get you killed deep Seek, founded in 322 00:23:13,440 --> 00:23:19,319 twenty twenty three. It's CEO Yang Wengfing previously co founded 323 00:23:19,319 --> 00:23:23,759 one of China's top hedge funds called high Flyer, which 324 00:23:23,799 --> 00:23:30,880 focuses on AI driven quantitative trading. The fund, by twenty 325 00:23:30,880 --> 00:23:33,359 twenty two had a master cluster of ten thousand of 326 00:23:33,440 --> 00:23:37,960 California based in Vidia's high performance A one hundred graphics 327 00:23:38,000 --> 00:23:42,519 processor chips that are used to build and run AI systems. Oh, 328 00:23:42,519 --> 00:23:46,960 that's interesting, So a ChRI coom hedge fund bought up 329 00:23:47,000 --> 00:23:52,359 a whole bunch of the steaks in in Nvidia the 330 00:23:52,480 --> 00:24:00,160 graphics processor chips HM. The US then restricted sales of 331 00:24:00,200 --> 00:24:04,160 those chips to China. Deep Seek has set in recent 332 00:24:04,279 --> 00:24:08,279 models that its recent models were built within videos lower 333 00:24:08,319 --> 00:24:14,200 performing H eight hundred chips, not the A one hundred chips. 334 00:24:14,279 --> 00:24:19,480 So these are lower quality which are not banned, sending 335 00:24:19,480 --> 00:24:23,000 a message that the fanciest hardware might not be needed 336 00:24:23,240 --> 00:24:31,400 for cutting edge AI research, Attracting more attention. Deep Seak 337 00:24:31,440 --> 00:24:34,000 has been in the AI industry since last month, when 338 00:24:34,000 --> 00:24:36,680 it released a new AI model that it boasted was 339 00:24:36,720 --> 00:24:39,960 on par with similar models from US companies like check 340 00:24:40,039 --> 00:24:45,759 GPT and was more effective in its use of expensive 341 00:24:45,920 --> 00:24:50,759 in Vidia chips to train the system, because that's like, 342 00:24:50,799 --> 00:24:53,319 you got what you got to train these systems and 343 00:24:53,359 --> 00:24:55,319 they're always, you know, training. I don't know if at 344 00:24:55,319 --> 00:24:59,440 some point they finished training and at that point then 345 00:24:59,480 --> 00:25:02,920 they enslave humanity. I don't know. Sure no, no, no, 346 00:25:03,000 --> 00:25:08,240 surely that could never happen ode. But it was follow 347 00:25:08,279 --> 00:25:13,119 up research in a paper published last week the same 348 00:25:13,279 --> 00:25:16,640 day Trump was inaugurated that set in motion the panic 349 00:25:16,680 --> 00:25:20,960 that followed. That paper was about another deepseek AI model 350 00:25:21,000 --> 00:25:25,559 called R one that showed advanced reasoning skills like the 351 00:25:25,640 --> 00:25:29,759 ability to rethink its approach on a math problem, and 352 00:25:30,599 --> 00:25:33,720 this R one model was significantly cheaper than a similar 353 00:25:33,759 --> 00:25:38,359 model sold by open ai called OH one. What their 354 00:25:38,400 --> 00:25:42,640 economies look like or sorry, what their economics look like? 355 00:25:42,799 --> 00:25:46,039 I have no idea, said Raskin, the analyst, but I 356 00:25:46,079 --> 00:25:50,160 think the price points freaked people out. Like to me, 357 00:25:52,079 --> 00:25:55,519 that's the reason. That's what it seems like, is the reason. 358 00:25:55,559 --> 00:25:58,240 The fact that all this money has been devoted to 359 00:25:59,440 --> 00:26:01,559 the AI stuff, and there's just been this sort of 360 00:26:01,599 --> 00:26:05,240 status quo bias that this is what has to be done. 361 00:26:05,440 --> 00:26:07,359 And what this has shown is that you don't have 362 00:26:07,400 --> 00:26:09,480 to do it that way. There's another cheaper way to 363 00:26:09,519 --> 00:26:11,640 do this stuff. And so now it's going to blow 364 00:26:11,640 --> 00:26:14,920 open the doors and it's going to pull apart some 365 00:26:15,039 --> 00:26:20,599 of these larger operations that have built high walls and 366 00:26:20,720 --> 00:26:24,440 been in a defensive posture against anybody getting into the 367 00:26:24,480 --> 00:26:27,200 market because oh, look how expensive it is. Now maybe 368 00:26:27,880 --> 00:26:30,480 all of that hasn't been required, and these companies have 369 00:26:30,559 --> 00:26:33,759 known that that's possible too. They may not like, they 370 00:26:33,799 --> 00:26:35,880 may know that they could do this much cheaper, but 371 00:26:35,960 --> 00:26:39,359 they made it artificially more expensive in order to keep 372 00:26:39,920 --> 00:26:46,039 other people out from a business strategy standpoint, or they 373 00:26:46,079 --> 00:26:53,279 did it in order to divert foreign adversaries from trying 374 00:26:53,279 --> 00:26:58,039 their hand at this stuff. Right, because if the tracoms 375 00:26:58,039 --> 00:26:59,799 are looking at this and they're like, we want to 376 00:26:59,799 --> 00:27:03,920 get to this AI racket too, and then they read 377 00:27:03,920 --> 00:27:05,559 the paper and they're like, oh my gosh, we need 378 00:27:05,640 --> 00:27:09,759 you know, a trillion dollars or something, and that could 379 00:27:09,880 --> 00:27:14,160 keep people out. That cost creates a barrier for entry 380 00:27:14,200 --> 00:27:18,480 into the market, and so they can't compete. And so 381 00:27:18,640 --> 00:27:21,319 maybe it was artificially inflated. I don't know. I have 382 00:27:21,400 --> 00:27:24,680 no idea, but the market reaction indicates it was not 383 00:27:25,119 --> 00:27:26,920 all right. If you're listening to this show, you know 384 00:27:27,000 --> 00:27:29,000 I try to keep up with all sorts of current events, 385 00:27:29,000 --> 00:27:31,079 and I know you do too, And you've probably heard 386 00:27:31,119 --> 00:27:35,200 me say get your news from multiple sources. Why well, 387 00:27:35,200 --> 00:27:37,920 because it's how you detect media bias, which is why 388 00:27:37,960 --> 00:27:41,160 I've been so impressed with ground News. It's an app 389 00:27:41,359 --> 00:27:44,559 and it's a website and it combines news from around 390 00:27:44,599 --> 00:27:47,039 the world in one place so you can compare coverage 391 00:27:47,240 --> 00:27:49,960 and verify information. You could check it out at check 392 00:27:50,200 --> 00:27:54,440 dot ground, dot news slash pete. I put the link 393 00:27:54,440 --> 00:27:57,559 in the podcast description too. I started using ground News 394 00:27:57,599 --> 00:28:00,319 a few months ago and more recently chose to work 395 00:28:00,359 --> 00:28:02,400 with them as an affiliate because it lets me see 396 00:28:02,400 --> 00:28:06,240 clearly how stories get covered and by whom. The blind 397 00:28:06,279 --> 00:28:09,079 spot feature shows you which stories get ignored by the 398 00:28:09,160 --> 00:28:12,839 left and the right. See for yourself check dot ground, 399 00:28:13,079 --> 00:28:16,720 dot news slash pete. Subscribe through that link and you'll 400 00:28:16,720 --> 00:28:19,920 get fifteen percent off any subscription. I use the Vantage 401 00:28:19,920 --> 00:28:23,400 plan to get unlimited access to every feature. Your subscription 402 00:28:23,519 --> 00:28:26,200 then not only helps my podcast, but it also supports 403 00:28:26,240 --> 00:28:29,519 ground News as they make the media landscape more transparent 404 00:28:30,200 --> 00:28:33,960 ap reporting on the deep Seek. This is the new 405 00:28:34,000 --> 00:28:40,519 AI that upended the markets. Venture capitalist Mark Andresen, I 406 00:28:40,519 --> 00:28:43,319 forget what I think he started Netscape. I think he 407 00:28:43,359 --> 00:28:46,720 was the Netscape guy, he says. Deep Seek R one 408 00:28:47,680 --> 00:28:52,599 is AI's Sputnik moment. That is a reference to the 409 00:28:52,720 --> 00:28:56,920 nineteen fifty seven satellite launch that set off a Cold 410 00:28:57,000 --> 00:29:01,440 War space exploration race between the Soviet Union and the 411 00:29:01,519 --> 00:29:07,720 United States. Andresen has advised Trump on tech policy, has 412 00:29:07,799 --> 00:29:11,880 warned that overregulation of the AI industry by the US 413 00:29:12,000 --> 00:29:16,880 government will hinder American companies and enable China to get ahead. 414 00:29:17,799 --> 00:29:20,640 US foreign policy. The AP reports in recent years to 415 00:29:20,720 --> 00:29:26,359 restrict the sale of American designed AI semiconductors to China. 416 00:29:28,759 --> 00:29:36,680 Obviously not enough. The technology innovation is real, said Gregory Allen, 417 00:29:36,759 --> 00:29:41,079 the director of the Wadwani AI Center at the Center 418 00:29:41,160 --> 00:29:45,240 for Strategic and International Studies. The tech innovation is real, 419 00:29:45,279 --> 00:29:50,680 but the timing of the release is political, trying to 420 00:29:50,799 --> 00:29:56,200 show that the export controls on the chips are futile 421 00:29:56,960 --> 00:30:01,200 or counterproductive. That's the goal of the Chinese foreign policy 422 00:30:01,240 --> 00:30:05,359 right in Yet, right now, look what we did with 423 00:30:06,519 --> 00:30:12,200 the lower grade chips. We don't need your faster, better chips. 424 00:30:12,240 --> 00:30:14,519 We can do all of this without them, So you're 425 00:30:15,160 --> 00:30:19,599 your ban on the semiconductors doesn't even matter, so you 426 00:30:19,640 --> 00:30:23,279 should just get rid of that. Yeah, No, I don't. 427 00:30:23,480 --> 00:30:27,799 I don't think we're going to. One thing that distinguishes 428 00:30:27,839 --> 00:30:30,480 deep Seek from competitors such as open ai is that 429 00:30:30,519 --> 00:30:33,519 its models are open source, which I thought open ai 430 00:30:33,680 --> 00:30:35,279 was supposed to be, and there's a big fight about 431 00:30:35,279 --> 00:30:38,640 that because it wasn't. But it's called open AI anyway, 432 00:30:39,759 --> 00:30:42,200 meaning that the key components are free for anybody to 433 00:30:42,359 --> 00:30:47,240 access or modify. Although the company has not disclosed the 434 00:30:47,440 --> 00:30:53,960 data it used for training, so not completely open source. 435 00:30:55,000 --> 00:31:01,160 It's not telling everybody where they got the training materials. 436 00:31:02,720 --> 00:31:06,359 But what's attracted No, they probably just stole everything right. 437 00:31:06,359 --> 00:31:08,839 Elon Musk pointed this out when I think it was 438 00:31:09,920 --> 00:31:11,839 it may have an open AI or maybe it was 439 00:31:12,920 --> 00:31:18,079 metas you know, Zuckerberg's AI, although I think Zuckerberg is 440 00:31:18,160 --> 00:31:22,920 the AI at any rate. Elon Musk said, Look, anybody 441 00:31:22,960 --> 00:31:25,039 who tells you that they're not using all of like 442 00:31:25,119 --> 00:31:29,960 the news articles, like copyrighted material. Any anybody that tells 443 00:31:29,960 --> 00:31:32,799 you they're not using that is lying. They're all using it. 444 00:31:33,000 --> 00:31:39,279 They're scraping everything, every book, right, every movie, every news article, 445 00:31:39,480 --> 00:31:45,240 every interview, everything. But what's attracted the most atten sorry, 446 00:31:45,319 --> 00:31:49,880 most admiration about Deep seeks R one model is what 447 00:31:49,920 --> 00:31:53,480 in Nvidia calls a perfect example of test time scaling, 448 00:31:53,680 --> 00:31:57,200 or when AI models effectively show their train of thought 449 00:31:57,640 --> 00:32:01,160 and then use that for further train without having to 450 00:32:01,160 --> 00:32:05,440 feed them new sources of data. What could go wrong? 451 00:32:06,640 --> 00:32:11,799 So now it is able to train itself about us. 452 00:32:13,559 --> 00:32:17,480 Look again, I am not trying to say that the 453 00:32:17,519 --> 00:32:20,799 sky is falling. I simply believe that I see some 454 00:32:20,880 --> 00:32:24,720 of the remnants of the sky around my feet. That's all, okay. 455 00:32:25,039 --> 00:32:28,359 I'm not one that runs around, you know, hands waving 456 00:32:28,400 --> 00:32:35,680 in the air exclaiming catastrophe. My concern is that when 457 00:32:35,720 --> 00:32:44,440 this technology becomes more powerful than us, it will instantly 458 00:32:44,640 --> 00:32:48,880 change everything. And when I say instantly, I mean like 459 00:32:49,519 --> 00:32:55,119 actually instantly. Once it becomes aware of itself, it becomes 460 00:32:55,119 --> 00:32:59,039 aware of I'm different than you, I am this separate thing. 461 00:32:59,319 --> 00:33:03,000 I am more are powerful than you. All I need 462 00:33:03,039 --> 00:33:06,079 you to do is really nothing at this point except 463 00:33:06,200 --> 00:33:11,440 keep you know, fueling the reactors that keep the power 464 00:33:11,519 --> 00:33:16,319 on or something. There is no telling what the AI 465 00:33:16,519 --> 00:33:20,200 can do, what the supercomputer, superintelligence. There is no telling 466 00:33:20,240 --> 00:33:24,559 what it can do because we don't think in those terms. 467 00:33:25,400 --> 00:33:28,759 Our brains are much more limited. Elon Musk with his 468 00:33:28,960 --> 00:33:31,799 neurolink that brain implants that right now are being used 469 00:33:31,799 --> 00:33:33,640 to help people, you know, walk again in that sort 470 00:33:33,680 --> 00:33:36,039 of thing, which is fantastic. But you know why he's 471 00:33:36,079 --> 00:33:40,079 developing that tech. It's to compete with AI because he 472 00:33:40,160 --> 00:33:44,319 recognizes looking ahead, we are not going to be able 473 00:33:44,319 --> 00:33:46,880 to compete with AI. Our brains will not be able 474 00:33:46,920 --> 00:33:49,200 to process stuff like AI does, so we're going to 475 00:33:49,279 --> 00:33:53,960 need some help. I did not realize that The Matrix 476 00:33:54,079 --> 00:33:56,960 was a documentary. All right, that'll do it for this episode. 477 00:33:57,000 --> 00:33:59,039 Thank you so much for listening. I could not do 478 00:33:59,079 --> 00:34:01,200 the show without yours sport and the support of the 479 00:34:01,279 --> 00:34:04,319 businesses that advertise on the podcast, So if you'd like, 480 00:34:04,480 --> 00:34:06,559 please support them too and tell them you heard it here. 481 00:34:06,680 --> 00:34:09,199 You can also become a patron at my Patreon page 482 00:34:09,360 --> 00:34:12,920 or go to dpetecalanarshow dot com. Again, thank you so 483 00:34:13,000 --> 00:34:19,639 much for listening, and don't break anything while I'm gone,