1 00:00:02,800 --> 00:00:08,319 Speaker 1: Zone Media. Hello, and welcome to Better Offline. I'm your 2 00:00:08,320 --> 00:00:22,759 Speaker 1: host ed Zitron. What In the last episode, I walked 3 00:00:22,760 --> 00:00:25,040 Speaker 1: you through how GENERALIFAI has hit its ceiling and how 4 00:00:25,040 --> 00:00:27,800 Speaker 1: big tech has burned billions of dollars chasing a technology 5 00:00:28,000 --> 00:00:30,000 Speaker 1: that really isn't going to give them the returns they 6 00:00:30,040 --> 00:00:33,560 Speaker 1: got from the smartphone and cloud computing revolutions, not that 7 00:00:33,560 --> 00:00:34,080 Speaker 1: that's going. 8 00:00:34,040 --> 00:00:35,440 Speaker 2: To stop them. 9 00:00:35,680 --> 00:00:37,800 Speaker 1: In this episode, I'm going to walk you through the 10 00:00:37,840 --> 00:00:41,159 Speaker 1: potentially massive problems they have ahead, the ones they've created 11 00:00:41,280 --> 00:00:44,320 Speaker 1: by building these massive supercomputers, the ones necessary to capture 12 00:00:44,320 --> 00:00:47,440 Speaker 1: this imaginary demand of this kind of bullshit AI revolution, 13 00:00:47,920 --> 00:00:50,080 Speaker 1: and I'm going to try and guess what might happen next. 14 00:00:50,720 --> 00:00:53,640 Speaker 1: All right, So, the AI boom helped the S and 15 00:00:53,680 --> 00:00:56,360 Speaker 1: P five hundred hit record levels in twenty twenty four, 16 00:00:56,600 --> 00:00:59,240 Speaker 1: largely thanks to chip giant Nvidia, a company that makes 17 00:00:59,280 --> 00:01:01,800 Speaker 1: both the GPUs necessary to train and run generative AI 18 00:01:02,080 --> 00:01:04,520 Speaker 1: and the software architecture behind them, as well as some 19 00:01:04,600 --> 00:01:08,440 Speaker 1: other server parts too. Part of Nvidia's remarkable growth has 20 00:01:08,440 --> 00:01:11,640 Speaker 1: been its ability to capitalize on Kuda architecture, the software 21 00:01:11,680 --> 00:01:15,160 Speaker 1: layer that lets you do complex computing on GPUs rather 22 00:01:15,200 --> 00:01:17,679 Speaker 1: than simply using them to render video games or complex 23 00:01:17,720 --> 00:01:20,080 Speaker 1: three D images. Someone's going to email and say that 24 00:01:20,080 --> 00:01:20,880 Speaker 1: that's partially wrong. 25 00:01:21,000 --> 00:01:22,120 Speaker 2: Just bear with me. 26 00:01:23,360 --> 00:01:25,840 Speaker 1: And of course, one of Nvidia's other things they do 27 00:01:25,959 --> 00:01:28,520 Speaker 1: is continually create new GPUs to sell for tens of 28 00:01:28,560 --> 00:01:30,679 Speaker 1: thousands of dollars to tech companies that want to burn 29 00:01:30,760 --> 00:01:34,080 Speaker 1: billions of dollars on GENERATIVII, which has led Nvidia's stock 30 00:01:34,120 --> 00:01:36,000 Speaker 1: to pop more than one hundred and seventy nine percent 31 00:01:36,040 --> 00:01:39,080 Speaker 1: over the last year. Back in May, in Vidia's CEO 32 00:01:39,120 --> 00:01:42,120 Speaker 1: and professional Carnival Barker Jensen Wang said that the company 33 00:01:42,160 --> 00:01:44,120 Speaker 1: was now and I quote on a one year rhythm 34 00:01:44,120 --> 00:01:48,120 Speaker 1: in AIGPU production with its latest Blackwell GPUs, specifically the 35 00:01:48,120 --> 00:01:50,760 Speaker 1: B one hundred, B two hundred and GBT two hundred 36 00:01:50,960 --> 00:01:53,360 Speaker 1: used of Regenerative AI, supposedly due at the end of 37 00:01:53,400 --> 00:01:56,360 Speaker 1: twenty twenty four, except they're now delayed until at least 38 00:01:56,400 --> 00:02:00,000 Speaker 1: March twenty twenty five. Now, before we go any firm, 39 00:02:00,200 --> 00:02:02,400 Speaker 1: it's worth noting that when I say GPU, I don't 40 00:02:02,440 --> 00:02:04,360 Speaker 1: mean the one you'd find in the gaming PC, but 41 00:02:04,400 --> 00:02:06,760 Speaker 1: a much much larger chip put on a specialized server 42 00:02:07,000 --> 00:02:10,239 Speaker 1: and actually a specialized board as well. With multiple other GPUs, 43 00:02:10,320 --> 00:02:13,560 Speaker 1: all integrated with special case in cooling and networking infrastructure, 44 00:02:13,880 --> 00:02:15,760 Speaker 1: all of which is sold by Nvidia. By the way, 45 00:02:15,840 --> 00:02:18,280 Speaker 1: you can use other things, but as well buy it 46 00:02:18,280 --> 00:02:21,840 Speaker 1: in one place, right, That's their product strategy. In simple terms, 47 00:02:21,880 --> 00:02:23,640 Speaker 1: these are the things necessary to make sure all these 48 00:02:23,720 --> 00:02:26,520 Speaker 1: chips work together efficiently and also stop them from overheating 49 00:02:26,800 --> 00:02:29,880 Speaker 1: because they get extremely hot, and generative AI makes them 50 00:02:29,960 --> 00:02:31,680 Speaker 1: run at full speed all the time. 51 00:02:32,560 --> 00:02:33,519 Speaker 2: Now, the initial. 52 00:02:33,240 --> 00:02:35,200 Speaker 1: Delay of the new Blackwell chips is caused by a 53 00:02:35,240 --> 00:02:38,720 Speaker 1: now fixed design, floor and production. But as I've previously suggested, 54 00:02:38,720 --> 00:02:41,760 Speaker 1: the problem isn't just creating the chips, it's making sure 55 00:02:41,800 --> 00:02:44,560 Speaker 1: they actually work at scale for the jobs they've. 56 00:02:44,400 --> 00:02:45,000 Speaker 2: Been bought for. 57 00:02:46,840 --> 00:02:52,560 Speaker 1: Now what if that wasn't possible a few weeks ago? 58 00:02:52,639 --> 00:02:55,200 Speaker 1: The Information Report and Video is currently grappling with the 59 00:02:55,200 --> 00:02:58,560 Speaker 1: oldest problem in computing, how to cool the fucking things. 60 00:02:59,080 --> 00:03:01,640 Speaker 1: According to their report, and Video has been asking supplies 61 00:03:01,680 --> 00:03:04,800 Speaker 1: to change the design of its three thousand pound seventy 62 00:03:04,840 --> 00:03:08,440 Speaker 1: five GPU server racks several times to overcome these heating problems, 63 00:03:08,600 --> 00:03:11,880 Speaker 1: which the Information calls quote the most complicated design that 64 00:03:11,960 --> 00:03:14,639 Speaker 1: Nvideo had ever come up with. According to the report. 65 00:03:14,680 --> 00:03:17,200 Speaker 1: A few months after revealing the racks, engineers found that 66 00:03:17,240 --> 00:03:21,120 Speaker 1: they didn't work properly even when they use the smaller 67 00:03:21,200 --> 00:03:24,079 Speaker 1: thirty six chip racks, and they've been kind of scrambling 68 00:03:24,080 --> 00:03:25,000 Speaker 1: to fix it ever since. 69 00:03:25,760 --> 00:03:27,560 Speaker 2: While one can dazzle investors with. 70 00:03:27,480 --> 00:03:31,640 Speaker 1: Buzzwords, charts and the term superintelligence, the laws of physics 71 00:03:31,680 --> 00:03:34,240 Speaker 1: are a little bit more of a harsh mistress. And 72 00:03:34,280 --> 00:03:36,560 Speaker 1: if in Video is struggling mere months before the first 73 00:03:36,560 --> 00:03:39,200 Speaker 1: installations to begin, though I've heard some are going out now, 74 00:03:39,680 --> 00:03:43,000 Speaker 1: it's kind of unclear how they practically plan to launch 75 00:03:43,040 --> 00:03:46,320 Speaker 1: this next generation of chips, let alone doing another one 76 00:03:46,320 --> 00:03:49,520 Speaker 1: in a year. The Information reports that these changes have 77 00:03:49,560 --> 00:03:52,520 Speaker 1: been made late in production, which is scaring customers that 78 00:03:52,560 --> 00:03:55,200 Speaker 1: desperately need them now so that their models can continue 79 00:03:55,240 --> 00:03:57,280 Speaker 1: to do something that they will work out later. 80 00:03:57,800 --> 00:03:59,320 Speaker 2: To quote the Information again. 81 00:04:00,000 --> 00:04:02,440 Speaker 1: Secutives at large cloud providers that have ordered the new 82 00:04:02,520 --> 00:04:05,760 Speaker 1: chips said they're concerned that such last minute difficulties might 83 00:04:05,760 --> 00:04:07,480 Speaker 1: push back the timeline for when they can get their 84 00:04:07,480 --> 00:04:11,000 Speaker 1: GPU clusters up and running next year. The fact that 85 00:04:11,000 --> 00:04:14,520 Speaker 1: in video is having such significant difficulties with thermal performance. 86 00:04:14,160 --> 00:04:15,320 Speaker 2: Is very, very bad. 87 00:04:15,760 --> 00:04:18,800 Speaker 1: These chips are very expensive thirty to seventy grand I 88 00:04:18,880 --> 00:04:21,400 Speaker 1: hear and will be running, as I've mentioned, at full speed, 89 00:04:21,600 --> 00:04:24,360 Speaker 1: generating an incredible amount of heat that must be dissipated 90 00:04:24,640 --> 00:04:27,200 Speaker 1: while sat next to anywhere from thirty five to seventy 91 00:04:27,240 --> 00:04:30,040 Speaker 1: one other chips, which will in turn be densely packed 92 00:04:30,120 --> 00:04:32,440 Speaker 1: so that you can cram more servers into a data center. 93 00:04:33,000 --> 00:04:36,280 Speaker 1: New more powerful chips require entirely new methods to rack, 94 00:04:36,320 --> 00:04:38,839 Speaker 1: mount them, operate them, and cool them, and all of 95 00:04:38,880 --> 00:04:44,520 Speaker 1: these parts must operate and sync as overheating GPUs will die. 96 00:04:44,600 --> 00:04:46,599 Speaker 1: And while these units are big, some of their internal 97 00:04:46,600 --> 00:04:49,600 Speaker 1: components and nanometers in size, and unless properly cool their 98 00:04:49,600 --> 00:04:51,680 Speaker 1: circus will start to crumble when roasted by a guy 99 00:04:51,720 --> 00:04:54,440 Speaker 1: typing Garfield with a guarn and a bra into chat GPT, 100 00:04:54,839 --> 00:04:58,360 Speaker 1: which I have never done. Of course, Remember Blackwell is 101 00:04:58,360 --> 00:05:01,200 Speaker 1: supposed to be in videos big thing. It's meant to 102 00:05:01,240 --> 00:05:04,000 Speaker 1: represent a major leap forward in performance. If in Vidia 103 00:05:04,040 --> 00:05:06,720 Speaker 1: doesn't solve its cooling problem, and solve it well, its 104 00:05:06,760 --> 00:05:10,400 Speaker 1: customers will undoubtedly encounter thermal throttling, where the chip reduces 105 00:05:10,440 --> 00:05:13,240 Speaker 1: its speed in order to avoid causing any permanent damage. 106 00:05:13,520 --> 00:05:17,080 Speaker 1: It could eliminate some more all of the performance gains 107 00:05:17,200 --> 00:05:20,560 Speaker 1: obtained from the new architecture a new manufacturing process, despite 108 00:05:20,600 --> 00:05:23,440 Speaker 1: costing much much more, both for the chips itself and 109 00:05:23,480 --> 00:05:26,799 Speaker 1: the housing for them. In Vidio's problem isn't just bringing 110 00:05:26,880 --> 00:05:29,680 Speaker 1: these thermal performance issues under control, but both keeping them 111 00:05:29,720 --> 00:05:32,360 Speaker 1: under control and being able to educate their customers on 112 00:05:32,400 --> 00:05:35,400 Speaker 1: how to do so. In Vidia has, according to the information, 113 00:05:35,720 --> 00:05:39,000 Speaker 1: repeatedly tried to influence its customer server integrations to follow 114 00:05:39,040 --> 00:05:42,240 Speaker 1: its designs because it thinks it will lead to better performance. 115 00:05:42,279 --> 00:05:43,839 Speaker 1: But in this case one has to worry. If in 116 00:05:43,920 --> 00:05:47,760 Speaker 1: Vidia's black Well chips can be reliably cooled, maybe they 117 00:05:47,800 --> 00:05:50,360 Speaker 1: can in videos worked out crazy things in the past, 118 00:05:50,720 --> 00:05:52,479 Speaker 1: and while in video might be able to fix this 119 00:05:52,520 --> 00:05:54,919 Speaker 1: problem in isolation within its racks, it remains to be 120 00:05:54,920 --> 00:05:56,599 Speaker 1: seen how this works at scale as they ship and 121 00:05:56,600 --> 00:05:59,760 Speaker 1: integrate hundreds of thousands of Blackwell GPUs starting in the 122 00:05:59,760 --> 00:06:03,719 Speaker 1: front half of twenty twenty five. Things also get a 123 00:06:03,720 --> 00:06:06,000 Speaker 1: little worse when you realize how these chips are being 124 00:06:06,040 --> 00:06:09,760 Speaker 1: installed in these giant supercomputer data centers, where tens of 125 00:06:09,800 --> 00:06:12,160 Speaker 1: thousands are as many as one hundred thousand of them. 126 00:06:12,160 --> 00:06:14,960 Speaker 1: In the case of Elon Musk's Colossus Data center, which 127 00:06:15,000 --> 00:06:18,200 Speaker 1: is just very funny that that's what it's for. 128 00:06:18,279 --> 00:06:20,120 Speaker 2: It's for GROC on a dying social network. 129 00:06:20,600 --> 00:06:23,160 Speaker 1: But yeah, these GPUs, they're running concert to power generative 130 00:06:23,200 --> 00:06:23,800 Speaker 1: AI models. 131 00:06:23,839 --> 00:06:25,039 Speaker 2: That's all they're really for. 132 00:06:25,760 --> 00:06:27,760 Speaker 1: And The Wall Street Journal reported a few weeks ago 133 00:06:27,800 --> 00:06:31,440 Speaker 1: that building these vast data centers creates entirely new engineering challenges, 134 00:06:31,560 --> 00:06:33,720 Speaker 1: with one expert saying that big tech companies could be 135 00:06:33,800 --> 00:06:36,800 Speaker 1: spending as much of half of their capital expenditures on 136 00:06:36,880 --> 00:06:39,520 Speaker 1: replacing parts that are broken down, in large part because 137 00:06:39,520 --> 00:06:41,880 Speaker 1: these clusters are running on GPUs that are running at 138 00:06:41,880 --> 00:06:44,240 Speaker 1: full speed that need to all be cooled and need 139 00:06:44,279 --> 00:06:44,880 Speaker 1: to be cooled. 140 00:06:44,880 --> 00:06:45,159 Speaker 2: Well. 141 00:06:46,040 --> 00:06:49,159 Speaker 1: Remember the capital expenditures on generative AI and the associated 142 00:06:49,200 --> 00:06:52,120 Speaker 1: infrastructure have gone over two hundred billion dollars in the 143 00:06:52,160 --> 00:06:55,520 Speaker 1: last year. If half of that's dedicated to replacing broken shit. 144 00:06:55,560 --> 00:06:58,719 Speaker 1: What happens when there's no path of profitability in any 145 00:06:58,760 --> 00:07:02,000 Speaker 1: case is fine, they'll work it out well enough to 146 00:07:02,040 --> 00:07:04,320 Speaker 1: keep flogging these things. They've already made billions of dollars 147 00:07:04,400 --> 00:07:06,919 Speaker 1: selling Blackwells. They're sold out for a year, in fact, 148 00:07:07,000 --> 00:07:09,960 Speaker 1: and they'll continue to do so for now. But any 149 00:07:10,000 --> 00:07:12,679 Speaker 1: manufacturing and cooling issues are going to be very costly, 150 00:07:13,320 --> 00:07:15,480 Speaker 1: and even then, at some point somebody has to ask 151 00:07:15,480 --> 00:07:18,720 Speaker 1: the question, why do we need all these GPUs if 152 00:07:18,760 --> 00:07:22,560 Speaker 1: we've reached PKI. Despite the remarkable power of these chips. 153 00:07:22,640 --> 00:07:25,600 Speaker 1: In Vidia's entire enterprise GPU business model centers around the 154 00:07:25,640 --> 00:07:28,200 Speaker 1: idea that throwing more power at these problems will finally 155 00:07:28,200 --> 00:07:29,280 Speaker 1: create some solutions. 156 00:07:30,400 --> 00:07:31,880 Speaker 2: What if that isn't the case? 157 00:07:32,000 --> 00:07:37,720 Speaker 1: Though? So as ever, I'm kind of pissed off. I'm 158 00:07:37,760 --> 00:07:40,080 Speaker 1: pissed off reading this stuff. And it's not just because 159 00:07:40,080 --> 00:07:42,600 Speaker 1: I was saying this months ago. It's because all of 160 00:07:42,600 --> 00:07:44,560 Speaker 1: this money could have gone literally anywhere else. They could 161 00:07:44,560 --> 00:07:47,000 Speaker 1: have done something else, they could have I don't know, 162 00:07:47,480 --> 00:07:49,520 Speaker 1: I'm not running a big tech company, but I would 163 00:07:49,560 --> 00:07:52,160 Speaker 1: have put it into climate stuff or new battery technology. 164 00:07:52,680 --> 00:07:55,040 Speaker 1: Everyone's saying, oh, we'll make our own chips now, Yes, 165 00:07:55,080 --> 00:07:57,440 Speaker 1: see you in a few years, dickhead, Well, you go 166 00:07:57,520 --> 00:08:00,000 Speaker 1: to TSMC and ask if they've got any spare slot. 167 00:08:00,120 --> 00:08:02,760 Speaker 1: I'm sure they've got plenty. I'm sure that you can 168 00:08:02,840 --> 00:08:05,200 Speaker 1: just go and build a chip tomorrow. Sam Moultman talking 169 00:08:05,200 --> 00:08:09,880 Speaker 1: about building his own goddamn chips. What an absolute fast? 170 00:08:09,920 --> 00:08:10,280 Speaker 2: This is? 171 00:08:10,520 --> 00:08:13,960 Speaker 1: This fucking fast? All of this money burned in search 172 00:08:14,000 --> 00:08:16,360 Speaker 1: of a technology that only kind of works on something. 173 00:08:16,520 --> 00:08:20,200 Speaker 1: Sometimes it's just kind of a joke. The tech industry 174 00:08:20,240 --> 00:08:23,040 Speaker 1: is over leverage. They've doubled, they've tripled the kadrupled down 175 00:08:23,040 --> 00:08:25,680 Speaker 1: on generative AI now and this technology does not do 176 00:08:25,880 --> 00:08:27,720 Speaker 1: much more than it did a few months ago, And 177 00:08:27,760 --> 00:08:29,400 Speaker 1: I don't think it's going to do very much more 178 00:08:29,440 --> 00:08:32,760 Speaker 1: in a few months. Every single big tech companies pulled 179 00:08:32,760 --> 00:08:35,520 Speaker 1: tens of billions of dollars into building out these massive 180 00:08:35,559 --> 00:08:39,719 Speaker 1: superclusters with the intent of capturing AI demand. Yet they 181 00:08:39,760 --> 00:08:42,280 Speaker 1: never seem to think whether they were actually building things 182 00:08:42,280 --> 00:08:45,040 Speaker 1: that people wanted or would pay for, or that would 183 00:08:45,040 --> 00:08:47,960 Speaker 1: make the money, or that would help you manity in 184 00:08:48,000 --> 00:08:50,760 Speaker 1: any way, or really anything to do with an outcome 185 00:08:50,800 --> 00:08:53,960 Speaker 1: other than line go up. While some have claimed that 186 00:08:54,000 --> 00:08:58,120 Speaker 1: agents are the next frontier AI, Agents frontier Jesus, Nope, 187 00:08:58,240 --> 00:09:00,439 Speaker 1: keep him, The reality is that agents may be the 188 00:09:00,559 --> 00:09:04,199 Speaker 1: last generative AI product. Multiple large language models and integrations 189 00:09:04,240 --> 00:09:06,520 Speaker 1: bouncing off of each other in attempt to simulate what 190 00:09:06,559 --> 00:09:08,480 Speaker 1: a human might do at a cost that won't be 191 00:09:08,480 --> 00:09:12,040 Speaker 1: sustainable for the majority of businesses. While Anthropics demo of 192 00:09:12,080 --> 00:09:15,160 Speaker 1: its alleged model that can control a few browser windows 193 00:09:15,160 --> 00:09:17,840 Speaker 1: with a prompt might have seemed impressive to credulous people, 194 00:09:18,520 --> 00:09:22,160 Speaker 1: specifically mean casing you. And here these were controlled demos 195 00:09:22,160 --> 00:09:25,440 Speaker 1: which Anthropic added with slow and make lots of mistakes. Hey, 196 00:09:25,559 --> 00:09:28,079 Speaker 1: almost like it's hallucinating. I sure hope they fixed that 197 00:09:28,200 --> 00:09:30,360 Speaker 1: unfixable problem I'd be talking about for months. 198 00:09:30,440 --> 00:09:31,080 Speaker 2: Jesus Christ. 199 00:09:32,360 --> 00:09:36,120 Speaker 1: Even if it does, Anthropic has now successfully replaced an 200 00:09:36,240 --> 00:09:40,079 Speaker 1: entry full data worker position at an indeterminate cost, and 201 00:09:40,280 --> 00:09:43,320 Speaker 1: likely an unprofitable one too. And in many organizations those 202 00:09:43,400 --> 00:09:45,840 Speaker 1: jobs have already been outsourced or automated or start with 203 00:09:45,920 --> 00:09:48,320 Speaker 1: cheap contractors. So the people who are going to lose 204 00:09:48,360 --> 00:09:50,520 Speaker 1: out here are people who are already getting shot on 205 00:09:50,559 --> 00:09:51,120 Speaker 1: by life. 206 00:09:51,320 --> 00:09:52,160 Speaker 2: It really sucks. 207 00:09:52,679 --> 00:09:55,880 Speaker 1: And the obscenity of this mass delusion, it's just nauseating. 208 00:09:56,080 --> 00:09:58,160 Speaker 1: It's a monolith to bad decision making and the herd 209 00:09:58,160 --> 00:10:01,440 Speaker 1: mentality of tech's most powerful people, as well as an 210 00:10:01,480 --> 00:10:04,160 Speaker 1: outright attempt to manipulate the media into believing something was 211 00:10:04,200 --> 00:10:07,280 Speaker 1: possible that never was, and the media bought it, hook 212 00:10:07,360 --> 00:10:08,240 Speaker 1: line and sinker. 213 00:10:09,120 --> 00:10:09,920 Speaker 2: You all bought it. 214 00:10:10,920 --> 00:10:13,480 Speaker 1: Hundreds of billions of dollars have been wasted building giant 215 00:10:13,559 --> 00:10:16,240 Speaker 1: data centers to crunch numbers for software that has no 216 00:10:16,320 --> 00:10:18,640 Speaker 1: real product market fit, or while trying to hammer it 217 00:10:18,640 --> 00:10:21,760 Speaker 1: into various shapes to make it pretend that it's alive, conscious, 218 00:10:21,840 --> 00:10:25,360 Speaker 1: or even fucking useful. There is no path, from what 219 00:10:25,440 --> 00:10:27,880 Speaker 1: I can see, to turning generative AI from what it 220 00:10:27,920 --> 00:10:31,520 Speaker 1: is today into anything resembling a sustainable business. And the 221 00:10:31,559 --> 00:10:34,760 Speaker 1: only path that big tech appeared to have seen or 222 00:10:34,840 --> 00:10:37,240 Speaker 1: thought about was to throw as much money, power and 223 00:10:37,320 --> 00:10:40,320 Speaker 1: data at the problem. An avenue that appeared to be 224 00:10:40,440 --> 00:10:42,880 Speaker 1: and definitely appears to be now a complete dead end, 225 00:10:43,640 --> 00:10:47,120 Speaker 1: and we're still nothing's really come out of this movement. 226 00:10:47,960 --> 00:10:50,640 Speaker 1: I've used a handful of AI products that I found useful, 227 00:10:50,720 --> 00:10:54,320 Speaker 1: and aipower journal, for example, but these are not products 228 00:10:54,320 --> 00:10:57,560 Speaker 1: that one associates with revolutions. They're useful tools that would 229 00:10:57,600 --> 00:11:00,000 Speaker 1: have been a welcome surprised if it didn't require burning 230 00:11:00,120 --> 00:11:03,080 Speaker 1: billions of dollars, blowing past emissions targets, and stealing the 231 00:11:03,080 --> 00:11:06,640 Speaker 1: creative works of millions of people to train them. It's 232 00:11:06,679 --> 00:11:10,360 Speaker 1: just sickening. It's sickening because this did not have to happen. 233 00:11:10,520 --> 00:11:13,320 Speaker 1: Had suned up Ashay seen what Microsoft did with open 234 00:11:13,360 --> 00:11:16,520 Speaker 1: AI and said not for me, mate, That's definitely not 235 00:11:16,559 --> 00:11:19,440 Speaker 1: how he talks. But nevertheless, none of this would have continued. 236 00:11:19,559 --> 00:11:21,360 Speaker 1: You think Microsoft was going to hold this shit up 237 00:11:21,360 --> 00:11:24,880 Speaker 1: on their own. God, No, they have no juice such 238 00:11:24,920 --> 00:11:27,640 Speaker 1: an Adela. He's just a he's a cult leader, something 239 00:11:27,679 --> 00:11:30,400 Speaker 1: I'm going to get into in a later episode. And 240 00:11:30,440 --> 00:11:33,160 Speaker 1: it's just it's all very sad. But if you want 241 00:11:33,160 --> 00:11:35,520 Speaker 1: to cheer yourself up, maybe you could listen to one 242 00:11:35,520 --> 00:11:38,200 Speaker 1: of the following advertisements. And this is a short episode, 243 00:11:38,240 --> 00:11:39,920 Speaker 1: so if they don't put one after this, you might 244 00:11:39,960 --> 00:11:43,560 Speaker 1: just have a short gap between me talking. That's up 245 00:11:43,600 --> 00:11:50,800 Speaker 1: to Matt. MAT's a genius. That's my producer, by the way, Mattasowski. Anyway, ads, 246 00:11:59,160 --> 00:12:04,079 Speaker 1: and we're back. Look, I truly don't know what happens next, 247 00:12:04,160 --> 00:12:05,920 Speaker 1: but I'm going to walk you through what I'm thinking. 248 00:12:06,640 --> 00:12:08,480 Speaker 2: If we're truly at the diminishing. 249 00:12:08,040 --> 00:12:10,720 Speaker 1: Return stage of transformer based models, it's going to be 250 00:12:10,760 --> 00:12:14,800 Speaker 1: extremely difficult to justify buying further iterations of Nvidia GPUs 251 00:12:14,840 --> 00:12:19,199 Speaker 1: past Blackwell, and also I have doubts they're going to 252 00:12:19,240 --> 00:12:21,719 Speaker 1: be able to make a new one every year. The 253 00:12:21,920 --> 00:12:24,679 Speaker 1: entire generative AI movement lives and dies by the idea 254 00:12:24,720 --> 00:12:27,079 Speaker 1: that more compute power and more training data makes these 255 00:12:27,120 --> 00:12:29,720 Speaker 1: things better. And if that's no longer the case, there's 256 00:12:29,960 --> 00:12:32,120 Speaker 1: not really a reason to keep buying bigger and better. 257 00:12:32,640 --> 00:12:36,760 Speaker 1: What's the point even now? What exactly happens when Microsoft 258 00:12:36,880 --> 00:12:39,600 Speaker 1: or Google has racks worth of Blackwell GPUs? What does 259 00:12:39,600 --> 00:12:43,640 Speaker 1: aren't getting better faster? I guess what does faster even mean? 260 00:12:43,760 --> 00:12:46,480 Speaker 1: What does better even mean? Better has yet to me 261 00:12:46,520 --> 00:12:49,800 Speaker 1: in a new product? Better has not seem to be 262 00:12:50,200 --> 00:12:54,680 Speaker 1: measurable by humans just by bullshit benchmarks. This also makes 263 00:12:54,720 --> 00:12:58,120 Speaker 1: the lives of open Ai and Anthropic a little more difficult. 264 00:12:58,440 --> 00:13:00,760 Speaker 1: Sam Altman has grown rich in power for lying about 265 00:13:00,800 --> 00:13:04,480 Speaker 1: how GPT will somehow lead to AGI some conscious computer. 266 00:13:04,920 --> 00:13:09,080 Speaker 1: But at this point, what exactly is open ai meant 267 00:13:09,080 --> 00:13:11,720 Speaker 1: to do? What are they doing? The only way it's 268 00:13:12,160 --> 00:13:14,480 Speaker 1: this company has ever been able to develop new models 269 00:13:14,559 --> 00:13:17,079 Speaker 1: is by throwing masses of compute and training data at them. 270 00:13:17,400 --> 00:13:19,839 Speaker 1: And there are only other choices to start stapling its 271 00:13:20,040 --> 00:13:23,120 Speaker 1: their reasoning model onto the side of their large language model, 272 00:13:23,240 --> 00:13:29,200 Speaker 1: at which point something happens something everyone It's happening something 273 00:13:29,400 --> 00:13:31,880 Speaker 1: that is so good that literally nobody working for open 274 00:13:31,920 --> 00:13:33,719 Speaker 1: aire in the media appears to be able to tell 275 00:13:33,720 --> 00:13:34,200 Speaker 1: you what it is. 276 00:13:35,160 --> 00:13:36,679 Speaker 2: Yeah. 277 00:13:36,760 --> 00:13:39,680 Speaker 1: Putting that aside, open ai is also a terrible business 278 00:13:40,160 --> 00:13:42,520 Speaker 1: that has to burn five billion dollars to make three 279 00:13:42,520 --> 00:13:45,040 Speaker 1: point seven billion dollars, and there's no proof that they're 280 00:13:45,080 --> 00:13:47,840 Speaker 1: able to bring down their costs. The constant thing I 281 00:13:47,840 --> 00:13:50,720 Speaker 1: hear from VC's and AI fantasist is that the chips 282 00:13:50,720 --> 00:13:53,600 Speaker 1: will bring down the cost of inference. Yet I don't 283 00:13:53,640 --> 00:13:57,400 Speaker 1: see that happening. I've yet to see that happening. There's 284 00:13:57,440 --> 00:14:00,040 Speaker 1: cerebrus dinner plate sized chips, but you know what, I 285 00:14:00,160 --> 00:14:02,520 Speaker 1: just don't think that shit's going to scale. And at 286 00:14:02,559 --> 00:14:05,400 Speaker 1: that point you have to wonder do they have like 287 00:14:05,640 --> 00:14:08,880 Speaker 1: where is this going? I know you're listening to an 288 00:14:08,920 --> 00:14:12,160 Speaker 1: episode where I tell you, but still even reading this 289 00:14:12,240 --> 00:14:14,800 Speaker 1: stuff all day, you just look at it too hard 290 00:14:15,040 --> 00:14:18,920 Speaker 1: and you start feeling a little crazy. I've read pretty 291 00:14:18,960 --> 00:14:21,680 Speaker 1: much every article on open ai at this point. I've 292 00:14:21,680 --> 00:14:24,880 Speaker 1: read Microsoft's earnings multiple quarters straight. Same with the rest 293 00:14:24,880 --> 00:14:27,000 Speaker 1: of them. I've read all their blog posts, I've read 294 00:14:27,000 --> 00:14:30,359 Speaker 1: all their shit. They don't have any idea. I'm confident 295 00:14:30,440 --> 00:14:33,480 Speaker 1: that they don't have any idea. It's terrifying and this 296 00:14:33,600 --> 00:14:36,880 Speaker 1: is just this dismal situation where the only other there's 297 00:14:36,920 --> 00:14:40,000 Speaker 1: really only a few options. If I'm honest, you stop now. 298 00:14:40,720 --> 00:14:45,160 Speaker 1: That is the biggest one. You just stop. Someone has 299 00:14:45,200 --> 00:14:48,200 Speaker 1: to admit, Microsoft, Google, They'll never do this. They need 300 00:14:48,240 --> 00:14:50,640 Speaker 1: to go this is not going to scale. We need 301 00:14:50,680 --> 00:14:52,560 Speaker 1: to pair this back or we need to stop it. 302 00:14:53,240 --> 00:14:55,800 Speaker 1: And it will hurt their stock, it will hurt their media. 303 00:14:56,040 --> 00:14:58,480 Speaker 1: But it's the right thing to do for the environment alone, 304 00:14:58,760 --> 00:15:03,360 Speaker 1: but as a sustainable busines. But assuming they don't do that, 305 00:15:03,880 --> 00:15:08,040 Speaker 1: we have another problem, which is the only other option 306 00:15:08,160 --> 00:15:10,880 Speaker 1: they can have is to keep flooring the gas. It 307 00:15:10,920 --> 00:15:13,800 Speaker 1: costs one hundred million dollars to train GPT four to zero, 308 00:15:14,000 --> 00:15:17,000 Speaker 1: and anthropic CEO Dario ami Day estimated a few months 309 00:15:17,040 --> 00:15:20,000 Speaker 1: ago that training future models would cost one billion, if 310 00:15:20,040 --> 00:15:23,080 Speaker 1: not ten billion dollars, with one researcher claiming that training 311 00:15:23,160 --> 00:15:26,760 Speaker 1: open AI's next model, GPT five will cost around a 312 00:15:26,760 --> 00:15:29,320 Speaker 1: billion dollars outside of a miracle. 313 00:15:29,320 --> 00:15:32,200 Speaker 2: We're about to enter an era of desperation in generative AI. 314 00:15:32,640 --> 00:15:34,760 Speaker 1: We're two years in and we have no killer apps, 315 00:15:34,760 --> 00:15:38,080 Speaker 1: no industry defining products other than chat GPT, a product 316 00:15:38,120 --> 00:15:40,520 Speaker 1: that burns billions of doddlers, and nobody. 317 00:15:40,200 --> 00:15:41,920 Speaker 2: Can really describe every episode. 318 00:15:41,960 --> 00:15:44,600 Speaker 1: I'm like, email me, tell me what chat GPT is, 319 00:15:44,800 --> 00:15:48,880 Speaker 1: and I'll get some smart us who sends me three paragraphs. No, no, no, no, no, 320 00:15:49,200 --> 00:15:51,680 Speaker 1: that's not a description, that's an essay. May you actually 321 00:15:51,680 --> 00:15:53,320 Speaker 1: need to tell me what the hell this is? And 322 00:15:53,400 --> 00:15:55,720 Speaker 1: if you can't, you've proven my point in my extremely 323 00:15:55,760 --> 00:16:00,280 Speaker 1: thin and rigged game. Anyway, neither Microsoft, nor Meta, nor Google, 324 00:16:00,320 --> 00:16:02,000 Speaker 1: no Amazon seemed to be able to come up with 325 00:16:02,080 --> 00:16:05,280 Speaker 1: any profitable use cases, let alone one their users actually like, 326 00:16:05,560 --> 00:16:07,560 Speaker 1: Nor have any of the people that have raised billions 327 00:16:07,560 --> 00:16:09,720 Speaker 1: of dollars in venture capital for anything with AI take 328 00:16:09,800 --> 00:16:13,240 Speaker 1: to the side. An investor interest in AI is already cooling. 329 00:16:13,280 --> 00:16:17,280 Speaker 1: According to the information, It's kind of unclear how far 330 00:16:17,400 --> 00:16:20,040 Speaker 1: this fast goes from here, if only because it isn't 331 00:16:20,080 --> 00:16:22,480 Speaker 1: obvious what it is that anybody gets by investing in 332 00:16:22,520 --> 00:16:25,560 Speaker 1: future rounds of open Aianthropic or any of these other 333 00:16:25,600 --> 00:16:29,000 Speaker 1: generative AI companies. At some point, they must make money, 334 00:16:29,080 --> 00:16:31,440 Speaker 1: and the entire dream has been built around the idea 335 00:16:31,480 --> 00:16:33,200 Speaker 1: that all of these GPUs and all of this money 336 00:16:33,400 --> 00:16:38,960 Speaker 1: would eventually spit out something, something revolutionary, something world changing. 337 00:16:40,200 --> 00:16:43,280 Speaker 1: What we have is this clunky, ugly, messy, lasceness, environmentally 338 00:16:43,280 --> 00:16:45,080 Speaker 1: destructive and mediocre. 339 00:16:44,560 --> 00:16:45,280 Speaker 2: Piece of shit. 340 00:16:46,240 --> 00:16:48,760 Speaker 1: Generative AI was a reckless pursuit, one that shows a 341 00:16:48,760 --> 00:16:50,960 Speaker 1: total lack of creativity and sense in the minds of 342 00:16:51,000 --> 00:16:53,200 Speaker 1: big tech in venture capital, one where there was never 343 00:16:53,320 --> 00:16:55,920 Speaker 1: anything really impressive other than the amount of money it 344 00:16:55,920 --> 00:16:57,800 Speaker 1: could burn on the amount of time samal And could 345 00:16:57,800 --> 00:17:00,800 Speaker 1: say something stupid and get quoted for it. I'll be 346 00:17:00,800 --> 00:17:03,920 Speaker 1: honest with you, I don't really know what happens here. 347 00:17:04,400 --> 00:17:06,840 Speaker 1: The future was always one that demanded big tech spend 348 00:17:06,840 --> 00:17:08,600 Speaker 1: more to make even bigger models that would at some 349 00:17:08,600 --> 00:17:11,840 Speaker 1: point become useful, and it isn't happening. In pursuit of 350 00:17:11,880 --> 00:17:14,359 Speaker 1: doing so, big tech invested hundreds of billions of dollars 351 00:17:14,400 --> 00:17:17,920 Speaker 1: into infrastructure specifically to follow one goal and put AI 352 00:17:18,040 --> 00:17:20,280 Speaker 1: front and center of their businesses, claiming it was the 353 00:17:20,280 --> 00:17:24,280 Speaker 1: future without ever checking if it was, if it actually 354 00:17:24,320 --> 00:17:28,600 Speaker 1: did anything useful. And really they, as I've said in 355 00:17:28,600 --> 00:17:30,440 Speaker 1: the rock com bubble a few months ago, they don't 356 00:17:30,520 --> 00:17:33,679 Speaker 1: have anything else. They don't have another growth market, and 357 00:17:33,720 --> 00:17:35,800 Speaker 1: maybe that's why they're shoving all the cash into this, 358 00:17:36,119 --> 00:17:37,000 Speaker 1: because they don't. 359 00:17:36,760 --> 00:17:38,159 Speaker 2: Have it anywhere. Else to put it. 360 00:17:38,359 --> 00:17:40,760 Speaker 1: They don't want to admit they've got nothing else. They 361 00:17:40,760 --> 00:17:43,080 Speaker 1: don't want to admit that their businesses are crumbling under 362 00:17:43,480 --> 00:17:47,520 Speaker 1: scrutiny from antitrust. They don't want to admit their services 363 00:17:47,560 --> 00:17:50,680 Speaker 1: are kind of deteriorating, especially in the case of Meta. 364 00:17:50,800 --> 00:17:53,760 Speaker 1: They don't have anything else. The revenue isn't coming, the 365 00:17:53,800 --> 00:17:58,040 Speaker 1: products aren't coming. Oryan open AI's next model, it will underwhelm, 366 00:17:58,119 --> 00:18:00,119 Speaker 1: as will its competitors model, and at some point, think 367 00:18:00,200 --> 00:18:02,560 Speaker 1: somebody is going to blink in one of the hyperscalers 368 00:18:02,560 --> 00:18:06,399 Speaker 1: and the AI era will end. Almost every single GENERAI 369 00:18:06,520 --> 00:18:09,040 Speaker 1: company that you've heard of is deeply unprofitable, and there 370 00:18:09,040 --> 00:18:11,639 Speaker 1: are very few innovations coming to save them from the 371 00:18:11,680 --> 00:18:14,960 Speaker 1: atrophy of the Foundation model system, which by the way, 372 00:18:15,400 --> 00:18:17,760 Speaker 1: is the main large language models that they're training in 373 00:18:17,800 --> 00:18:21,960 Speaker 1: all of these cases. Now, I've tried to keep my 374 00:18:22,000 --> 00:18:24,400 Speaker 1: emotions in check, but you might you might be able 375 00:18:24,400 --> 00:18:26,760 Speaker 1: to tell that this pisses me off a little. I 376 00:18:26,840 --> 00:18:30,040 Speaker 1: just feel sad and exhausted about it all. I feel 377 00:18:30,160 --> 00:18:32,000 Speaker 1: drained as I look at how many times I've tried 378 00:18:32,000 --> 00:18:34,399 Speaker 1: to warm people. I'm pissed off at the many members 379 00:18:34,400 --> 00:18:36,000 Speaker 1: of the media that failed to push back against the 380 00:18:36,040 --> 00:18:38,760 Speaker 1: over promises and outright lies of people like Sam Altman 381 00:18:38,840 --> 00:18:41,600 Speaker 1: and Dario Ami Day. Warrio ami Day is what I'm 382 00:18:41,640 --> 00:18:44,159 Speaker 1: calling him now. It's not particularly funny ira acurate, but 383 00:18:44,160 --> 00:18:46,320 Speaker 1: it's funny to say for me. And I'm just full 384 00:18:46,359 --> 00:18:48,679 Speaker 1: of dread as I consider the economic ramifications of this 385 00:18:48,720 --> 00:18:51,280 Speaker 1: industry collapsing, as well as the damage it's already done 386 00:18:51,400 --> 00:18:54,720 Speaker 1: to our fucking environment. Once the AI bubble pops, there 387 00:18:54,760 --> 00:18:57,720 Speaker 1: are no hyper growth markets left, which will in turn 388 00:18:57,800 --> 00:18:59,679 Speaker 1: lead to a blood bath in big tech stocks as 389 00:18:59,680 --> 00:19:02,000 Speaker 1: they really that they're out of big ideas to convince 390 00:19:02,040 --> 00:19:05,000 Speaker 1: the street that they're going to grow forever. There are 391 00:19:05,040 --> 00:19:07,360 Speaker 1: some that will boast about being right here, and yes, 392 00:19:07,400 --> 00:19:10,840 Speaker 1: there is some satisfaction in being so, of course, of 393 00:19:10,880 --> 00:19:12,960 Speaker 1: course you take the moment, you do the victory lap. 394 00:19:13,040 --> 00:19:16,159 Speaker 1: Whatever I said a thing, and I was right. I 395 00:19:16,160 --> 00:19:18,560 Speaker 1: don't know if I'm I'm really doing the joker kick 396 00:19:18,640 --> 00:19:20,720 Speaker 1: down the stairs here. I don't know if I'm really 397 00:19:21,280 --> 00:19:24,560 Speaker 1: happy about being right here. I think generative AI is 398 00:19:24,600 --> 00:19:26,439 Speaker 1: a huge waste of money. I think the damage it's 399 00:19:26,480 --> 00:19:29,800 Speaker 1: doing to the environment is disgusting. I think the stealing 400 00:19:30,000 --> 00:19:33,280 Speaker 1: required to make the training data side is disgusting. I 401 00:19:33,280 --> 00:19:35,880 Speaker 1: think all of that ending is good, and I think 402 00:19:35,920 --> 00:19:37,920 Speaker 1: it needs to just to be clear or I sit. 403 00:19:38,920 --> 00:19:43,360 Speaker 1: But you know what, the other problem is that when 404 00:19:43,400 --> 00:19:45,400 Speaker 1: this all falls apart, it's going to be rough. It's 405 00:19:45,440 --> 00:19:47,240 Speaker 1: going to be rough for people in the tech industry, 406 00:19:47,440 --> 00:20:03,440 Speaker 1: and it's going to be rough for the economy. So 407 00:20:03,440 --> 00:20:05,240 Speaker 1: I'm gonna start wrapping up here. I'm gonna read you 408 00:20:05,280 --> 00:20:07,320 Speaker 1: a quote from Bubble Trouble, a piece I wrote in April. 409 00:20:08,600 --> 00:20:13,040 Speaker 1: Hum how do you solve all of these incredibly difficult problems? 410 00:20:13,440 --> 00:20:15,640 Speaker 1: What does open ai or Anthropic do when they run 411 00:20:15,640 --> 00:20:18,240 Speaker 1: out of data and the synthetic data doesn't fill the gap, 412 00:20:18,320 --> 00:20:21,239 Speaker 1: or worse, massively degrades the quality of their outputs. What 413 00:20:21,280 --> 00:20:24,120 Speaker 1: does Sam Moltman do if GPT five, like GPT four, 414 00:20:24,400 --> 00:20:27,480 Speaker 1: doesn't significantly improve its performance and he can't find enough 415 00:20:27,480 --> 00:20:30,040 Speaker 1: compute to make the next step. What do open ai 416 00:20:30,119 --> 00:20:32,480 Speaker 1: and Anthropic do when they realize they'll likely never turn 417 00:20:32,480 --> 00:20:35,560 Speaker 1: a profit. What does Microsoft, gra Amazon or Google do 418 00:20:35,600 --> 00:20:37,320 Speaker 1: if demand never really takes off and they' left with 419 00:20:37,359 --> 00:20:40,760 Speaker 1: billions of dollars of underutilized data centers. What does Nvidia 420 00:20:40,840 --> 00:20:42,600 Speaker 1: do if the demand for its chips drops off a 421 00:20:42,600 --> 00:20:45,680 Speaker 1: cliff as a result. I don't know why more people 422 00:20:45,720 --> 00:20:48,720 Speaker 1: aren't screaming from the rooftops about how unsustainable the AI 423 00:20:48,760 --> 00:20:51,600 Speaker 1: boom is and how impossible some of the challenges are 424 00:20:51,600 --> 00:20:54,960 Speaker 1: that it faces. There is no way to create enough 425 00:20:55,000 --> 00:20:57,840 Speaker 1: training data for these models, and little that we've seen 426 00:20:57,920 --> 00:21:00,760 Speaker 1: so far suggests that generative AI will make anybody but 427 00:21:00,880 --> 00:21:04,159 Speaker 1: Nvidio money. We're reaching the point where physics, things like 428 00:21:04,160 --> 00:21:06,439 Speaker 1: heat and electricity are getting in the way of progressing 429 00:21:06,560 --> 00:21:09,800 Speaker 1: much further, and it's hard to stomach investing more considering 430 00:21:09,840 --> 00:21:12,040 Speaker 1: where we're at right now. Once you cut through the noise, 431 00:21:13,080 --> 00:21:16,959 Speaker 1: it just seems kind of fairly goddamn mediocre. There's no 432 00:21:17,000 --> 00:21:20,840 Speaker 1: iPhone moment coming, I'm afraid, so all right, not gonna 433 00:21:20,880 --> 00:21:22,280 Speaker 1: be too smug, but reading that. 434 00:21:22,320 --> 00:21:24,320 Speaker 2: Back pretty accurate, Pretty good for April. 435 00:21:25,520 --> 00:21:28,800 Speaker 1: I was right then, I'm right now. Generative AI isn't 436 00:21:28,800 --> 00:21:31,640 Speaker 1: a revolution. It's an evolution of a tech industry overtaken 437 00:21:31,640 --> 00:21:34,320 Speaker 1: by growth hungry management consultant freaks that I neither know 438 00:21:34,400 --> 00:21:37,080 Speaker 1: the problems that real people face nor how to fix them. 439 00:21:37,560 --> 00:21:38,280 Speaker 2: It's a waste. 440 00:21:38,440 --> 00:21:41,359 Speaker 1: It's a sickening waste, a monument to the corrupting force 441 00:21:41,400 --> 00:21:43,520 Speaker 1: of growth and a sign that people in power and 442 00:21:43,600 --> 00:21:45,960 Speaker 1: tech no longer work for you, the customer, but for 443 00:21:46,040 --> 00:21:49,679 Speaker 1: venture capitalists and for the markets. I also want to 444 00:21:49,760 --> 00:21:52,040 Speaker 1: be clear that none of these companies seem to have 445 00:21:52,080 --> 00:21:55,000 Speaker 1: ever had a plan. They believed that if they threw 446 00:21:55,119 --> 00:21:58,000 Speaker 1: enough GPUs together, they would turn generative AI with a 447 00:21:58,000 --> 00:22:02,199 Speaker 1: probabilistic model for generating stef either into a sentient computer 448 00:22:02,280 --> 00:22:04,600 Speaker 1: in open AI's case, or into some sort of useful 449 00:22:04,600 --> 00:22:08,520 Speaker 1: product in Microsoft's case. Neither of them are right, And 450 00:22:08,600 --> 00:22:11,000 Speaker 1: I get it. I get some of you that you're like, well, 451 00:22:11,280 --> 00:22:13,200 Speaker 1: this is all a big plan, so they can proliferate 452 00:22:13,280 --> 00:22:15,480 Speaker 1: data centers. This is all a big plans. They can 453 00:22:15,480 --> 00:22:17,919 Speaker 1: steal everything and ship it back to us, so they 454 00:22:17,960 --> 00:22:22,080 Speaker 1: can do this to us. It's much easier and more 455 00:22:22,080 --> 00:22:24,760 Speaker 1: comfortable to look at the world as a series of conspiracies, 456 00:22:24,800 --> 00:22:28,520 Speaker 1: these grand strategies, and that all of these people are powerful, 457 00:22:28,800 --> 00:22:31,240 Speaker 1: which they are, but powerful in a way that is 458 00:22:31,320 --> 00:22:34,040 Speaker 1: kind of the dread hand controlling your everything. 459 00:22:35,080 --> 00:22:35,960 Speaker 2: But honestly, it's. 460 00:22:35,880 --> 00:22:39,240 Speaker 1: Far scarier to see reality for what it is. Extremely 461 00:22:39,359 --> 00:22:41,400 Speaker 1: rich and powerful people that are willing to bed insanely 462 00:22:41,480 --> 00:22:43,680 Speaker 1: large amounts of money on what amounts to a few PDFs. 463 00:22:43,680 --> 00:22:48,040 Speaker 1: In their fucking gut, these people have no plan. This 464 00:22:48,200 --> 00:22:50,960 Speaker 1: is not big takes big plan, or they're excused to 465 00:22:50,960 --> 00:22:53,399 Speaker 1: build more data centers. It's the death throws of twenty 466 00:22:53,480 --> 00:22:56,120 Speaker 1: years of growth or costs thinking, because throwing a bunch 467 00:22:56,160 --> 00:22:58,480 Speaker 1: of money at more servers and more engineers always seem 468 00:22:58,520 --> 00:22:59,120 Speaker 1: to create more. 469 00:22:59,040 --> 00:22:59,840 Speaker 2: Growth in the past. 470 00:23:00,480 --> 00:23:02,720 Speaker 1: In practice, this means that the people in charge and 471 00:23:02,760 --> 00:23:05,040 Speaker 1: the strategies they employ are born not of an interest 472 00:23:05,040 --> 00:23:07,280 Speaker 1: in improving the lives of you their customer, but in 473 00:23:07,280 --> 00:23:10,600 Speaker 1: increasing revenue growth, which means products they create aren't really 474 00:23:10,640 --> 00:23:14,000 Speaker 1: about solving any problem other than what will make somebody 475 00:23:14,040 --> 00:23:17,360 Speaker 1: give me more money, which doesn't necessarily mean provide them 476 00:23:17,359 --> 00:23:21,439 Speaker 1: with a service. Generative AI is the perfect monster of 477 00:23:21,440 --> 00:23:24,240 Speaker 1: the row economy. It's a technology that lacks any real purpose, 478 00:23:24,280 --> 00:23:26,879 Speaker 1: sold as if it could do literally anything, one without 479 00:23:26,880 --> 00:23:29,359 Speaker 1: a real business model or a killer app, proliferated because 480 00:23:29,359 --> 00:23:34,000 Speaker 1: big tech no longer innovates, They clone, and they monopolize. Yes, 481 00:23:34,240 --> 00:23:37,119 Speaker 1: this much money can be this stupid, and yes they 482 00:23:37,160 --> 00:23:39,760 Speaker 1: will burn billions more in pursuit of a non specific 483 00:23:39,840 --> 00:23:42,280 Speaker 1: dream that involves charging you more money and trapping you 484 00:23:42,320 --> 00:23:46,080 Speaker 1: in their ecosystem. But they're out of ideas. This was 485 00:23:46,119 --> 00:23:48,439 Speaker 1: their big thing. If they had anything else, they wouldn't 486 00:23:48,440 --> 00:23:48,840 Speaker 1: have done this. 487 00:23:49,400 --> 00:23:51,680 Speaker 2: They don't. They're kind of screwed. 488 00:23:52,200 --> 00:23:54,480 Speaker 1: I don't know how long it will take for you 489 00:23:54,600 --> 00:23:57,400 Speaker 1: to see how much they're screwed. Could be a year, 490 00:23:57,680 --> 00:24:01,000 Speaker 1: could be two quarters. I honestly don't know. But it 491 00:24:01,080 --> 00:24:04,800 Speaker 1: cannot go on like this. It cannot because at some point, 492 00:24:04,840 --> 00:24:08,320 Speaker 1: burning billions of dollars every month is not going to work. 493 00:24:08,480 --> 00:24:11,080 Speaker 1: At some point, the markets will react, or at some point, 494 00:24:11,520 --> 00:24:14,480 Speaker 1: I don't know, CFO of Microsoft Damian Hood might say, hey, 495 00:24:14,520 --> 00:24:17,200 Speaker 1: we're burning billions of dollars to lose billions of dollars. 496 00:24:17,520 --> 00:24:19,960 Speaker 1: We don't have the customers, these products aren't useful enough, 497 00:24:20,680 --> 00:24:22,480 Speaker 1: or maybe she keeps driving to hell. 498 00:24:23,200 --> 00:24:23,760 Speaker 2: I don't know. 499 00:24:24,400 --> 00:24:26,399 Speaker 1: But the longer it takes, the worse it's going to 500 00:24:26,480 --> 00:24:29,760 Speaker 1: be for the industry and for the markets. I'm not 501 00:24:29,760 --> 00:24:31,960 Speaker 1: trying to be a doom set, just like I wasn't 502 00:24:32,000 --> 00:24:34,399 Speaker 1: trying to be one in March. I believe all of 503 00:24:34,400 --> 00:24:37,560 Speaker 1: this is going nowhere, and that at some point, Google, Microsoft, Meta, 504 00:24:38,240 --> 00:24:39,520 Speaker 1: one of them is going to blink and they're going 505 00:24:39,520 --> 00:24:42,240 Speaker 1: to pull back on the capex. And before then, you're 506 00:24:42,240 --> 00:24:44,440 Speaker 1: going to see a lot of desperate stories about how 507 00:24:44,480 --> 00:24:47,240 Speaker 1: AI gains can be found outside of training you models 508 00:24:47,240 --> 00:24:49,960 Speaker 1: to try and keep the party going despite reality, flicking 509 00:24:50,000 --> 00:24:51,720 Speaker 1: the lights on and off and threatening to call the 510 00:24:51,760 --> 00:24:54,600 Speaker 1: police on them. Really, though, you're going to see so 511 00:24:54,680 --> 00:24:56,960 Speaker 1: much of that. You're going to see people in the 512 00:24:57,040 --> 00:25:01,479 Speaker 1: media who should know better still pumping, still pumping that bag. 513 00:25:01,840 --> 00:25:03,879 Speaker 1: Because it gets back to the conspiracy theory I was 514 00:25:03,920 --> 00:25:07,199 Speaker 1: talking about before it gets back there, because it's so 515 00:25:07,320 --> 00:25:10,800 Speaker 1: much easier to think, Ah, they'll work it out, they're smart. 516 00:25:11,200 --> 00:25:14,399 Speaker 1: I trust them. I trust these people. They're going to 517 00:25:14,480 --> 00:25:16,920 Speaker 1: work out and make Generative AI make so much money. 518 00:25:16,960 --> 00:25:19,040 Speaker 1: It's going to be so profitable. Alw it's going to 519 00:25:19,080 --> 00:25:21,880 Speaker 1: be so good. Because when you don't accept that premise, 520 00:25:21,960 --> 00:25:24,960 Speaker 1: everything you write about it feels like writing about a 521 00:25:25,040 --> 00:25:29,440 Speaker 1: dying person. Oh, anthropic claims it can write in your 522 00:25:29,800 --> 00:25:32,480 Speaker 1: writing style. Now, great, where's anthropic going to be in 523 00:25:32,520 --> 00:25:35,720 Speaker 1: two years? The toilet probably going to need a pretty 524 00:25:35,720 --> 00:25:39,040 Speaker 1: big toilet for all those GPUs. Oh open AI. They 525 00:25:39,080 --> 00:25:41,560 Speaker 1: have this idea, Yeah, their ideas have fucking suck, so 526 00:25:41,680 --> 00:25:44,919 Speaker 1: far and the irony is large. Language models as an 527 00:25:44,960 --> 00:25:50,240 Speaker 1: idea aren't a bad one. The concept is interesting, a word, calculator, sexy, orto, correct, 528 00:25:50,280 --> 00:25:52,360 Speaker 1: whatever you call it. There are meaningful things you could 529 00:25:52,359 --> 00:25:55,560 Speaker 1: build with that, but not at this cost, not like this, 530 00:25:56,280 --> 00:25:59,720 Speaker 1: not this much money. Not stealing from artists, not stealing 531 00:26:00,359 --> 00:26:03,600 Speaker 1: from writers. I don't know if mine have been used, 532 00:26:03,640 --> 00:26:06,600 Speaker 1: but if they have, I got the legal resources to fight, 533 00:26:06,600 --> 00:26:08,960 Speaker 1: and I fucking will. I'm sick of this shit. I'm 534 00:26:09,000 --> 00:26:11,240 Speaker 1: sick of what I see in being done to contractors. 535 00:26:11,320 --> 00:26:13,840 Speaker 1: I'm sick of what I see happening to art directors 536 00:26:13,840 --> 00:26:17,640 Speaker 1: who lose their jobs to some shitthead using day three mini. 537 00:26:18,680 --> 00:26:23,640 Speaker 1: It's grotesque. It's enough to turn your stomach. And as 538 00:26:23,640 --> 00:26:25,960 Speaker 1: we come towards the end of the year, you may 539 00:26:25,960 --> 00:26:27,280 Speaker 1: have at the beginning of the show and been like, 540 00:26:27,320 --> 00:26:29,520 Speaker 1: why Ed is so Why is he so annoyed? I 541 00:26:29,520 --> 00:26:32,800 Speaker 1: think I've done a good job explaining it, but I 542 00:26:32,840 --> 00:26:36,600 Speaker 1: want to end on a good note and look, I 543 00:26:37,000 --> 00:26:39,480 Speaker 1: fear for the future for many reasons, but I always 544 00:26:39,480 --> 00:26:41,480 Speaker 1: have hope because I believe that there are still good 545 00:26:41,480 --> 00:26:44,080 Speaker 1: people in the tech industry, and I love technology. I'm 546 00:26:44,080 --> 00:26:46,239 Speaker 1: going to say every episode that I can, because I 547 00:26:46,280 --> 00:26:48,359 Speaker 1: really do. I love my computer. I love the people 548 00:26:48,359 --> 00:26:49,919 Speaker 1: are fine through it. I love the friends and the 549 00:26:49,920 --> 00:26:52,439 Speaker 1: families loved one I've found through it, and I know 550 00:26:52,480 --> 00:26:53,919 Speaker 1: a lot of you do too. And I want to 551 00:26:54,000 --> 00:26:56,080 Speaker 1: make that clear because none of this is a problem 552 00:26:56,119 --> 00:26:59,440 Speaker 1: with tech as a whole. The people who have taken 553 00:26:59,480 --> 00:27:03,600 Speaker 1: over tech are the problem, the management consultant elite, the 554 00:27:03,680 --> 00:27:06,159 Speaker 1: business people who just want to see growth at all costs. 555 00:27:06,880 --> 00:27:07,760 Speaker 2: But I do have hope. 556 00:27:07,760 --> 00:27:09,520 Speaker 1: I do have hope because of Blue Sky, which is 557 00:27:09,520 --> 00:27:11,560 Speaker 1: a social network that I've posted on for over a 558 00:27:11,640 --> 00:27:13,720 Speaker 1: year that's finally taking route. 559 00:27:13,760 --> 00:27:14,640 Speaker 2: And this isn't an ad. 560 00:27:14,920 --> 00:27:17,920 Speaker 1: This is just somewhere I spend eleven hours a day posting, 561 00:27:18,359 --> 00:27:19,439 Speaker 1: but it feels different. 562 00:27:19,480 --> 00:27:20,360 Speaker 2: It's growing rapidly. 563 00:27:20,359 --> 00:27:23,400 Speaker 1: It's competing with Threads and with Twitter, and they're doing 564 00:27:23,440 --> 00:27:25,920 Speaker 1: someone on an honest product and an open protocol. I'm 565 00:27:25,920 --> 00:27:29,159 Speaker 1: not saying it's the solution for everything. I'm just saying 566 00:27:29,920 --> 00:27:34,800 Speaker 1: solutions are possible. People are building them, people are trying. 567 00:27:35,520 --> 00:27:38,600 Speaker 1: It doesn't have to be like this. The rot economy 568 00:27:38,680 --> 00:27:42,520 Speaker 1: does not have to control the destiny of tech, and 569 00:27:42,600 --> 00:27:44,240 Speaker 1: I think better things can happen. 570 00:27:44,440 --> 00:27:46,800 Speaker 2: It's going to take a while. Things are going to 571 00:27:46,880 --> 00:27:47,320 Speaker 2: blow up. 572 00:27:47,680 --> 00:27:51,920 Speaker 1: I'm scared for this industry for what happens after Generative 573 00:27:51,960 --> 00:27:55,040 Speaker 1: AA collapses, and not just for the initial problem, but 574 00:27:55,080 --> 00:27:58,399 Speaker 1: when they don't have anything else the next quarter. But 575 00:27:58,440 --> 00:28:01,040 Speaker 1: there are other ideas out there. There are other ideas 576 00:28:01,040 --> 00:28:03,520 Speaker 1: of the future that aren't just born out of this shitty, 577 00:28:03,600 --> 00:28:07,439 Speaker 1: scuzzy billionaire mindset from shit heels like Sandar Pashai and 578 00:28:07,480 --> 00:28:10,560 Speaker 1: Sam Ortman and Prabagar Ragavan. And they can and they 579 00:28:10,600 --> 00:28:13,679 Speaker 1: will grow out of the ruins that these people create. 580 00:28:14,600 --> 00:28:18,200 Speaker 1: I want you to have hope. Hope isn't just about 581 00:28:18,680 --> 00:28:22,400 Speaker 1: blindly being optimistic. You can feel hopeful while still being 582 00:28:22,440 --> 00:28:25,680 Speaker 1: pissed off at everything. Hope is about getting up every 583 00:28:25,760 --> 00:28:28,280 Speaker 1: day and being willing to say what you think better 584 00:28:28,320 --> 00:28:32,760 Speaker 1: looks like, what you think a better world could be. 585 00:28:33,240 --> 00:28:35,639 Speaker 1: You don't have to do it at scale. You don't 586 00:28:35,680 --> 00:28:37,880 Speaker 1: have to be an influencer. You don't have to write, 587 00:28:37,920 --> 00:28:39,680 Speaker 1: you don't have to have a tech podcast. You just 588 00:28:39,760 --> 00:28:42,200 Speaker 1: have to talk to your friends. You have to talk 589 00:28:42,240 --> 00:28:44,040 Speaker 1: to the people you know. You have to fight within 590 00:28:44,080 --> 00:28:48,479 Speaker 1: the institutions you are in to find the joy and 591 00:28:48,520 --> 00:28:51,560 Speaker 1: to push out things like generative AI. We can have 592 00:28:51,720 --> 00:28:54,959 Speaker 1: better and you are the beginning of fixing these problems. 593 00:28:55,560 --> 00:28:58,640 Speaker 1: And I actually believe that with enough pushback, these big 594 00:28:58,680 --> 00:29:01,360 Speaker 1: tech companies can change too. I'm not relying on them. 595 00:29:01,520 --> 00:29:04,440 Speaker 1: I hate that I have to use their products, but 596 00:29:04,560 --> 00:29:08,400 Speaker 1: you don't. Perhaps you can find alternatives. I always say 597 00:29:08,400 --> 00:29:10,240 Speaker 1: it's a great day to join blue Sky or use 598 00:29:10,280 --> 00:29:13,920 Speaker 1: Signal proton males apparently great. I'm very much stuck in 599 00:29:13,960 --> 00:29:17,920 Speaker 1: the Gmail ecosystem, which sucks a few decades in there. 600 00:29:17,920 --> 00:29:19,280 Speaker 2: It's tough, is it decades now? 601 00:29:19,400 --> 00:29:22,440 Speaker 1: Either way, It's hard to escape the rot economy whole, 602 00:29:22,960 --> 00:29:25,120 Speaker 1: but there are bits of yourself you can remove from it. 603 00:29:25,960 --> 00:29:30,360 Speaker 2: Please do not give up hope. Please stay angry. 604 00:29:30,480 --> 00:29:33,320 Speaker 1: If you are angry, not telling you to be angry. 605 00:29:33,440 --> 00:29:39,360 Speaker 1: Your indignation is necessary. Your righteous indignation is necessary to 606 00:29:39,400 --> 00:29:43,640 Speaker 1: tell these people to go fucking pounds sand your little changes. 607 00:29:43,720 --> 00:29:46,240 Speaker 1: Those discussions you have with colleagues and friends and loved 608 00:29:46,280 --> 00:29:50,320 Speaker 1: ones about these problems. That knowledge is powerful, and I 609 00:29:50,440 --> 00:29:54,240 Speaker 1: encourage you to talk about these things. The tech industry 610 00:29:54,240 --> 00:29:57,200 Speaker 1: has grown big and strong, telling you you're too stupid 611 00:29:57,200 --> 00:30:00,000 Speaker 1: to understand what they're doing. And unless I'm very much 612 00:30:00,080 --> 00:30:02,760 Speaker 1: mistaken based on the readers and listeners to talk to 613 00:30:03,240 --> 00:30:06,720 Speaker 1: you all get this. I just explained GPUs and server architecture, 614 00:30:06,760 --> 00:30:09,280 Speaker 1: and shit, I think you mostly got it right. I'm 615 00:30:09,320 --> 00:30:12,200 Speaker 1: assuming you did. You keep listening, and that's because this 616 00:30:12,240 --> 00:30:15,200 Speaker 1: stuff isn't that complex. They want you to believe it 617 00:30:15,240 --> 00:30:17,480 Speaker 1: is so that they can stay powerful. But you're a 618 00:30:17,480 --> 00:30:19,880 Speaker 1: lot goddamn smarter than you give yourself credit for. You're 619 00:30:19,880 --> 00:30:23,520 Speaker 1: a lot goddamn stronger than you are too. Don't let 620 00:30:23,560 --> 00:30:26,920 Speaker 1: these people make you feel oppressed. Don't let them make 621 00:30:27,000 --> 00:30:27,800 Speaker 1: you feel weak. 622 00:30:28,160 --> 00:30:28,640 Speaker 2: You are not. 623 00:30:29,280 --> 00:30:32,080 Speaker 1: You are so much more valuable to them than they'll 624 00:30:32,120 --> 00:30:34,040 Speaker 1: ever be to you. 625 00:30:34,040 --> 00:30:34,920 Speaker 2: You have been conned. 626 00:30:35,120 --> 00:30:39,280 Speaker 1: You are the victim, and you will succeed long term. 627 00:30:39,800 --> 00:30:41,800 Speaker 1: Spread the good ideas. Talk about what a good world 628 00:30:41,880 --> 00:30:52,360 Speaker 1: looks like, and thank you for listening. Thank you for 629 00:30:52,400 --> 00:30:55,080 Speaker 1: listening to Better Offline. The editor and composer of the 630 00:30:55,080 --> 00:30:58,160 Speaker 1: Better Offline theme song is Matosowski. You can check out 631 00:30:58,200 --> 00:31:01,200 Speaker 1: more of his music and audio projects at Mattasowski dot 632 00:31:01,240 --> 00:31:03,200 Speaker 1: com m A T T O. 633 00:31:03,400 --> 00:31:06,920 Speaker 2: S O W s ki dot com. 634 00:31:06,960 --> 00:31:09,600 Speaker 1: You can email me at easy at Better offline dot com, 635 00:31:09,680 --> 00:31:12,320 Speaker 1: or visit Better Offline dot com to find more podcast links. 636 00:31:12,320 --> 00:31:13,480 Speaker 2: And of course my newsletter. 637 00:31:13,880 --> 00:31:16,400 Speaker 1: I also really recommend you go to chat dot where's 638 00:31:16,400 --> 00:31:18,880 Speaker 1: youreaed dot at to visit the discord, and go to 639 00:31:18,920 --> 00:31:21,360 Speaker 1: our slash Better Offline to check out our reddit. 640 00:31:22,120 --> 00:31:25,400 Speaker 2: Thank you so much for listening. Better Offline is a 641 00:31:25,400 --> 00:31:26,760 Speaker 2: production of cool Zone Media. 642 00:31:26,920 --> 00:31:29,760 Speaker 1: For more from cool Zone Media, visit our website cool 643 00:31:29,840 --> 00:31:33,200 Speaker 1: Zonemedia dot com, or check us out on the iHeartRadio app, 644 00:31:33,240 --> 00:31:35,920 Speaker 1: Apple Podcasts, or wherever you get your podcasts.