1 00:00:02,800 --> 00:00:06,880 Speaker 1: Zone Media Hell, and welcome to Better Offline. I'm your 2 00:00:06,880 --> 00:00:21,799 Speaker 1: host ed Zeitron. Subscribe to the newsletter by the merchandise. 3 00:00:21,840 --> 00:00:23,560 Speaker 1: It's all in the notes. And we're on the second 4 00:00:23,640 --> 00:00:26,680 Speaker 1: installment of our three part Hater's Guide to the AI 5 00:00:26,800 --> 00:00:30,040 Speaker 1: bubble and the cracks within the generative AI industry and 6 00:00:30,040 --> 00:00:32,600 Speaker 1: how they're becoming bigger and scarier and the potential economic 7 00:00:32,680 --> 00:00:35,760 Speaker 1: meltdown course by a collapse in generative AI spending. Well, 8 00:00:35,960 --> 00:00:38,440 Speaker 1: it's not really general if AI spending. It's literally just 9 00:00:38,479 --> 00:00:41,159 Speaker 1: fucking GPUs, and I think it might be sooner and 10 00:00:41,320 --> 00:00:44,360 Speaker 1: likelier than many think. Toward the end of the last episode, 11 00:00:44,360 --> 00:00:46,360 Speaker 1: we talked about one of the inane comparisons we hear 12 00:00:46,400 --> 00:00:49,760 Speaker 1: between today's nation state size spending on jen ai capital 13 00:00:49,760 --> 00:00:52,840 Speaker 1: expenditures and the investments that Amazon made when scaling Amazon 14 00:00:52,880 --> 00:00:56,440 Speaker 1: Web Services, which was literally the foundation of Dow Computing 15 00:00:56,480 --> 00:00:58,920 Speaker 1: at scale. I would say someone's going to email and 16 00:00:58,920 --> 00:01:01,280 Speaker 1: say I'm wrong, not going to read it, and I 17 00:01:01,320 --> 00:01:03,640 Speaker 1: had to cut things short because we ran out of time. 18 00:01:03,760 --> 00:01:04,200 Speaker 2: But I want to. 19 00:01:04,160 --> 00:01:06,600 Speaker 1: Continue the conversation because I think it's important to examine 20 00:01:06,600 --> 00:01:10,040 Speaker 1: this comparison thoroughly, if not just to explain why it 21 00:01:10,080 --> 00:01:12,080 Speaker 1: doesn't work. It's also I want to stop. I want 22 00:01:12,080 --> 00:01:13,880 Speaker 1: to stop hearing it. I want when people say it 23 00:01:13,880 --> 00:01:15,280 Speaker 1: to me, I just want to send them this fucking 24 00:01:15,319 --> 00:01:17,520 Speaker 1: episode and say leave me alone, buddy boy. 25 00:01:18,720 --> 00:01:20,360 Speaker 2: But but bye. 26 00:01:20,400 --> 00:01:22,160 Speaker 1: But the first point I want to make in this 27 00:01:22,200 --> 00:01:24,800 Speaker 1: episode is that generative AI and large language models do 28 00:01:24,880 --> 00:01:28,840 Speaker 1: not resemble Amazon Web Services or the greater cloud compute boom, 29 00:01:28,920 --> 00:01:32,520 Speaker 1: and generative AI is not infrastructure. Now, some people compare 30 00:01:32,640 --> 00:01:35,720 Speaker 1: llms and their associated services to Amazon Web Services or 31 00:01:35,760 --> 00:01:39,279 Speaker 1: services like Microsoft Zero or Google Cloud, their giant, multi 32 00:01:39,280 --> 00:01:43,280 Speaker 1: billion dollar operations that basically share their server capacity with 33 00:01:43,319 --> 00:01:46,120 Speaker 1: companies wanting to run stuff on the Internet or within 34 00:01:46,160 --> 00:01:49,280 Speaker 1: their own within their own systems. A very fudgy way 35 00:01:49,280 --> 00:01:52,440 Speaker 1: of putting. They help make sure that applications work online. 36 00:01:52,720 --> 00:01:56,600 Speaker 1: These are very very useful services, and by the way, 37 00:01:56,600 --> 00:01:58,280 Speaker 1: people are wrong to make the comparison between them and 38 00:01:58,280 --> 00:02:01,880 Speaker 1: the l lms. As I'll get into now. Amazon Web 39 00:02:01,920 --> 00:02:04,440 Speaker 1: Services when it launched, comprised of things like and forgive 40 00:02:04,600 --> 00:02:07,400 Speaker 1: me how much I'm going to dilute this, Amazon's Elastic 41 00:02:07,440 --> 00:02:10,000 Speaker 1: Compute Cloud EC two, where you rent space in Amazon 42 00:02:10,080 --> 00:02:12,480 Speaker 1: service to run applications in the cloud, or Amazon Simple 43 00:02:12,520 --> 00:02:15,680 Speaker 1: Storage S three, which is enterprise level storage for applications 44 00:02:15,680 --> 00:02:18,679 Speaker 1: and storing things, is not just like a simple hard drive. 45 00:02:18,720 --> 00:02:21,560 Speaker 1: It's redundancy, it's making sure it's copied in places, so 46 00:02:21,639 --> 00:02:25,200 Speaker 1: latency comes down tons of other things. But in simpler terms, 47 00:02:25,800 --> 00:02:27,840 Speaker 1: if you were providing a cloud based service, you used 48 00:02:27,880 --> 00:02:30,280 Speaker 1: Amazon to both stored a stuff that the service needed 49 00:02:30,360 --> 00:02:33,320 Speaker 1: and the cloud actual cloud based processing. So of compute, 50 00:02:33,320 --> 00:02:35,680 Speaker 1: so like your compute loads and runs applications, but delivered 51 00:02:35,680 --> 00:02:37,760 Speaker 1: the thousands of millions of people online. 52 00:02:38,200 --> 00:02:39,600 Speaker 2: And this is a huge industry. 53 00:02:40,280 --> 00:02:42,680 Speaker 1: Amazon Web services are alone brought in WHEB revenues are 54 00:02:42,680 --> 00:02:44,840 Speaker 1: over one hundred billion dollars in twenty twenty four. And 55 00:02:44,840 --> 00:02:47,280 Speaker 1: while Microsoft and Google don't break out their cloud revenues, 56 00:02:47,280 --> 00:02:50,200 Speaker 1: they're similarly large parts of their companies, and Microsoft is 57 00:02:50,240 --> 00:02:52,440 Speaker 1: used as zero and the past to patch over shoddy growth. 58 00:02:53,000 --> 00:02:56,440 Speaker 1: These services are also selling infrastructure. You aren't just paying 59 00:02:56,480 --> 00:02:58,799 Speaker 1: for compute, but the ability to access storage and deliver 60 00:02:58,880 --> 00:03:02,080 Speaker 1: services with low lateent so users have a snappy experiences 61 00:03:02,680 --> 00:03:04,360 Speaker 1: wherever they are in the world. And I know I 62 00:03:04,440 --> 00:03:07,680 Speaker 1: just said a snappy experiences. I'm not editing it. The 63 00:03:07,720 --> 00:03:10,440 Speaker 1: subtle magic of the Internet is that it works at all, 64 00:03:10,480 --> 00:03:12,359 Speaker 1: and a large part of that is the cloud compute 65 00:03:12,360 --> 00:03:15,040 Speaker 1: infrastructure and oligopoly of the main cloud providers. Having such 66 00:03:15,040 --> 00:03:17,919 Speaker 1: a vast data centers, this is much cheaper than doing 67 00:03:17,919 --> 00:03:20,520 Speaker 1: it yourself, and to a certain point, Jobbox moved away 68 00:03:20,520 --> 00:03:22,880 Speaker 1: from Amazon Web Services is it at scale, for example, 69 00:03:23,440 --> 00:03:25,200 Speaker 1: but this also allows someone to take care of the 70 00:03:25,240 --> 00:03:27,960 Speaker 1: maintenance of the hardware and make sure it actually gets 71 00:03:28,040 --> 00:03:31,160 Speaker 1: your stuff to your customers. You also don't have to 72 00:03:31,160 --> 00:03:34,000 Speaker 1: worry about spikes and usage because these things are usage based, 73 00:03:34,000 --> 00:03:36,440 Speaker 1: hence the elastic and you can always add more compute 74 00:03:36,440 --> 00:03:38,560 Speaker 1: to meet demand or just have it in a particular time. 75 00:03:38,960 --> 00:03:42,560 Speaker 1: There is, of course nuance, security specific features, content specific 76 00:03:42,600 --> 00:03:46,680 Speaker 1: delivery services, data based services. There's nuance behind these clouds. 77 00:03:47,080 --> 00:03:49,640 Speaker 1: You're buying into the infrastructure of the infrastructure provider, and 78 00:03:49,680 --> 00:03:52,120 Speaker 1: the reason these products are so profitable is that in 79 00:03:52,200 --> 00:03:54,480 Speaker 1: part you are handing off the problems and responsibility to 80 00:03:54,480 --> 00:03:57,680 Speaker 1: somebody else. And also, most web applications are not that 81 00:03:57,800 --> 00:04:01,520 Speaker 1: demanding of cloud compute. They might be at scale expensive 82 00:04:01,560 --> 00:04:04,680 Speaker 1: to provide to millions of people, but Facebook was not 83 00:04:04,840 --> 00:04:09,200 Speaker 1: a super complex, I don't know website depending on thousands 84 00:04:09,280 --> 00:04:12,400 Speaker 1: or millions of GPUs, and based on the idea, there 85 00:04:12,400 --> 00:04:14,560 Speaker 1: are multiple product categories you can build on top of 86 00:04:14,560 --> 00:04:17,920 Speaker 1: something like edblys Because ultimately cloud services are about Amazon, 87 00:04:18,000 --> 00:04:22,000 Speaker 1: Microsoft and Google running your infrastructure for you. Large language 88 00:04:22,000 --> 00:04:25,160 Speaker 1: models and their associated services are completely different, despite these 89 00:04:25,160 --> 00:04:27,280 Speaker 1: companies attempting to prove otherwise. And it starts with a 90 00:04:27,360 --> 00:04:31,279 Speaker 1: very very simple problem. Why did any of these companies 91 00:04:31,279 --> 00:04:33,400 Speaker 1: build these giant data centers and why did they fill 92 00:04:33,400 --> 00:04:36,840 Speaker 1: them full of GPUs? Amazon Web Services was created out 93 00:04:36,839 --> 00:04:40,320 Speaker 1: of necessity. Amazon's infrastructure needs were so great that it 94 00:04:40,360 --> 00:04:42,799 Speaker 1: effectively had to build out the software and hardware necessary 95 00:04:42,839 --> 00:04:46,240 Speaker 1: to deliver a store that sold theoretically everything, the theoretically anywhere, 96 00:04:46,520 --> 00:04:49,320 Speaker 1: handling both the traffic and customers, delivering the software that 97 00:04:49,400 --> 00:04:52,920 Speaker 1: runs Amazon dot Com quickly and reliably and well, making 98 00:04:52,960 --> 00:04:56,039 Speaker 1: sure things kept working, making sure they were stable. And 99 00:04:56,080 --> 00:04:57,920 Speaker 1: it didn't need to come up with a reason for 100 00:04:58,000 --> 00:05:01,159 Speaker 1: people to run web applications. They were already running applications 101 00:05:01,160 --> 00:05:04,680 Speaker 1: client side on their computers. They realized that doing so 102 00:05:04,760 --> 00:05:07,159 Speaker 1: at scale would be cool, or they were already doing 103 00:05:07,200 --> 00:05:10,479 Speaker 1: so in a way that was likely not particularly cost effective. 104 00:05:11,240 --> 00:05:13,039 Speaker 1: And the ways that we're doing so, they were inflexible, 105 00:05:13,080 --> 00:05:16,720 Speaker 1: and they required specially skills and indeed physical infrastructure personnel. 106 00:05:16,720 --> 00:05:20,120 Speaker 1: They were quite expensive. So Amazon Web Services took something 107 00:05:20,120 --> 00:05:22,400 Speaker 1: that people already did and what there was actually a 108 00:05:22,400 --> 00:05:25,360 Speaker 1: proven demand for, and made it better and scaled it. Eventually, 109 00:05:25,440 --> 00:05:28,160 Speaker 1: Google and Microsoft copied done because that's all they can do. 110 00:05:28,880 --> 00:05:31,520 Speaker 1: And that appears to be the only similarity with generative AI. 111 00:05:31,640 --> 00:05:33,640 Speaker 1: That due to the ridiculous costs of both data centers 112 00:05:33,640 --> 00:05:36,520 Speaker 1: and GPUs necessary to provide these services, it's largely impossible 113 00:05:36,560 --> 00:05:37,760 Speaker 1: for others to enter the market. 114 00:05:38,520 --> 00:05:38,719 Speaker 2: You know. 115 00:05:38,760 --> 00:05:41,120 Speaker 1: After that, generative AI feels more like a feature of 116 00:05:41,160 --> 00:05:45,120 Speaker 1: cloud infrastructure rather than the infrastructure itself. AWS and similar 117 00:05:45,120 --> 00:05:48,279 Speaker 1: medic clouds are versatile, flexible, and multi faceted. Generative AI 118 00:05:48,440 --> 00:05:52,880 Speaker 1: does what generative AI does well, that's about it. You 119 00:05:52,920 --> 00:05:55,480 Speaker 1: can run lots of different things in AWS. What are 120 00:05:55,480 --> 00:05:58,480 Speaker 1: the different things you can run using large language models? 121 00:05:58,480 --> 00:06:01,599 Speaker 1: What are the different use cases and indeed user requirements 122 00:06:01,680 --> 00:06:04,880 Speaker 1: that make this the supposed next big thing. Perhaps the 123 00:06:05,000 --> 00:06:07,520 Speaker 1: argument is that generator of AI is the next AWS 124 00:06:07,600 --> 00:06:09,440 Speaker 1: or similar cloud service because you can build the next 125 00:06:09,440 --> 00:06:11,880 Speaker 1: great companies on the infrastructure of others. The models of 126 00:06:11,960 --> 00:06:16,920 Speaker 1: say open AI and anthropic and the service of Microsoft. Okay, okay, 127 00:06:17,839 --> 00:06:20,200 Speaker 1: let's humor this point too. You can build the next 128 00:06:20,200 --> 00:06:22,560 Speaker 1: great AI startup, and you have to build it on 129 00:06:22,560 --> 00:06:24,960 Speaker 1: one of the megaclouds because they're the only ones that 130 00:06:25,000 --> 00:06:29,839 Speaker 1: can afford to build the infrastructure. One inc wincteny small problem. 131 00:06:30,400 --> 00:06:32,600 Speaker 1: Companies built on top of large language models don't make 132 00:06:32,720 --> 00:06:35,200 Speaker 1: much money, and in fact they're almost all deeply unprofitable. 133 00:06:35,640 --> 00:06:40,320 Speaker 1: But let's establish a few flats to get going. I said, flats, flats, 134 00:06:40,520 --> 00:06:41,240 Speaker 1: Jesus Christ. 135 00:06:41,480 --> 00:06:41,880 Speaker 2: Facts. 136 00:06:42,520 --> 00:06:45,880 Speaker 1: Here are the flats I'm establishing. Outside of one exception, 137 00:06:46,080 --> 00:06:48,600 Speaker 1: mid Journey, which claimed it was profitable in twenty twenty two, 138 00:06:48,640 --> 00:06:50,839 Speaker 1: which may not still be the case. I've actually reached 139 00:06:50,839 --> 00:06:52,159 Speaker 1: out to ask them and they didn't get. 140 00:06:52,000 --> 00:06:52,359 Speaker 2: Back to me. 141 00:06:52,680 --> 00:06:58,560 Speaker 1: Every single LLM model is company is unprofitable, often wildly so. 142 00:06:58,600 --> 00:07:00,960 Speaker 1: Outside of open ai and oropic, in any sphere which 143 00:07:00,960 --> 00:07:03,520 Speaker 1: makes the AI coding app cursor, there are no large 144 00:07:03,560 --> 00:07:06,680 Speaker 1: language model companies either building models or services on top 145 00:07:06,720 --> 00:07:08,599 Speaker 1: of others models that make more than five hundred million 146 00:07:08,640 --> 00:07:12,360 Speaker 1: dollars in annualized revenue meaning month times twelve. Outside mid 147 00:07:12,400 --> 00:07:14,960 Speaker 1: Journeys two hundred million arr and Iron clouds one hundred 148 00:07:14,960 --> 00:07:21,120 Speaker 1: and fifty million arr Also fucking perplexity, there are only 149 00:07:21,200 --> 00:07:24,160 Speaker 1: twelve generative AI powered companies making one hundred million dollars 150 00:07:24,200 --> 00:07:27,320 Speaker 1: annualized or eighteen point three million dollars a month in revenue. 151 00:07:27,560 --> 00:07:31,200 Speaker 1: The database then, this is the Information's AI. Generative AI 152 00:07:31,280 --> 00:07:34,480 Speaker 1: database doesn't have replt, which also announced it hit one 153 00:07:34,520 --> 00:07:36,960 Speaker 1: hundred million in analyzed revenue. I've included it in my 154 00:07:37,040 --> 00:07:41,000 Speaker 1: statement of facts. Of these companies, two of them have 155 00:07:41,040 --> 00:07:43,360 Speaker 1: been acquired, move Works acquired by service Now in March 156 00:07:43,360 --> 00:07:45,800 Speaker 1: twenty twenty five after the company shit the Big Big Time, 157 00:07:46,040 --> 00:07:48,640 Speaker 1: and Windsurf, which was acquired by Google and Cognition in 158 00:07:48,720 --> 00:07:50,600 Speaker 1: July twenty twenty five and one of the most annoying 159 00:07:50,640 --> 00:07:53,360 Speaker 1: deals of all time. But for the sake of simplicity, 160 00:07:53,400 --> 00:07:55,920 Speaker 1: I've left out companies like Surge, Scale, Cheering, and Together, 161 00:07:56,000 --> 00:07:59,120 Speaker 1: all of whom run consultancies selling services and training stuff 162 00:07:59,120 --> 00:08:02,600 Speaker 1: for training models. Otherwise, there are seven companies total that 163 00:08:02,640 --> 00:08:05,200 Speaker 1: make fifty million dollars or more annual recurring revenue, which 164 00:08:05,240 --> 00:08:07,320 Speaker 1: is four point one six million dollars a month. 165 00:08:08,440 --> 00:08:09,680 Speaker 2: Now, none of this is to say. 166 00:08:09,520 --> 00:08:11,440 Speaker 1: That one hundred million dollars isn't a lot of money 167 00:08:11,480 --> 00:08:12,080 Speaker 1: to you and me. 168 00:08:12,520 --> 00:08:13,320 Speaker 2: I just want to be clear. 169 00:08:13,320 --> 00:08:15,000 Speaker 1: If you want to give me one hundred million dollars, 170 00:08:15,080 --> 00:08:18,280 Speaker 1: I'll do anything. I'll wink like a pig for you anyway. 171 00:08:18,360 --> 00:08:20,040 Speaker 1: But in the world of software as a service or 172 00:08:20,120 --> 00:08:23,640 Speaker 1: enterprise software, this is jump change HubSpot At revenues are 173 00:08:23,640 --> 00:08:26,200 Speaker 1: two point six three billion dollars in its twenty twenty 174 00:08:26,240 --> 00:08:29,520 Speaker 1: four financial year. Three years into this crap, and Generative 175 00:08:29,520 --> 00:08:32,800 Speaker 1: AI's highest grossing companies outside of open Ai ten billion 176 00:08:33,040 --> 00:08:36,400 Speaker 1: annualized as of June and Anthropic four billion annualized as July. 177 00:08:36,559 --> 00:08:39,000 Speaker 1: Don't like saying that word. Both of them loose billions 178 00:08:39,000 --> 00:08:42,640 Speaker 1: a year after revenue. There are really three problems here. 179 00:08:42,960 --> 00:08:45,920 Speaker 1: Businesses powered by Generative AI do not seem to be popular, 180 00:08:46,280 --> 00:08:49,680 Speaker 1: Those businesses that are remotely popular are deeply unprofitable, and 181 00:08:49,720 --> 00:08:52,920 Speaker 1: even the less popular generative AI powered businesses are also 182 00:08:53,000 --> 00:08:56,280 Speaker 1: deeply unprofitable. But I want to start somewhere because I 183 00:08:56,440 --> 00:09:01,120 Speaker 1: keep hearing about fucking Cursor. Fucking's start with any sphere 184 00:09:01,120 --> 00:09:05,520 Speaker 1: and Cursor and their app Cursor. It's an AI powered 185 00:09:05,559 --> 00:09:08,040 Speaker 1: coding app and they have five hundred million dollars of 186 00:09:08,120 --> 00:09:09,080 Speaker 1: annualized revenue. 187 00:09:09,200 --> 00:09:10,720 Speaker 2: Pretty great, right, ha. 188 00:09:11,240 --> 00:09:13,640 Speaker 1: It hit two hundred million dollars in annualized revenue in 189 00:09:13,679 --> 00:09:16,560 Speaker 1: March and then hit five hundred million in June after 190 00:09:16,640 --> 00:09:20,800 Speaker 1: raising nine hundred million dollars. That's amazing, ed, ed it's 191 00:09:20,880 --> 00:09:24,000 Speaker 1: time walk to the garage. ED, it's over for you. Wrong, 192 00:09:24,240 --> 00:09:26,760 Speaker 1: it's a mirage. Cursor's growth was the result of an 193 00:09:26,800 --> 00:09:29,000 Speaker 1: unsustainable business model that it's now had to replace with 194 00:09:29,040 --> 00:09:32,600 Speaker 1: opaque terms of service, dramatically restricting access to models, and 195 00:09:32,679 --> 00:09:35,040 Speaker 1: rate limits that effectively stop its users using the product 196 00:09:35,080 --> 00:09:37,040 Speaker 1: at the price point they were used to go to 197 00:09:37,160 --> 00:09:40,640 Speaker 1: Arsnash Cursor on red app Take a look. Take a 198 00:09:40,640 --> 00:09:42,120 Speaker 1: look at how happy everyone is. 199 00:09:42,440 --> 00:09:43,240 Speaker 2: I want to know one. 200 00:09:43,240 --> 00:09:44,960 Speaker 1: My peers in the media don't seem to have the 201 00:09:44,960 --> 00:09:48,200 Speaker 1: ability to talk to actual fucking customers. It's ridiculous. This 202 00:09:48,240 --> 00:09:51,120 Speaker 1: company is circling the drain, and nobody seems to want 203 00:09:51,160 --> 00:09:53,880 Speaker 1: to talk about it, despite how big a deal that is. 204 00:09:54,240 --> 00:09:54,440 Speaker 2: Oh. 205 00:09:54,480 --> 00:09:57,160 Speaker 1: Also, Curse is horribly unprofitable, and I believe there are 206 00:09:57,160 --> 00:09:59,520 Speaker 1: a sign of things to come in generative AI. A 207 00:09:59,520 --> 00:10:01,319 Speaker 1: couple of weeks weeks ago, I wrote up the dramatic 208 00:10:01,400 --> 00:10:03,600 Speaker 1: changes that Cursor made to its service in the middle 209 00:10:03,600 --> 00:10:06,199 Speaker 1: of June or my premium newsletter and discovered that they 210 00:10:06,480 --> 00:10:09,520 Speaker 1: timed these changes precisely with Anthropic and open Ai to 211 00:10:09,559 --> 00:10:13,240 Speaker 1: a lesser extent, adding service tiers and priority processing, which 212 00:10:13,280 --> 00:10:15,360 Speaker 1: is tech language for pay us extra if you have 213 00:10:15,400 --> 00:10:17,920 Speaker 1: a lot of customers or face rate limits or service delays. 214 00:10:17,960 --> 00:10:18,480 Speaker 2: Asshole. 215 00:10:19,160 --> 00:10:21,480 Speaker 1: These price ships have also led to companies like replt 216 00:10:21,600 --> 00:10:24,000 Speaker 1: having to make significant changes to their pricing model that 217 00:10:24,040 --> 00:10:29,360 Speaker 1: disfavors users. People are finding in really simple terms that 218 00:10:29,520 --> 00:10:32,079 Speaker 1: what they used to get for twenty bucks is much much, much, 219 00:10:32,200 --> 00:10:35,640 Speaker 1: much much smaller curse the users hit rate limits. Replit 220 00:10:35,720 --> 00:10:38,199 Speaker 1: users are hitting rate limits, and even then when they 221 00:10:38,200 --> 00:10:39,840 Speaker 1: try and do the same things, they're spending way more 222 00:10:39,880 --> 00:10:41,560 Speaker 1: money if they go pay as you go. It's a 223 00:10:41,559 --> 00:10:44,400 Speaker 1: complete fast But I'm going to repeat some of this 224 00:10:44,440 --> 00:10:46,520 Speaker 1: stuff from the premium newsletter because there is a time 225 00:10:46,520 --> 00:10:49,000 Speaker 1: of events that I believe are going to be in 226 00:10:49,040 --> 00:10:52,000 Speaker 1: the big short to AI Boogloo all right. In or 227 00:10:52,080 --> 00:10:54,440 Speaker 1: around May fifth, twenty twenty five, Cursor closed the nine 228 00:10:54,480 --> 00:10:57,320 Speaker 1: hundred million dollar funding round in a Around May twenty second, 229 00:10:57,360 --> 00:11:00,640 Speaker 1: twenty twenty five, Anthropic launched Clawed four Opus and new 230 00:11:00,679 --> 00:11:03,480 Speaker 1: models with Sona and Opus, both of them kind of 231 00:11:03,480 --> 00:11:05,800 Speaker 1: well known for coding, and on May thirtieth, twenty twenty five, 232 00:11:05,840 --> 00:11:09,440 Speaker 1: they added service tiers, including priority pricing specifically focused on 233 00:11:09,520 --> 00:11:12,160 Speaker 1: cash heavy products like Cursor and the cash is when 234 00:11:12,160 --> 00:11:14,240 Speaker 1: you put stuff that you're going to be looking at regularly, 235 00:11:14,440 --> 00:11:15,880 Speaker 1: take a look at it, and you can use it 236 00:11:15,880 --> 00:11:19,200 Speaker 1: more readily. Cash is the CAC eight G, by the way, 237 00:11:19,840 --> 00:11:21,800 Speaker 1: is generally something that's for efficiency. 238 00:11:22,360 --> 00:11:23,160 Speaker 2: The idea that you. 239 00:11:23,160 --> 00:11:26,920 Speaker 1: Would add a toll onto the cash is fucking disgusting 240 00:11:26,920 --> 00:11:30,560 Speaker 1: and only targeted coding startups. But on May thirtieth, twenty 241 00:11:30,600 --> 00:11:33,960 Speaker 1: twenty five, Reutter's reported the Anthropics annualized revenue hit three 242 00:11:34,000 --> 00:11:37,760 Speaker 1: billion dollars, with a key driver being code generation. This 243 00:11:37,800 --> 00:11:40,200 Speaker 1: translates to around two hundred and fifty million dollars in 244 00:11:40,280 --> 00:11:43,880 Speaker 1: monthly revenue. June ninth, twenty twenty five, CNBC reported open 245 00:11:43,880 --> 00:11:47,680 Speaker 1: Ai'd hit ten billion dollars in annualized revenue. And yeah, 246 00:11:47,720 --> 00:11:50,360 Speaker 1: when they said ann your recurring revenue, they meant annualized. 247 00:11:50,559 --> 00:11:52,319 Speaker 1: But the very same day they cut the price of 248 00:11:52,360 --> 00:11:54,960 Speaker 1: their three model by eighty percent, which competes directly with 249 00:11:55,000 --> 00:11:56,760 Speaker 1: Clawed four Opus by the way, and This was a 250 00:11:56,800 --> 00:11:59,760 Speaker 1: direct and aggressive attempt to force Anthropic to kind of 251 00:11:59,800 --> 00:12:03,120 Speaker 1: like make too ether lower prices or compete. It's just 252 00:12:03,400 --> 00:12:07,160 Speaker 1: shtheads fuckinging around with assholes. But on or around June sixteen, 253 00:12:07,200 --> 00:12:09,600 Speaker 1: twenty twenty five, Cursor changed its pricing, added a new 254 00:12:09,640 --> 00:12:11,880 Speaker 1: two hundred dollar a month Ultra tier that, in their 255 00:12:11,920 --> 00:12:14,800 Speaker 1: own words, was made possible by multi year partnerships with 256 00:12:14,880 --> 00:12:18,680 Speaker 1: open Ai, Anthropic, Google, an Xai, which translates to multi 257 00:12:18,720 --> 00:12:21,960 Speaker 1: year commitments to spend which can be amortized as monthly amounts. 258 00:12:22,400 --> 00:12:25,440 Speaker 1: A day later, on June seventeenth, Cursor dramatically changed its 259 00:12:25,440 --> 00:12:28,679 Speaker 1: offering to it for its twenty dollars a month subscriptions 260 00:12:28,720 --> 00:12:31,520 Speaker 1: to usage base, where one got at least the value 261 00:12:31,520 --> 00:12:33,400 Speaker 1: of their subscription, so a twenty bus a month person 262 00:12:33,440 --> 00:12:37,520 Speaker 1: would get more than twenty dollars of API course in compute, 263 00:12:37,720 --> 00:12:41,319 Speaker 1: along with arbitrary rate limits and unlimited access to Cursor's 264 00:12:41,320 --> 00:12:44,280 Speaker 1: own slow model that its users hey. Then on June eighteenth, 265 00:12:44,360 --> 00:12:46,679 Speaker 1: Repler and other vibe coding company that I had previously 266 00:12:46,720 --> 00:12:50,079 Speaker 1: mentioned announced their effort based pricing increases that were massive. 267 00:12:50,559 --> 00:12:53,280 Speaker 2: July first, the Information reported. 268 00:12:52,920 --> 00:12:56,320 Speaker 1: The Anthropic hit four billion dollars of annualized revenue making 269 00:12:56,360 --> 00:12:59,080 Speaker 1: three hundred and thirty million dollars a month, an increase 270 00:12:59,120 --> 00:13:01,520 Speaker 1: of eighty three million dollars a month. We'll just under 271 00:13:01,520 --> 00:13:03,240 Speaker 1: twenty five percent in the space of a month. 272 00:13:04,920 --> 00:13:05,160 Speaker 2: Hmm. 273 00:13:05,960 --> 00:13:18,800 Speaker 1: Where could that money have come from? In simpler terms, 274 00:13:18,840 --> 00:13:21,080 Speaker 1: Cursor raised nine hundred million dollars and very likely had 275 00:13:21,120 --> 00:13:23,000 Speaker 1: to hand large amounts of that money over to Open 276 00:13:23,040 --> 00:13:25,480 Speaker 1: Air and Anthropic to keep doing business with them, then 277 00:13:25,480 --> 00:13:28,000 Speaker 1: immediately change the terms of service to make them worse 278 00:13:28,000 --> 00:13:30,320 Speaker 1: for their customers. And as I said at the time, 279 00:13:30,360 --> 00:13:32,320 Speaker 1: and this is a direct quote from my news there, 280 00:13:33,440 --> 00:13:35,640 Speaker 1: while some met, no, I can't do the Kevin Ruth's 281 00:13:35,679 --> 00:13:38,679 Speaker 1: voice and doing my own stuff, pardon me. While some 282 00:13:38,760 --> 00:13:41,520 Speaker 1: may believe that Open AI and Anthropic hitting annualized revenue 283 00:13:41,559 --> 00:13:43,920 Speaker 1: milestones is good news, you have to consider how these 284 00:13:43,960 --> 00:13:46,559 Speaker 1: milestones were hit. Based on my reporting, I believe that 285 00:13:46,600 --> 00:13:50,200 Speaker 1: both companies are effectively doing steroids, forcing massive infrastructural costs 286 00:13:50,200 --> 00:13:52,600 Speaker 1: onto big customers as a means of covering the increasing 287 00:13:52,640 --> 00:13:55,160 Speaker 1: costs of their own models. There is simply no other 288 00:13:55,200 --> 00:13:58,160 Speaker 1: aid to read this situation. By making these changes, Anthropic 289 00:13:58,200 --> 00:14:00,800 Speaker 1: is intentionally making it harder for its larger costs largest 290 00:14:00,800 --> 00:14:03,320 Speaker 1: customer to do business. By the way, Cursor is their 291 00:14:03,400 --> 00:14:06,920 Speaker 1: largest customer, creating the extra revenue by making Cursors product 292 00:14:07,000 --> 00:14:10,240 Speaker 1: worse by proxy. What's sickening about this particular situation. It 293 00:14:10,240 --> 00:14:12,720 Speaker 1: doesn't really matter if Curs's customers are happy or sad. 294 00:14:13,080 --> 00:14:17,920 Speaker 1: They like open AI's Enterprise Priority Access API Anthropic in 295 00:14:17,920 --> 00:14:20,280 Speaker 1: this case, require a long term commitment which involves a 296 00:14:20,280 --> 00:14:22,680 Speaker 1: minimum through put of tokens per second as part of 297 00:14:22,680 --> 00:14:26,000 Speaker 1: their tiered access program. If Curs's customers drop off, both 298 00:14:26,040 --> 00:14:28,360 Speaker 1: Anthropic and open Ai still get their cut, and if 299 00:14:28,400 --> 00:14:31,120 Speaker 1: curses customers somehow out spend those commitments, they'll either still 300 00:14:31,120 --> 00:14:34,280 Speaker 1: get rate limited or any sphere willkin cur more costs. 301 00:14:35,480 --> 00:14:36,600 Speaker 2: Why do you care about this? 302 00:14:37,040 --> 00:14:39,600 Speaker 1: Well, Cursor is the largest and most successful genetive AI 303 00:14:39,760 --> 00:14:42,080 Speaker 1: company by far. In these aggressive and desperate changes to 304 00:14:42,120 --> 00:14:45,080 Speaker 1: its products suggest that a that its products are deeply unprofitable, 305 00:14:45,080 --> 00:14:46,880 Speaker 1: and b that its current growth was the result of 306 00:14:46,880 --> 00:14:48,560 Speaker 1: offering a product that it was not the one it 307 00:14:48,560 --> 00:14:51,920 Speaker 1: would sell in the long term. Cursor misled its customers, 308 00:14:52,120 --> 00:14:55,920 Speaker 1: and its current revenue is as a result, highly unlikely 309 00:14:55,960 --> 00:14:59,240 Speaker 1: to stay at this level. Worse still, two anthropic engineers 310 00:14:59,360 --> 00:15:01,560 Speaker 1: left from the the Clawed Code team to go and 311 00:15:01,600 --> 00:15:04,160 Speaker 1: work at Cursor two weeks ago, and they have already 312 00:15:04,200 --> 00:15:06,600 Speaker 1: come back. This heavily suggests that whatever they saw over 313 00:15:06,600 --> 00:15:09,240 Speaker 1: there wasn't compelling enough to make them stay. As I 314 00:15:09,280 --> 00:15:12,480 Speaker 1: also said, while Cursor may have raised nine hundred million dollars, 315 00:15:12,640 --> 00:15:15,760 Speaker 1: it was really open Aianthropic XAI and Google that got 316 00:15:15,800 --> 00:15:18,640 Speaker 1: that money. At this point, there are no profitable price 317 00:15:18,680 --> 00:15:21,040 Speaker 1: AI startups, and it's highly unlikely that the new pricing 318 00:15:21,120 --> 00:15:23,120 Speaker 1: models by both Cursor and Replet are going to help. 319 00:15:23,720 --> 00:15:24,600 Speaker 2: These are now the. 320 00:15:24,600 --> 00:15:27,160 Speaker 1: New terms of doing business with the big model companies, 321 00:15:27,320 --> 00:15:29,840 Speaker 1: a shakedown where you pay for priority access or tears, 322 00:15:29,920 --> 00:15:33,640 Speaker 1: or face indeterminate delays or rate limits. Any start up 323 00:15:33,680 --> 00:15:36,640 Speaker 1: scaling into an enterprise integration of General AVII, which means 324 00:15:36,680 --> 00:15:39,280 Speaker 1: in this case anything that requires a level of service 325 00:15:39,360 --> 00:15:41,880 Speaker 1: uptime has to commit to both a minimum amount of 326 00:15:41,920 --> 00:15:43,720 Speaker 1: months and the throughput of tokens, which means that the 327 00:15:43,720 --> 00:15:46,280 Speaker 1: price of starting an AI company that gets any kind 328 00:15:46,320 --> 00:15:50,120 Speaker 1: of real market traction just dramatically increased. Well, one could say, oh, 329 00:15:50,160 --> 00:15:52,280 Speaker 1: perhaps you don't need priority access. The need here is 330 00:15:52,320 --> 00:15:54,920 Speaker 1: something that can be entirely judged by anthropic and open 331 00:15:54,960 --> 00:15:58,000 Speaker 1: ai in a totally opaque manner. They can and they 332 00:15:58,040 --> 00:16:00,520 Speaker 1: will throttle companies that are two demanding on their systems. 333 00:16:00,600 --> 00:16:02,120 Speaker 1: It's proven by the fact that they've done this to 334 00:16:02,120 --> 00:16:06,120 Speaker 1: curse them multiple times. But okay, why does curse them 335 00:16:06,120 --> 00:16:09,160 Speaker 1: out so much? And it's simple. Generative AI will not 336 00:16:09,240 --> 00:16:13,600 Speaker 1: get big on selling consumer software without an enterprise SaaS story, 337 00:16:13,840 --> 00:16:19,080 Speaker 1: they're dead And I realize, I know, okay, folks, it's 338 00:16:19,160 --> 00:16:21,120 Speaker 1: kind of a little boring hearing about software as a 339 00:16:21,160 --> 00:16:23,840 Speaker 1: service despite the fact that it's a huge, several hundred 340 00:16:23,880 --> 00:16:26,800 Speaker 1: billion dollar industry. But this is the only place where 341 00:16:26,840 --> 00:16:29,760 Speaker 1: generative AI can really make money. Companies buying hundreds of 342 00:16:29,800 --> 00:16:33,120 Speaker 1: thousands of seats or how industries that rely on compute 343 00:16:33,160 --> 00:16:35,920 Speaker 1: grow and without that growth, they're going nowhere. To give 344 00:16:35,920 --> 00:16:38,840 Speaker 1: you some context, Netflix makes about thirty nine billion dollars 345 00:16:38,880 --> 00:16:42,280 Speaker 1: a year in subscription new from consumers, and Spotify about 346 00:16:42,280 --> 00:16:42,920 Speaker 1: eighteen billion. 347 00:16:43,360 --> 00:16:43,840 Speaker 2: These are the. 348 00:16:43,800 --> 00:16:46,800 Speaker 1: Single most popular consumer software subscriptions in the world, and 349 00:16:46,840 --> 00:16:49,800 Speaker 1: open ai is fifteen point five million subscribers. Suggest that 350 00:16:50,000 --> 00:16:51,760 Speaker 1: open ai can't rely on them for the kind of 351 00:16:51,760 --> 00:16:54,360 Speaker 1: growth that would actually make the company worth three hundred billion. 352 00:16:54,160 --> 00:16:55,160 Speaker 2: Dollars or more. 353 00:16:55,960 --> 00:16:58,400 Speaker 1: Cuzer, as it stands, is the one example of a 354 00:16:58,440 --> 00:17:02,640 Speaker 1: company thriving using GENERATIVEA a software company selling software, and 355 00:17:02,680 --> 00:17:04,920 Speaker 1: it appears its rapid growth was the result of selling 356 00:17:04,960 --> 00:17:07,560 Speaker 1: a product at a massive loss. As it stands today, 357 00:17:07,640 --> 00:17:10,280 Speaker 1: Curs's product is significantly worse and it's ready it's full 358 00:17:10,320 --> 00:17:12,639 Speaker 1: of people furious at the company for the changes. In 359 00:17:12,720 --> 00:17:15,439 Speaker 1: simpler terms, Curser was the company that people mentioned to 360 00:17:15,440 --> 00:17:17,720 Speaker 1: prove that startups could make money by building on top 361 00:17:17,760 --> 00:17:20,320 Speaker 1: products on top of open AI and Anthropics models. Yet 362 00:17:20,359 --> 00:17:22,159 Speaker 1: the truth is the only way to do so is 363 00:17:22,200 --> 00:17:24,359 Speaker 1: to grow, and grow is to burn tons of money. 364 00:17:25,119 --> 00:17:27,719 Speaker 1: While the tempting argument is to say that Curs's customers 365 00:17:27,760 --> 00:17:30,280 Speaker 1: are addicted and will keep paying, this is clearly not 366 00:17:30,320 --> 00:17:33,600 Speaker 1: the case, nor is it an actual business model. Like 367 00:17:33,680 --> 00:17:36,199 Speaker 1: people that say this, I've never had a drug addiction, 368 00:17:36,240 --> 00:17:39,320 Speaker 1: but I know people that do it. It's nothing like software. 369 00:17:39,400 --> 00:17:42,520 Speaker 1: Stop making that comparison. It's insulting to the victims of addiction. 370 00:17:43,119 --> 00:17:46,040 Speaker 1: But anyway, this story showed that open A and Anthropics 371 00:17:46,080 --> 00:17:48,320 Speaker 1: are actually their bigger the biggest threats to their customers 372 00:17:48,359 --> 00:17:50,360 Speaker 1: and will actively rent seek can punish any of their 373 00:17:50,359 --> 00:17:53,000 Speaker 1: success stories, looking to loose as much as they can from. 374 00:17:52,840 --> 00:17:54,200 Speaker 2: Them before they copy their products. 375 00:17:54,680 --> 00:17:57,880 Speaker 1: To put it bluntly, curses growth story was a fucking lie. 376 00:17:58,000 --> 00:18:00,680 Speaker 1: It reached five hundred million dollars in annulif revenue selling 377 00:18:00,680 --> 00:18:02,520 Speaker 1: a product it can no longer afford to sell and 378 00:18:02,560 --> 00:18:06,119 Speaker 1: could not afford to sell long term, suggesting material weakness 379 00:18:06,160 --> 00:18:09,400 Speaker 1: in its business and any and all coding startups. It's 380 00:18:09,440 --> 00:18:11,919 Speaker 1: also remarkable, in the shocking failure of journalism that this 381 00:18:12,040 --> 00:18:16,000 Speaker 1: isn't in every single article about any sphere. I'm doing 382 00:18:16,040 --> 00:18:18,520 Speaker 1: this part time? Why am I in the asshole here? 383 00:18:18,560 --> 00:18:22,760 Speaker 1: Like I'm I don't know, really, though, I do have 384 00:18:22,800 --> 00:18:25,640 Speaker 1: a question. Where are all the consumer AI starts? I'm 385 00:18:25,800 --> 00:18:26,720 Speaker 1: genuinely serious. 386 00:18:27,080 --> 00:18:28,000 Speaker 2: What have you got for me? 387 00:18:28,119 --> 00:18:32,280 Speaker 1: Perplexity Perplexity. Perplexity only has one hundred and fifty million 388 00:18:32,320 --> 00:18:34,639 Speaker 1: dollars in the annualized revenue, and they spent one hundred 389 00:18:34,640 --> 00:18:37,360 Speaker 1: and sixty seven percent of their revenue in twenty twenty four, 390 00:18:37,560 --> 00:18:40,119 Speaker 1: or fifty seven million dollars of spending on revenues of 391 00:18:40,200 --> 00:18:43,480 Speaker 1: thirty four million dollars on computer services from Anthropic, Open 392 00:18:43,520 --> 00:18:47,879 Speaker 1: AI and Amazon. They lost sixty eight million dollars and 393 00:18:47,920 --> 00:18:50,520 Speaker 1: worse still, they still have no path to profitability and 394 00:18:50,560 --> 00:18:53,439 Speaker 1: it's not even making anything new. They're a search engine, 395 00:18:53,480 --> 00:18:57,160 Speaker 1: they have an AI browser. But don't worry. Professional gas 396 00:18:57,160 --> 00:19:00,399 Speaker 1: bag Alex Heath just did this insane and flumm mixing 397 00:19:00,400 --> 00:19:04,560 Speaker 1: interview with CEO Aravins Ravinas, who, when asked how it 398 00:19:04,600 --> 00:19:08,200 Speaker 1: perplexed you would become profitable, appeared to experience what seems 399 00:19:08,200 --> 00:19:12,240 Speaker 1: to be a stroke like I'm about to read something 400 00:19:12,240 --> 00:19:15,480 Speaker 1: to you and it's gonna sound strange, but this is 401 00:19:15,520 --> 00:19:18,600 Speaker 1: exactly what was said. Maybe let me give you another example. 402 00:19:18,720 --> 00:19:20,359 Speaker 1: You want to put an ad on meta Instagram, and 403 00:19:20,400 --> 00:19:22,200 Speaker 1: you want to look at ads done by similar brands, 404 00:19:22,400 --> 00:19:24,560 Speaker 1: pull that, study that, or look at AdWords pricing of 405 00:19:24,600 --> 00:19:26,720 Speaker 1: one hundred different keywords and figure out how to price 406 00:19:26,760 --> 00:19:29,320 Speaker 1: your thing comparatively. These are tasks that could definitely save 407 00:19:29,359 --> 00:19:31,119 Speaker 1: you hours and hours and maybe even give you up 408 00:19:31,160 --> 00:19:33,600 Speaker 1: an arbitrage over what you could do yourself, because AI 409 00:19:33,720 --> 00:19:35,679 Speaker 1: is able to do a lot more and at scale. 410 00:19:35,720 --> 00:19:37,639 Speaker 1: If it helps you to make a few million bugs, 411 00:19:37,680 --> 00:19:39,680 Speaker 1: does it not make sense to spend two thousand dollars 412 00:19:39,680 --> 00:19:41,840 Speaker 1: for that prompt? It does, right, So I think we're 413 00:19:41,880 --> 00:19:44,320 Speaker 1: going to be able to monetize in many more interesting 414 00:19:44,359 --> 00:19:46,960 Speaker 1: ways than chatbots for the browser. I want to be 415 00:19:47,040 --> 00:19:50,280 Speaker 1: fucking clear about something. Alex seems like a nice guy. 416 00:19:50,440 --> 00:19:52,359 Speaker 1: If someone said that to me, I'd ask them if 417 00:19:52,359 --> 00:19:56,280 Speaker 1: they could smell toast. I'd be like, Aravin, Mate, are 418 00:19:56,320 --> 00:19:58,960 Speaker 1: you okay? How many fingers I'm holding up? 419 00:19:58,960 --> 00:19:59,719 Speaker 2: Aravin? You're right? 420 00:19:59,760 --> 00:20:01,800 Speaker 1: Did you hit your head on something? The ceilings don't 421 00:20:01,800 --> 00:20:04,240 Speaker 1: seem that low in here. But mate, you're just spewing 422 00:20:04,359 --> 00:20:07,960 Speaker 1: utter fucking nonsense. I've read this paragraph multiple times. I 423 00:20:08,000 --> 00:20:09,800 Speaker 1: do not know what he's getting at. I think he's 424 00:20:09,800 --> 00:20:14,320 Speaker 1: suggesting something about how you could ask it to tell 425 00:20:14,359 --> 00:20:15,520 Speaker 1: you what to do with ads. 426 00:20:15,600 --> 00:20:18,439 Speaker 2: I don't know. I don't know. 427 00:20:18,800 --> 00:20:22,040 Speaker 1: This is the big probably the biggest consumer AI company 428 00:20:22,040 --> 00:20:24,639 Speaker 1: that isn't open AI, and they speak like they're an 429 00:20:24,720 --> 00:20:28,080 Speaker 1: insane person or a stupid person. Check out the Business 430 00:20:28,119 --> 00:20:31,439 Speaker 1: Idiot Trilogy for what I think there. I also mentioned 431 00:20:31,440 --> 00:20:33,119 Speaker 1: them earlier, but I don't I don't want you to 432 00:20:33,119 --> 00:20:36,240 Speaker 1: talk to me about AI browsers. Anyone humoring AI browsers 433 00:20:36,280 --> 00:20:40,359 Speaker 1: is being an imbecile for some reason. They are not 434 00:20:40,440 --> 00:20:42,920 Speaker 1: a business model. How are people going to make money 435 00:20:42,960 --> 00:20:45,439 Speaker 1: on the browser. Hm hmm, what do these products actually do? 436 00:20:45,760 --> 00:20:51,560 Speaker 2: Oh? They can poorly automate accepting linked invites. Wow. Wow, 437 00:20:51,600 --> 00:20:53,879 Speaker 2: it's like God himself has personally best my computer. A 438 00:20:53,880 --> 00:20:54,760 Speaker 2: big fucking deal. 439 00:20:55,160 --> 00:20:56,760 Speaker 1: In any case, it doesn't seem like you can really 440 00:20:56,800 --> 00:20:59,720 Speaker 1: build a consumer AI startup that makes any real money 441 00:20:59,800 --> 00:21:02,800 Speaker 1: or approach being a real company other than chat GPT, 442 00:21:03,240 --> 00:21:06,520 Speaker 1: I guess, and that's because the GENERATIVEAI software market is small, 443 00:21:06,560 --> 00:21:10,040 Speaker 1: with little room for growth and no profits to be seen. Arguably, 444 00:21:10,080 --> 00:21:12,280 Speaker 1: the biggest sign that things are are troubling in the 445 00:21:12,280 --> 00:21:15,199 Speaker 1: generative AI spaces that we use the term annualized revenue 446 00:21:15,200 --> 00:21:18,199 Speaker 1: at all, which, as I've mentioned repeatedly, means multiplying a 447 00:21:18,240 --> 00:21:21,199 Speaker 1: month by twelve and saying that's our annualized baby. The 448 00:21:21,240 --> 00:21:24,520 Speaker 1: problem with this number is that, well, people cancel things. 449 00:21:24,840 --> 00:21:27,119 Speaker 1: While your June might look great, if ten percent of 450 00:21:27,119 --> 00:21:29,240 Speaker 1: your subscribers churning a bad month due to a change 451 00:21:29,280 --> 00:21:31,120 Speaker 1: in your terms of service, for example, that's a huge 452 00:21:31,160 --> 00:21:33,800 Speaker 1: chunk of your annualized revenue gone and likely gone forever. 453 00:21:34,240 --> 00:21:36,080 Speaker 1: But the worst sign is that nobody is saying the 454 00:21:36,119 --> 00:21:39,280 Speaker 1: monthly figures, mostly because the monthly figures fucking suck. One 455 00:21:39,320 --> 00:21:41,600 Speaker 1: hundred million dollars of anualized revenue is eight point three 456 00:21:41,680 --> 00:21:44,280 Speaker 1: three million dollars a month. To give you some scale, 457 00:21:44,320 --> 00:21:46,600 Speaker 1: Amazon Web Services hit one hundred and eighty nine million 458 00:21:46,640 --> 00:21:48,720 Speaker 1: dollars fifteen point seventy five million dollars a month in 459 00:21:48,760 --> 00:21:51,320 Speaker 1: revenue in two thousand and eight, two years after founding, 460 00:21:51,440 --> 00:21:53,639 Speaker 1: and while it took until twenty fifteen to hit profitability, 461 00:21:53,640 --> 00:21:55,680 Speaker 1: it actually hit break even in two thousand and nine, 462 00:21:55,840 --> 00:21:57,800 Speaker 1: though were invested in cash and growth for a few 463 00:21:57,840 --> 00:22:00,640 Speaker 1: years later. And I should be clear them doing that 464 00:22:01,000 --> 00:22:04,280 Speaker 1: justified so many startups burning cash, so many starts like yeah, 465 00:22:04,280 --> 00:22:07,720 Speaker 1: look at aws. They were investing in growth, which is 466 00:22:07,760 --> 00:22:09,880 Speaker 1: a fair thing for companies to do. But I'm being 467 00:22:09,920 --> 00:22:13,320 Speaker 1: an asshole. But right now there is not a single 468 00:22:13,359 --> 00:22:16,119 Speaker 1: generative AI software company that's profitable, and none of them 469 00:22:16,119 --> 00:22:17,919 Speaker 1: are showing the signs of the kind of hypergrowth that 470 00:22:17,960 --> 00:22:21,639 Speaker 1: previous big software companies had or Cursor technically is the 471 00:22:21,720 --> 00:22:24,480 Speaker 1: fastest growing software as a service company of all time. 472 00:22:24,920 --> 00:22:29,080 Speaker 1: It got there by basically lying. Cursor is never bringing 473 00:22:29,160 --> 00:22:34,119 Speaker 1: back the product at the twenty dollars price point that 474 00:22:34,200 --> 00:22:36,480 Speaker 1: they were selling. They're never doing it. The money they 475 00:22:36,600 --> 00:22:40,240 Speaker 1: earned was earned it's not fraud because they didn't do it. 476 00:22:40,920 --> 00:22:42,800 Speaker 2: I guess it was deceptive, but it's not really to 477 00:22:42,840 --> 00:22:44,359 Speaker 2: the it's just fucking lying. 478 00:22:44,640 --> 00:22:47,480 Speaker 1: It's just lying. And who knows what happens to curser now. 479 00:22:47,800 --> 00:22:50,000 Speaker 1: But you know what, I'm harping on cursor a bit. 480 00:22:50,040 --> 00:22:51,800 Speaker 1: What other software startups are there? 481 00:22:51,920 --> 00:22:57,240 Speaker 2: Glean, Glean, fucking Glean, Glean, everyone loves to talk about. 482 00:22:57,359 --> 00:23:00,040 Speaker 1: Enterprise search company Glean, a company that uses AI to 483 00:23:00,040 --> 00:23:02,680 Speaker 1: search and generate answers from your company's files and documents. 484 00:23:02,880 --> 00:23:06,040 Speaker 1: Fun fact, also Salesforce's own Slack has now blocked them 485 00:23:06,080 --> 00:23:10,600 Speaker 1: from searching Slack. Just arshole on arsehole violence. In December 486 00:23:10,640 --> 00:23:13,119 Speaker 1: twenty twenty four, Glean raised two hundred and sixty million dollars, 487 00:23:13,119 --> 00:23:15,480 Speaker 1: broadly stating that it had over five hundred and fifty 488 00:23:15,520 --> 00:23:18,560 Speaker 1: million dollars in cash with best in class ARR growth. 489 00:23:18,840 --> 00:23:20,920 Speaker 1: A few months later, in February twenty twenty five, Glean 490 00:23:20,920 --> 00:23:23,959 Speaker 1: announced it achieved one hundred million dollars in annual recurring 491 00:23:23,960 --> 00:23:27,520 Speaker 1: revenue in fourth arter FY twenty five, cementing its position 492 00:23:27,600 --> 00:23:29,640 Speaker 1: is one of the fastest growing sas startups and reflecting 493 00:23:29,640 --> 00:23:33,320 Speaker 1: a searching demand for AI powered workplace intelligence. In any case, 494 00:23:33,560 --> 00:23:35,960 Speaker 1: AR could literally mean anything, as it appears to be 495 00:23:36,000 --> 00:23:38,119 Speaker 1: based on quarters, meaning it could be an average of 496 00:23:38,160 --> 00:23:41,000 Speaker 1: the last three months. I guess anyway. In June twenty 497 00:23:41,040 --> 00:23:43,800 Speaker 1: twenty five, Glean announced it had raised another funding round, 498 00:23:43,800 --> 00:23:45,840 Speaker 1: this time raising one hundred and fifty million dollars in 499 00:23:45,880 --> 00:23:49,560 Speaker 1: It troublingly added that since its last round, it had 500 00:23:49,760 --> 00:23:54,080 Speaker 1: surpassed one hundred million dollars in AR raw five months 501 00:23:54,080 --> 00:23:56,680 Speaker 1: into the fucking year. And your revenue is basically the same. 502 00:23:57,119 --> 00:24:00,440 Speaker 1: That isn't good. That isn't good at all. Also happened 503 00:24:00,440 --> 00:24:02,440 Speaker 1: to that five hundred and fifty million dollars in cash? 504 00:24:02,440 --> 00:24:05,120 Speaker 1: Why did Glean need more? Hey, wait a second, take 505 00:24:05,160 --> 00:24:07,560 Speaker 1: a look at this. Glean announced their raise on June eighteenth, 506 00:24:07,560 --> 00:24:10,000 Speaker 1: twenty twenty five, two days after Curses price increase, in 507 00:24:10,040 --> 00:24:12,200 Speaker 1: the same day that Repler announced the similar price act. 508 00:24:12,400 --> 00:24:13,359 Speaker 2: It's almost as if the. 509 00:24:13,400 --> 00:24:17,520 Speaker 1: Dramatic pricing increase has affected them due to the introduction 510 00:24:17,560 --> 00:24:20,359 Speaker 1: of Anthropic Service TRES and Opening Eyes priority processing. 511 00:24:20,359 --> 00:24:22,800 Speaker 2: But I'm guessing. I know, I'm guessing. 512 00:24:22,960 --> 00:24:24,479 Speaker 1: But it is kind of where that all of these 513 00:24:24,520 --> 00:24:26,800 Speaker 1: companies raise money and all announced these things around the 514 00:24:26,840 --> 00:24:27,320 Speaker 1: same time. 515 00:24:42,840 --> 00:24:45,520 Speaker 2: Hey, that reminds me, I got another problem. 516 00:24:45,680 --> 00:24:47,760 Speaker 1: I got another problem here because I think that there 517 00:24:47,840 --> 00:24:51,760 Speaker 1: is another reason why the cycles kind of keep repeating. 518 00:24:51,760 --> 00:24:53,119 Speaker 1: You get a company of that grows, and then they 519 00:24:53,160 --> 00:24:56,520 Speaker 1: kind of go nowhere, because well, the company doesn't really 520 00:24:56,560 --> 00:24:58,560 Speaker 1: seem to have a total addressable market much bigger than 521 00:24:58,560 --> 00:25:02,400 Speaker 1: one hundred million AR and I think it's a little simple. 522 00:25:03,119 --> 00:25:03,840 Speaker 2: It's quite simple. 523 00:25:03,840 --> 00:25:07,600 Speaker 1: In fact, there really are no unique generative AI companies, 524 00:25:07,640 --> 00:25:10,080 Speaker 1: and building a moat on top of l elms is 525 00:25:10,119 --> 00:25:10,959 Speaker 1: near impossible. 526 00:25:11,560 --> 00:25:13,080 Speaker 2: If you look a man, am I going to get 527 00:25:13,119 --> 00:25:15,399 Speaker 2: some emails about this, but bring them on. 528 00:25:15,920 --> 00:25:18,479 Speaker 1: If you look at what GENERATIVEAI companies do, now that 529 00:25:18,520 --> 00:25:21,520 Speaker 1: the following is not a quality barometer, it's probably one 530 00:25:21,560 --> 00:25:24,840 Speaker 1: of the following things. They're either chatbot one, either you 531 00:25:24,880 --> 00:25:29,200 Speaker 1: ask questions or talk to This includes customer service bots, searching, summarizing, 532 00:25:29,440 --> 00:25:32,440 Speaker 1: or comparing documents with increased amounts of complexity of documents 533 00:25:32,480 --> 00:25:34,760 Speaker 1: or quantity of documents to be compared. This includes being 534 00:25:34,800 --> 00:25:39,119 Speaker 1: able to ask questions of documents. Web search deep research, 535 00:25:39,240 --> 00:25:41,600 Speaker 1: meaning long form web search that generates a document where 536 00:25:41,600 --> 00:25:44,240 Speaker 1: some parts of it will inevitably be hallucinated or derived 537 00:25:44,240 --> 00:25:48,520 Speaker 1: from low quality sources, generating text, images, voice, or in 538 00:25:48,600 --> 00:25:52,640 Speaker 1: some rare cases video, Using AI to generative AII mean 539 00:25:52,720 --> 00:25:56,439 Speaker 1: to write, edit or maintain code, transcription, translation, or photo 540 00:25:56,440 --> 00:25:59,280 Speaker 1: and video editing. Every single generative AI company that is 541 00:25:59,320 --> 00:26:02,200 Speaker 1: an open Aireanthropic and honestly kind of those two does 542 00:26:02,359 --> 00:26:04,080 Speaker 1: one or a few of these things, and I mean 543 00:26:04,359 --> 00:26:06,639 Speaker 1: every one of them. And it's because every single generative 544 00:26:06,680 --> 00:26:09,919 Speaker 1: AI company uses large language models, which have inherent limits 545 00:26:09,920 --> 00:26:12,840 Speaker 1: on what they can do. Llms can generate, they can search, 546 00:26:12,880 --> 00:26:16,480 Speaker 1: they can kind of edit, they can sometimes transcribe accurately, 547 00:26:16,520 --> 00:26:21,120 Speaker 1: and they can sometimes translate much more well, much less accurately. 548 00:26:21,160 --> 00:26:24,480 Speaker 1: I guess within weeks of Curses changed to its services, 549 00:26:24,680 --> 00:26:27,520 Speaker 1: Amazon and byte Dance release competitors that, for the most part, do. 550 00:26:27,480 --> 00:26:28,400 Speaker 2: Exactly the same thing. 551 00:26:28,960 --> 00:26:31,159 Speaker 1: Sure, there's a few differences in how they're designed, but 552 00:26:31,200 --> 00:26:33,120 Speaker 1: design is not a moat, especially in a high cost, 553 00:26:33,200 --> 00:26:36,280 Speaker 1: negative profit business were your only way of growing is 554 00:26:36,280 --> 00:26:39,360 Speaker 1: to offer a product you can't sustain. The only other 555 00:26:39,400 --> 00:26:41,400 Speaker 1: moat you can build is the services you provide, which, 556 00:26:41,440 --> 00:26:43,600 Speaker 1: when your services are dependent on a large language model, 557 00:26:43,600 --> 00:26:45,400 Speaker 1: are dependent on the model developer, who, in the case 558 00:26:45,400 --> 00:26:47,880 Speaker 1: of open AI and Anthropic, could simply clone your startup, 559 00:26:48,040 --> 00:26:51,080 Speaker 1: because the only valuable intellectual property is the models, and 560 00:26:51,160 --> 00:26:54,200 Speaker 1: those models are theirs. You may say, well, nobody else 561 00:26:54,240 --> 00:26:56,920 Speaker 1: has any ideas either, to which I say, I fully agree. 562 00:26:57,160 --> 00:26:59,640 Speaker 1: My rock com bubble thesis suggests that we're all out 563 00:26:59,640 --> 00:27:02,240 Speaker 1: of hyper growth ideas, and yeah, I think we're out 564 00:27:02,240 --> 00:27:05,399 Speaker 1: of ideas related to any large language models too. At 565 00:27:05,440 --> 00:27:07,280 Speaker 1: this point, I think it's fair to ask, are there 566 00:27:08,040 --> 00:27:10,960 Speaker 1: any good businesses you can build on top of generative 567 00:27:11,000 --> 00:27:14,520 Speaker 1: AI or large language models. I don't mean ad features 568 00:27:14,520 --> 00:27:17,440 Speaker 1: related to I mean an AI company that actually sells 569 00:27:17,440 --> 00:27:20,120 Speaker 1: a product that people buy at scale that isn't called chat, 570 00:27:20,160 --> 00:27:23,560 Speaker 1: GPT or claude. In previous tech booms, companies would make 571 00:27:23,560 --> 00:27:25,960 Speaker 1: their own models, their own infrastructure, or the things that 572 00:27:26,200 --> 00:27:28,359 Speaker 1: make them distinct from other companies. But the generative AI 573 00:27:28,400 --> 00:27:31,520 Speaker 1: boom effectively changes that by making everybody build on stuff 574 00:27:31,520 --> 00:27:34,040 Speaker 1: on top of somebody else's models, because training your own 575 00:27:34,080 --> 00:27:37,359 Speaker 1: models is both extremely expensive and requires vast amounts of 576 00:27:37,359 --> 00:27:41,240 Speaker 1: infrastructure and just pure power. As a result, much of 577 00:27:41,280 --> 00:27:43,560 Speaker 1: this boom is about a few companies, really too, if 578 00:27:43,560 --> 00:27:46,680 Speaker 1: we're honest, getting other companies to try and build functional 579 00:27:46,680 --> 00:27:49,920 Speaker 1: software for them, and these companies Open ai and Anthropic 580 00:27:50,040 --> 00:27:52,359 Speaker 1: are their customers weak point in a relationship that veers 581 00:27:52,359 --> 00:27:55,480 Speaker 1: from symbiotic to parasitic at a moment's notice. I cannot 582 00:27:55,480 --> 00:27:57,960 Speaker 1: stress enough how bad open ai and Anthropic are for 583 00:27:58,000 --> 00:28:00,879 Speaker 1: their business customers. Their models are popular, by which I 584 00:28:00,920 --> 00:28:03,960 Speaker 1: mean their customers customers will expect access to them, meaning 585 00:28:04,000 --> 00:28:06,479 Speaker 1: the open ai and Anthropic can, as they did to Cursor, 586 00:28:06,600 --> 00:28:10,159 Speaker 1: arbitrarily change pricing, service availability, and functionality based on how 587 00:28:10,200 --> 00:28:12,119 Speaker 1: they feel that day or whether they need to pump 588 00:28:12,160 --> 00:28:14,040 Speaker 1: their annualized revenue for investors. 589 00:28:14,560 --> 00:28:15,199 Speaker 2: Don't believe me. 590 00:28:16,000 --> 00:28:19,280 Speaker 1: Anthropic cut off access to AI coding platform Windsurf because 591 00:28:19,320 --> 00:28:21,560 Speaker 1: it looked like they might get acquired by open Ai. 592 00:28:21,880 --> 00:28:24,720 Speaker 1: They never were. They just harmed that business. They just 593 00:28:25,119 --> 00:28:28,440 Speaker 1: cut a hole in them. Why because they might touch 594 00:28:28,480 --> 00:28:31,480 Speaker 1: another business, the most anti competitive shit in the world. 595 00:28:31,520 --> 00:28:34,880 Speaker 1: And everyone sat there clapping like a fucking seal. Disgusting 596 00:28:35,560 --> 00:28:38,440 Speaker 1: even by big tech standards. This fucking sucks, and these 597 00:28:38,440 --> 00:28:40,760 Speaker 1: companies will do it again. But you know what, Let's 598 00:28:40,800 --> 00:28:43,120 Speaker 1: talk about the actual uses of generative AI, because the 599 00:28:43,160 --> 00:28:45,840 Speaker 1: limited number of use cases are because large language models 600 00:28:45,840 --> 00:28:49,240 Speaker 1: are all really really similar. Because all large language models 601 00:28:49,240 --> 00:28:51,920 Speaker 1: require more data than anyone who's ever needed, including like 602 00:28:52,000 --> 00:28:54,000 Speaker 1: four times the amount of data on the Internet. They 603 00:28:54,040 --> 00:28:56,240 Speaker 1: all basically have to use the same thing, either taken 604 00:28:56,280 --> 00:28:58,640 Speaker 1: from the Internet or bought from one of the few 605 00:28:58,680 --> 00:29:02,640 Speaker 1: companies that scale surge during together or whoever. While they 606 00:29:02,720 --> 00:29:06,240 Speaker 1: can get customized data or do customized training and reinforcement learning, 607 00:29:06,400 --> 00:29:09,440 Speaker 1: these models are all transformer based and they all function similarly, 608 00:29:09,440 --> 00:29:11,120 Speaker 1: and the only way to make them different is by 609 00:29:11,160 --> 00:29:13,960 Speaker 1: training them, which doesn't make them that much different, just 610 00:29:14,000 --> 00:29:17,440 Speaker 1: better things they already do. And good lord, is it 611 00:29:17,680 --> 00:29:20,760 Speaker 1: so is general IFAI is so ungodly expensive and the 612 00:29:20,800 --> 00:29:22,880 Speaker 1: training is as well. By the way, they have to 613 00:29:22,920 --> 00:29:25,160 Speaker 1: pay real humans as well, which they hate doing. And 614 00:29:25,200 --> 00:29:27,920 Speaker 1: even when they're paying outsourced labor and ken youre at 615 00:29:27,920 --> 00:29:30,560 Speaker 1: two dollars a pop, they're still losing a ton of money. 616 00:29:30,960 --> 00:29:34,120 Speaker 1: It's really crazy, actually, how badly built all of this is. 617 00:29:34,520 --> 00:29:37,000 Speaker 1: And I already mentioned open AI and Anthropics costs as 618 00:29:37,000 --> 00:29:39,640 Speaker 1: well as perplex The's fifty million dollar bill in a 619 00:29:39,720 --> 00:29:42,080 Speaker 1: year to Anthropic Amazon and open Ai off of a 620 00:29:42,160 --> 00:29:45,640 Speaker 1: easily thirty four dollars million dollars in revenue. These companies 621 00:29:45,680 --> 00:29:48,440 Speaker 1: cost too much to run and their functionality doesn't make 622 00:29:48,520 --> 00:29:51,360 Speaker 1: enough money to make them make sense. And the problem 623 00:29:51,400 --> 00:29:53,760 Speaker 1: isn't just the pricing, but how unpredictable it is. As 624 00:29:53,760 --> 00:29:57,200 Speaker 1: Matterscheer wrote for cio Dive last year, generative AI makes 625 00:29:57,200 --> 00:29:59,360 Speaker 1: a lot of companies lives difficult for the massive spikes 626 00:29:59,360 --> 00:30:02,080 Speaker 1: and costs that from the power users, with few ways 627 00:30:02,120 --> 00:30:04,680 Speaker 1: to mitigate those costs. One of the ways that company 628 00:30:04,680 --> 00:30:07,520 Speaker 1: manages their cloud bills is by having some degree of predictability, 629 00:30:07,520 --> 00:30:09,400 Speaker 1: which is difficult to do with the constant sleu of 630 00:30:09,400 --> 00:30:11,440 Speaker 1: new models and demands some new products to go with them, 631 00:30:11,600 --> 00:30:14,960 Speaker 1: especially when send models can and can and do often 632 00:30:15,040 --> 00:30:19,840 Speaker 1: cost more with subsequent iterations, not necessarily for much return, 633 00:30:20,040 --> 00:30:22,760 Speaker 1: except if you're a company like a coding company, your 634 00:30:22,800 --> 00:30:25,680 Speaker 1: customers are going to actually ask you for the new models. 635 00:30:26,160 --> 00:30:28,520 Speaker 1: As a result, it's half AI companies to actually budge in. 636 00:30:29,080 --> 00:30:31,120 Speaker 1: But ed, What was that? Ed? 637 00:30:31,600 --> 00:30:32,680 Speaker 2: What about agents? 638 00:30:32,880 --> 00:30:34,920 Speaker 1: Aren't they the thing that will eventually make the insane 639 00:30:34,920 --> 00:30:37,200 Speaker 1: broken calculus behind generative AI actually work? 640 00:30:37,480 --> 00:30:41,280 Speaker 2: What is your accent made? Anyway? Anyway? 641 00:30:42,200 --> 00:30:44,920 Speaker 1: Let me tell you about agents. The term agent is 642 00:30:44,960 --> 00:30:47,160 Speaker 1: one of the most egregious acts of fraud I've seen 643 00:30:47,160 --> 00:30:49,640 Speaker 1: in my entire career writing about this crap, and that 644 00:30:49,640 --> 00:30:52,800 Speaker 1: includes the metavers. When you hear the word agent, you 645 00:30:52,840 --> 00:30:54,680 Speaker 1: were meant to think of an autonomous AI that can 646 00:30:54,680 --> 00:30:57,480 Speaker 1: go and do stuff without oversight, replacing someone's job in 647 00:30:57,520 --> 00:30:59,880 Speaker 1: the process. And companies have been pushing the boundaries of 648 00:30:59,880 --> 00:31:03,040 Speaker 1: good taste and financial crimes in pursuit of them. Most 649 00:31:03,040 --> 00:31:05,640 Speaker 1: egregious of them as Salesforce's Agent Force, which leads you 650 00:31:05,680 --> 00:31:08,560 Speaker 1: deploy AI agents at scale. That's a quote, and brings 651 00:31:08,600 --> 00:31:11,760 Speaker 1: digital labor to every employee, department and business process. Another 652 00:31:11,840 --> 00:31:16,520 Speaker 1: quote from Salesforce's website. These are two blatant fucking lies. 653 00:31:16,880 --> 00:31:21,000 Speaker 1: Agent Force is a goddamn chatbot program. It's a platform 654 00:31:21,040 --> 00:31:24,000 Speaker 1: for launching chatbots. They can sometimes plug into APIs that 655 00:31:24,040 --> 00:31:27,000 Speaker 1: allow them to access other information, but they're neither autonomous 656 00:31:27,000 --> 00:31:30,720 Speaker 1: nor agents by any reasonable definition. Not only does Salesforce 657 00:31:30,800 --> 00:31:33,440 Speaker 1: not actually sell agents, its own research shows that the 658 00:31:33,480 --> 00:31:37,080 Speaker 1: agents and agents in general only achieve around fifty eight 659 00:31:37,080 --> 00:31:41,320 Speaker 1: percent success rate on single step tasks. And I'm going 660 00:31:41,360 --> 00:31:43,680 Speaker 1: to quote the register here. This means tasks that can 661 00:31:43,720 --> 00:31:45,760 Speaker 1: be completed in the single step without needing follow up 662 00:31:45,760 --> 00:31:49,280 Speaker 1: actions and more information or multi step tasks, So you know, 663 00:31:49,440 --> 00:31:52,640 Speaker 1: most tasks they succeed a depressing thirty five percent of 664 00:31:52,760 --> 00:31:56,160 Speaker 1: the time. Last week, open Ai announced its own chat 665 00:31:56,200 --> 00:31:58,760 Speaker 1: GPT agent that can allegedly go and do tasks on 666 00:31:58,800 --> 00:32:01,720 Speaker 1: a virtual computer. In its own demo, the agent took 667 00:32:01,760 --> 00:32:03,560 Speaker 1: twenty one minutes or so to spit out a plan 668 00:32:03,600 --> 00:32:06,640 Speaker 1: for a wedding with destinations of a cander and some 669 00:32:06,640 --> 00:32:08,680 Speaker 1: suit options, and then showed a pre prepared demo of 670 00:32:08,720 --> 00:32:10,720 Speaker 1: the agent preparing an itinery of how to visit every 671 00:32:10,720 --> 00:32:13,400 Speaker 1: major league ballpark. And that's baseball for the non Americans 672 00:32:13,440 --> 00:32:16,640 Speaker 1: out there. In this example's case, agents took twenty three 673 00:32:16,640 --> 00:32:18,960 Speaker 1: minutes and produced arguably the most confusing map I've seen 674 00:32:19,000 --> 00:32:20,640 Speaker 1: in my life. You can see the map in the 675 00:32:20,640 --> 00:32:25,200 Speaker 1: newsletter version of this episode. It's hilarious. It missed out 676 00:32:25,240 --> 00:32:28,200 Speaker 1: every single major ballpark on the East Coast, including Yankee 677 00:32:28,240 --> 00:32:31,360 Speaker 1: Stadium and Femway Park, which are two of the most 678 00:32:31,440 --> 00:32:34,080 Speaker 1: well known stadiums in sports, and added a bunch of 679 00:32:34,200 --> 00:32:36,240 Speaker 1: roundom ones and like one in the middle of the 680 00:32:36,280 --> 00:32:39,560 Speaker 1: Gulf of Mexico. What team is that, Sammy the deep 681 00:32:39,600 --> 00:32:42,600 Speaker 1: Water Horizon Devils. Is there a baseball team in North Dakota? 682 00:32:42,680 --> 00:32:43,120 Speaker 2: Clammy? 683 00:32:43,160 --> 00:32:46,760 Speaker 1: Sammy Samy. I also should be clear this was a 684 00:32:46,800 --> 00:32:50,720 Speaker 1: pre prepared example. This is the best they had. I 685 00:32:50,800 --> 00:32:52,920 Speaker 1: want to see the cutting room footage on this, because 686 00:32:52,960 --> 00:32:55,800 Speaker 1: you best bet that that map looked like straight dogshit, 687 00:32:57,080 --> 00:33:00,280 Speaker 1: as with every large language model product, And yes, that's 688 00:33:00,640 --> 00:33:02,680 Speaker 1: what this is, even if open ai won't talk about 689 00:33:02,680 --> 00:33:06,960 Speaker 1: what model results are. Extremely variable agents are difficult because 690 00:33:06,960 --> 00:33:09,080 Speaker 1: tasks are if for coal, even if they can be 691 00:33:09,080 --> 00:33:12,040 Speaker 1: completed by a human being, that the CEO thinks is stupid. 692 00:33:12,400 --> 00:33:14,160 Speaker 1: What open ai appears to be doing is using a 693 00:33:14,200 --> 00:33:17,720 Speaker 1: virtual machine to run scripts that its models trigger regardless 694 00:33:17,720 --> 00:33:19,760 Speaker 1: of how will it works, and it works very very 695 00:33:19,880 --> 00:33:23,480 Speaker 1: very very poorly and inconsistently. It's also very likely expensive 696 00:33:23,520 --> 00:33:26,480 Speaker 1: to run. In any case, every single company you see 697 00:33:26,560 --> 00:33:29,080 Speaker 1: using the word agent is trying to mislead you. They're 698 00:33:29,120 --> 00:33:32,880 Speaker 1: lying gleans ai agents to chatbots with If this, then 699 00:33:32,960 --> 00:33:36,640 Speaker 1: that functions that trigger events using APIs, which means if 700 00:33:36,680 --> 00:33:40,280 Speaker 1: an event happens, another thing will be triggered, not taking 701 00:33:40,320 --> 00:33:42,680 Speaker 1: actual actions, because that is not what lms can do. 702 00:33:43,200 --> 00:33:46,680 Speaker 1: Service Now's ai agents that allegedly act autonomously and proactively 703 00:33:46,720 --> 00:33:49,600 Speaker 1: on your behalf are despite claiming they go beyond better 704 00:33:49,680 --> 00:33:52,800 Speaker 1: chatbots still ultimately better chatbots that use APIs to trigger 705 00:33:52,840 --> 00:33:55,959 Speaker 1: different events using if this, then that functions. Sometimes these 706 00:33:56,040 --> 00:33:58,400 Speaker 1: chatbox can also answer questions that people might have or 707 00:33:58,400 --> 00:34:02,440 Speaker 1: trigger an event somewhere. Oh right, that's literally the same thing. 708 00:34:03,320 --> 00:34:05,840 Speaker 1: The closest we have to an agent is any kind 709 00:34:05,880 --> 00:34:08,000 Speaker 1: of coding agent, which is they can make a list 710 00:34:08,040 --> 00:34:10,040 Speaker 1: of things that you might do on our software project 711 00:34:10,120 --> 00:34:12,120 Speaker 1: and go and generate code and push stuff to get 712 00:34:12,200 --> 00:34:13,520 Speaker 1: help when you ask them to. And they can do 713 00:34:13,600 --> 00:34:16,399 Speaker 1: so autonomously in the sense that you can just let 714 00:34:16,440 --> 00:34:19,560 Speaker 1: them do what a model that doesn't know anything and 715 00:34:19,640 --> 00:34:23,120 Speaker 1: has no consciousness thinks is right based on its corpus 716 00:34:23,160 --> 00:34:25,680 Speaker 1: of data and the things you can access to. And 717 00:34:26,040 --> 00:34:28,359 Speaker 1: it's about as safe as that sounds. When I say 718 00:34:28,440 --> 00:34:30,680 Speaker 1: ask them to and go, and I mean that these 719 00:34:30,719 --> 00:34:33,520 Speaker 1: agents are not intelligent at all. They do not have intelligence, 720 00:34:33,560 --> 00:34:36,279 Speaker 1: and when let run rampant, fuck up everything and create 721 00:34:36,280 --> 00:34:38,279 Speaker 1: a bunch of extra work or so. A study found 722 00:34:38,280 --> 00:34:42,800 Speaker 1: that AI coding tools made engineers nineteen percent slower. Nevertheless, 723 00:34:42,800 --> 00:34:45,680 Speaker 1: none of these products are autonomous agents, and anybody using 724 00:34:45,680 --> 00:34:48,680 Speaker 1: the term agent likely means chat bond. And all of 725 00:34:48,719 --> 00:34:51,520 Speaker 1: this is working because the media keeps repeating everything these 726 00:34:51,560 --> 00:34:55,040 Speaker 1: companies say. It's a disgrace. We need to stop this. 727 00:34:56,040 --> 00:34:59,200 Speaker 1: I realize we've taken a kind of a scenic route here, though, 728 00:34:59,600 --> 00:35:01,520 Speaker 1: but I know he needed to lay the groundwork, because 729 00:35:01,680 --> 00:35:04,560 Speaker 1: I really am alarmed. According to a UBS report from 730 00:35:04,560 --> 00:35:06,640 Speaker 1: the twenty sixth of June, the public companies running AI 731 00:35:06,719 --> 00:35:10,760 Speaker 1: services are making absolutely pathetic amounts of money from AI. Microsoft, 732 00:35:10,800 --> 00:35:13,720 Speaker 1: according to the UBS, is making annual revenues of somehow 733 00:35:13,800 --> 00:35:16,280 Speaker 1: less than the Information report at two point one billion dollars. 734 00:35:16,360 --> 00:35:18,399 Speaker 1: Service now is making less than two hundred and fifty million, 735 00:35:18,440 --> 00:35:21,000 Speaker 1: Adobe less than one hundred and twenty five million salesforce 736 00:35:21,040 --> 00:35:24,440 Speaker 1: less than one hundred million now service now said two 737 00:35:24,520 --> 00:35:27,440 Speaker 1: hundred and fifty million dollar ACV annual contract value. This 738 00:35:27,560 --> 00:35:30,040 Speaker 1: may be one of the more honest explanations of revenue 739 00:35:30,160 --> 00:35:32,399 Speaker 1: I've seen, putting them in the upper echelons of AI 740 00:35:32,440 --> 00:35:34,879 Speaker 1: revenue and less. Of course, you think about it for 741 00:35:35,000 --> 00:35:37,719 Speaker 1: a couple of seconds and think, are these all AI 742 00:35:37,800 --> 00:35:40,600 Speaker 1: specific contracts or perhaps they're in contracts where you've taped 743 00:35:40,640 --> 00:35:41,640 Speaker 1: AI on to the side. 744 00:35:41,880 --> 00:35:44,800 Speaker 2: It gives a shit. It's also year. 745 00:35:44,600 --> 00:35:47,120 Speaker 1: Long agreements that could churn, and according to Gartner, over 746 00:35:47,160 --> 00:35:49,640 Speaker 1: forty percent of Agenetic AI products will be canceled by 747 00:35:49,760 --> 00:35:52,560 Speaker 1: end of twenty twenty seven. And really, you gotta laugh 748 00:35:52,600 --> 00:35:55,239 Speaker 1: at Adobe and Salesforce, both of whom to talk such 749 00:35:55,280 --> 00:35:58,080 Speaker 1: a goddamn fuck ton about jenerit of Ai. And yeah, 750 00:35:58,080 --> 00:36:00,880 Speaker 1: I have only made amazed the hundred twenty five million 751 00:36:00,880 --> 00:36:02,080 Speaker 1: in analyzed revenue from it. 752 00:36:02,120 --> 00:36:05,000 Speaker 2: Pathetic crap, dog shit. 753 00:36:05,080 --> 00:36:08,000 Speaker 1: These aren't futuristic numbers, they're barely product categories, and none 754 00:36:08,040 --> 00:36:10,160 Speaker 1: of this seems to include costs. 755 00:36:11,160 --> 00:36:13,160 Speaker 2: Oh well, good grief. 756 00:36:14,000 --> 00:36:16,280 Speaker 1: Look, a lot of what I've been saying is reminiscent 757 00:36:16,320 --> 00:36:19,160 Speaker 1: the previous podcasts, and I've gone over this a lot, 758 00:36:19,280 --> 00:36:20,719 Speaker 1: so I really want to make it clear that the 759 00:36:20,719 --> 00:36:23,200 Speaker 1: signs are very troubling, and that the things I've warned 760 00:36:23,200 --> 00:36:25,880 Speaker 1: you about the past couple of years are only getting worse, 761 00:36:26,120 --> 00:36:29,640 Speaker 1: and the cliff's coming up things are only getting closer. 762 00:36:29,800 --> 00:36:31,959 Speaker 1: When we double off of it, things may get really, 763 00:36:31,960 --> 00:36:34,200 Speaker 1: really bad, And then the next episode we'll talk about 764 00:36:34,200 --> 00:36:36,879 Speaker 1: how and what that tumble might look like and the 765 00:36:36,920 --> 00:36:38,640 Speaker 1: noises I'm going to make when it happens. 766 00:36:47,320 --> 00:36:49,719 Speaker 3: Thank you for listening to Better Offline, The editor and 767 00:36:49,760 --> 00:36:52,920 Speaker 3: composer of the Better Offline theme song is Matasowski. You 768 00:36:52,960 --> 00:36:55,239 Speaker 3: can check out more of his music and audio projects 769 00:36:55,360 --> 00:37:00,720 Speaker 3: at Matasowski dot com, m a Ttoso w s Ki 770 00:37:01,120 --> 00:37:04,000 Speaker 3: dot com. You can email me at easy at Better 771 00:37:04,000 --> 00:37:06,440 Speaker 3: Offline dot com or visit Better Offline dot com to 772 00:37:06,480 --> 00:37:09,319 Speaker 3: find more podcast links and of course, my newsletter. I 773 00:37:09,360 --> 00:37:12,080 Speaker 3: also really recommend you go to chat dot where's youreaed 774 00:37:12,160 --> 00:37:15,120 Speaker 3: dot at to visit the discord, and go to our slash. 775 00:37:14,840 --> 00:37:17,960 Speaker 2: Better Offline to check out I'll Reddit. Thank you so 776 00:37:18,080 --> 00:37:21,480 Speaker 2: much for listening. Better Offline is a production of cool 777 00:37:21,560 --> 00:37:22,120 Speaker 2: Zone Media. 778 00:37:22,280 --> 00:37:25,120 Speaker 1: For more from cool Zone Media, visit our website cool 779 00:37:25,160 --> 00:37:28,520 Speaker 1: Zonemedia dot com, or check us out on the iHeartRadio app, 780 00:37:28,600 --> 00:37:31,280 Speaker 1: Apple Podcasts, or wherever you get your podcasts.