WEBVTT - The AI Bubble Is Bursting 0:00:02.480 --> 0:00:06.800 Call Zone Media. Hi, I'm Edzeitron and welcome back to 0:00:06.880 --> 0:00:20.160 Better Offline. As I've discussed in my last episode, there 0:00:20.200 --> 0:00:22.840 are four intractable problems that are going to stop generative 0:00:22.880 --> 0:00:26.920 AI from going much further. It's massive energy demands, its 0:00:26.960 --> 0:00:31.000 massive computation demands. It's hallucinations when it authoritatively tells you 0:00:31.080 --> 0:00:34.040 something that isn't true or makes horrible mistakes in images, 0:00:34.840 --> 0:00:37.040 and the fact that these large language models have this 0:00:37.080 --> 0:00:41.080 insatiable need for more training data. Yet I think what 0:00:41.200 --> 0:00:44.720 might pop this bubble is a far simpler problem. Generative 0:00:44.720 --> 0:00:48.600 AI simply does not deliver the magical automation that everybody 0:00:48.600 --> 0:00:52.080 has been fantasizing about, and I don't think consumers or 0:00:52.200 --> 0:00:56.680 enterprises are actually impressed. A year and a half after launch, 0:00:57.200 --> 0:00:59.600 it seems like the kind of immediate and unquestioning in 0:00:59.640 --> 0:01:02.360 fact youation with chat GBT and for that matter, or 0:01:02.400 --> 0:01:08.760 other generative AIS has softened. Instead, there's this rising undercurrent 0:01:08.840 --> 0:01:12.760 of apathy and mistrust and of course failure that's kind 0:01:12.760 --> 0:01:16.480 of hard to ignore. In June twenty twenty three, traffic 0:01:16.480 --> 0:01:19.440 to chat GPT's website, where people access the chat GPT 0:01:19.600 --> 0:01:22.200 bot in a web browser fell for the first time 0:01:22.240 --> 0:01:25.160 since launch, starting a trend that's continued for five of 0:01:25.160 --> 0:01:27.880 the following eight months. According to data from similar Web, 0:01:28.720 --> 0:01:31.760 people are becoming more aware of the technology's limitations, like 0:01:31.800 --> 0:01:35.839 as I mentioned, hallucinations, which, as I note, is when 0:01:35.959 --> 0:01:39.560 chat gpt confidently asserts things that aren't true, which can 0:01:39.600 --> 0:01:42.039 be in writing, when it gives you an incorrect fact, 0:01:42.600 --> 0:01:44.759 or in an image when it gives a dog eighteen 0:01:44.840 --> 0:01:49.280 legs to make matters worse. According to data from data Ai, 0:01:49.360 --> 0:01:52.520 which used to be known as Appanny, chat gbt's downloads 0:01:52.520 --> 0:01:54.760 in iOS have begun to drop from a high of 0:01:54.880 --> 0:01:57.440 just over seven hundred thousand a week to a plateau 0:01:57.440 --> 0:02:00.280 of around four hundred and fifty thousand to five hundred 0:02:00.320 --> 0:02:04.639 thousand a week since early twenty twenty three, which sounds 0:02:04.680 --> 0:02:07.480 impressive until you hear that only seven point three five 0:02:07.560 --> 0:02:11.240 percent of people who downloaded chatgbt in January twenty twenty 0:02:11.320 --> 0:02:14.200 four actually used the app again thirty days after they 0:02:14.280 --> 0:02:18.000 downloaded it, cratering from a high of twenty eight percent 0:02:18.040 --> 0:02:20.400 a month after the app launched in June twenty twenty three. 0:02:21.480 --> 0:02:25.679 In fact, things immediately appear to have fallen apart in 0:02:25.800 --> 0:02:28.280 July twenty twenty three, only two months after launch, only 0:02:28.320 --> 0:02:31.200 four point five nine percent of users opened the app 0:02:31.200 --> 0:02:34.680 for a second time. Numbers like these tell the story 0:02:34.680 --> 0:02:38.320 of a buzzy new application that isn't actually providing users 0:02:38.360 --> 0:02:41.840 with much utility. I think the generative AI engine has 0:02:41.840 --> 0:02:45.919 started to sputter for customers, for businesses, and indeed for 0:02:45.960 --> 0:02:49.120 the startups that create them. That's bad news for any 0:02:49.160 --> 0:02:53.880 industry that's yet to reach profitability or indeed sustainability, and 0:02:54.080 --> 0:02:57.360 especially for generative AI, which remains relying on a kind 0:02:57.400 --> 0:03:01.000 of an indefinite supply of cash to operate. Back in 0:03:01.040 --> 0:03:05.360 April twenty twenty three, Dylan Patel, chief analyst at Semi Analysis, 0:03:05.400 --> 0:03:09.120 calculated that GPT three, the previous generation of chech GPT 0:03:09.280 --> 0:03:12.600 current ones known as GPT four, cost around seven hundred 0:03:12.639 --> 0:03:15.440 thousand dollars a day to run. It's about twenty one 0:03:15.440 --> 0:03:17.680 million dollars a month, or two hundred and fifty million 0:03:17.720 --> 0:03:21.240 dollars a year. In October twenty twenty three, Richard Windsor, 0:03:21.440 --> 0:03:24.680 the research director at large of Counterpoint Research, which is 0:03:24.720 --> 0:03:28.240 one of the more reliable analyst houses, hypothesized that open 0:03:28.280 --> 0:03:30.960 AI's monthly cash burn was in the region of one 0:03:31.000 --> 0:03:34.880 point one billion dollars a month, based on them having 0:03:34.880 --> 0:03:37.440 to raise thirteen billion dollars from Microsoft, most of it, 0:03:37.520 --> 0:03:40.720 as I noted in credits for its Azure cloud computing 0:03:40.760 --> 0:03:44.000 service to run their models. It could be more, it 0:03:44.000 --> 0:03:47.280 could be less. As a private company, only investors and 0:03:47.320 --> 0:03:51.880 other insiders can possibly know what's going on in open Ai. However, 0:03:52.040 --> 0:03:55.240 four months later, Reuter's would report that open aye made 0:03:55.240 --> 0:03:57.600 about two billion dollars in revenue in twenty twenty three, 0:03:57.800 --> 0:04:00.720 a remarkable sum that much like every other story about 0:04:00.720 --> 0:04:05.000 open ai, never mentions profit. In fact, I can't find 0:04:05.000 --> 0:04:07.880 a single reporter that appears to have asked Sam Mormon 0:04:07.920 --> 0:04:11.280 about how much profit open ai makes, only breathless hype 0:04:11.520 --> 0:04:15.680 with no consideration of its sustainability. Even if open ai 0:04:15.840 --> 0:04:18.760 burns a tenth of windsors estema about one hundred million 0:04:18.800 --> 0:04:22.200 dollars a month, that's still far more money than they're making. 0:04:22.839 --> 0:04:25.000 There's not a single story out there talking about them 0:04:25.000 --> 0:04:28.719 making a profit, and I don't think they make in one. 0:04:28.960 --> 0:04:32.000 Here's one thing we can be certain of. Though things 0:04:32.040 --> 0:04:37.080 are getting more expensive, progress in generative AI means increasingly 0:04:37.120 --> 0:04:40.599 complex models, and as I previously mentioned open AI's attempted 0:04:40.920 --> 0:04:45.000 God damn it a Rakis model one, built specifically to 0:04:45.040 --> 0:04:48.359 wow Microsoft by making chatgiputy more efficient, failed to actually 0:04:48.400 --> 0:04:51.719 make it more efficient. Their attempts to make this a 0:04:51.760 --> 0:04:57.520 better company are not working. Windsor the aforementioned analyst in 0:04:57.560 --> 0:05:00.400 a separate blog also pointed out that there's nothing really 0:05:00.440 --> 0:05:05.560 sticky about these companies. There's nothing stopping someone from switching from, say, 0:05:05.680 --> 0:05:09.760 chat GPT to anthropics clawed two model. They're all trained 0:05:09.760 --> 0:05:13.280 on similar data sets, and they all produce very similar answers. 0:05:13.680 --> 0:05:16.200 And what one model might be better a one thing than another, 0:05:16.920 --> 0:05:22.480 they're fundamentally very very similar. There's also nothing stopping someone 0:05:22.520 --> 0:05:27.120 from simply giving up on generative AI altogether. It doesn't 0:05:27.160 --> 0:05:29.880 seem to be the plug and play automation god that 0:05:29.920 --> 0:05:34.160 everybody's been replicate to be, and judging by the plateauing 0:05:34.320 --> 0:05:37.719 Chat GPT user numbers, I think that might already be happening. 0:05:39.279 --> 0:05:41.880 It's also important to remember that while generative AI is 0:05:41.920 --> 0:05:46.160 shiny and new artificially, intelligence is absolutely not, and over 0:05:46.240 --> 0:05:49.160 the past decade it's found a number of homes from 0:05:49.240 --> 0:05:51.760 expensive security apps that detect when a hacker is trying 0:05:51.800 --> 0:05:54.800 to break into a corporate network. The spam filters proof 0:05:54.800 --> 0:05:58.279 for reading tools like grammarly, Plenty of Things, even Siri 0:05:58.400 --> 0:06:01.560 on your iPhone. In these context, AI is either a 0:06:01.640 --> 0:06:04.640 small component of a larger product or something that directly 0:06:04.640 --> 0:06:08.400 builds on human efforts. This stuff is actually valuable. AI 0:06:08.440 --> 0:06:10.960 based spam filters are typically better than those reliant on 0:06:11.000 --> 0:06:13.919 hand coded rules, for example, But it's also from a 0:06:13.960 --> 0:06:19.719 marketing perspective, can of boring generative AIS A law is 0:06:19.760 --> 0:06:23.360 that it can supplant humans either partially or entirely, producing 0:06:23.520 --> 0:06:26.760 entire creative works that otherwise would have taken hours and 0:06:26.800 --> 0:06:30.960 carried a real financial cost. But behind this glitzy technology 0:06:30.960 --> 0:06:34.760 and media hype, the unspoken truth is that generative AI 0:06:34.920 --> 0:06:38.080 holds way over the financial markets because it's regarded as 0:06:38.160 --> 0:06:41.120 a tool to eliminate an entire swarth of jobs in 0:06:41.160 --> 0:06:44.880 the creative and knowledge economies. It's a ghastly promise and 0:06:44.920 --> 0:06:48.880 it underpins the vast market value of otherwise commercially unviable 0:06:48.880 --> 0:06:52.240 generative AI companies like open Ai and Anthropic, and it's 0:06:52.240 --> 0:06:55.400 what is driving I believe the multi billion dollar investments 0:06:55.600 --> 0:06:59.920 we've seen from Microsoft, Amazon, and Google yeah, I see 0:07:00.000 --> 0:07:03.400 no evidence of mass adoption of generative AI, and by 0:07:03.440 --> 0:07:06.359 research suggests the enterprise adoption, which is the meat of 0:07:06.360 --> 0:07:09.640 what would actually make these companies money, it just isn't there. 0:07:10.840 --> 0:07:13.800 Deep within the earnings reports and the quotes of every 0:07:13.840 --> 0:07:16.840 major cloud provider claiming that the AI revolution is here 0:07:17.080 --> 0:07:21.200 is a deeply worrying trend. The AI revenue really isn't 0:07:21.200 --> 0:07:25.000 contributing much to the bottom line outside of vacuous media coverage, 0:07:26.280 --> 0:07:28.000 and I think the internal story is going to be 0:07:28.080 --> 0:07:39.560 much bleaker. In early March, The Information published the story 0:07:39.560 --> 0:07:43.800 about Amazon and Google tamping down GENERATIVEAI expectations, with these 0:07:43.840 --> 0:07:47.760 companies dousing their salespeople's excitement about the capabilities of the 0:07:47.800 --> 0:07:51.000 tech they're selling. A tech executive is quoted in the 0:07:51.080 --> 0:07:54.120 article saying that customers are beginning to struggle with questions, 0:07:54.240 --> 0:07:57.840 simple questions like is AI actually providing value? And how 0:07:57.880 --> 0:08:01.160 do I evaluate how AI is doing? And a Gartner 0:08:01.160 --> 0:08:05.520 analyst told Amazon Web Services sales staff that the AI 0:08:05.600 --> 0:08:08.280 industry was at the peak of the hype cycle around 0:08:08.320 --> 0:08:13.000 large language models and other generative AI, which is a 0:08:13.000 --> 0:08:16.560 somewhat specific code for it's not going to get much 0:08:16.600 --> 0:08:21.760 better anytime soon. This article confirms many of my suspicions 0:08:22.320 --> 0:08:26.320 that and I quote the information here, other software companies 0:08:26.320 --> 0:08:29.320 that have touted generative AI as a boon to enterprises 0:08:29.440 --> 0:08:33.040 are still waiting for revenue to emerge, citing the example 0:08:33.080 --> 0:08:37.679 of professional services firm KPMG buying forty seven thousand subscriptions 0:08:37.880 --> 0:08:41.360 to Microsoft's Copilot AI at a significant discount on the 0:08:41.360 --> 0:08:46.920 thirty dollars a c sticker price. Except KPMG bought the 0:08:46.960 --> 0:08:50.840 subscriptions without really have engaged whether their employees actually got 0:08:50.880 --> 0:08:53.400 anything out of it. They bought it, and I'm not 0:08:53.520 --> 0:08:57.679 kidding you entirely so that if any KPMG customers ask 0:08:57.800 --> 0:09:03.559 questions about AI, they'll be able to answer them. Was 0:09:03.640 --> 0:09:06.400 so clearly in the bot Oh my god. Anyway, as 0:09:06.400 --> 0:09:09.600 I've hinted, it's also not obvious how much AI actually 0:09:09.640 --> 0:09:13.120 contributes to the bottom line. In Microsoft's Q four twenty 0:09:13.200 --> 0:09:16.720 twenty three earnings report, gief financial officer Amy Hood reported 0:09:16.720 --> 0:09:19.559 that six points of revenue growth in its Azureine Cloud 0:09:19.559 --> 0:09:23.400 Services division was attributed to AI services. I went around 0:09:23.400 --> 0:09:25.920 the web and I read every bloody article about their earnings. 0:09:26.400 --> 0:09:29.120 I looked, and I looked and everyone was saying, Oh, 0:09:29.160 --> 0:09:31.080 this is really good. I found someone who said it 0:09:31.120 --> 0:09:33.559 was six percent of their revenue, and I went, that 0:09:33.679 --> 0:09:36.040 sounds like complete bollocks to me. So I went and 0:09:36.080 --> 0:09:39.520 spoke with Jordan Novette, who's covered Microsoft for many years. 0:09:39.520 --> 0:09:41.439 He is a great cloud reporter over at CNBC, and 0:09:41.480 --> 0:09:45.160 he actually covered Microsoft's earnings for the NBC itself, and 0:09:45.200 --> 0:09:49.000 he confirmed that what this means is that AI contributed 0:09:49.160 --> 0:09:52.480 six percent of the thirty percent of year over year 0:09:52.520 --> 0:09:56.840 growth in Microsoft's as your Cloud services. That is a 0:09:56.880 --> 0:10:01.160 percentage of a percentage. So by the way that that means, 0:10:01.200 --> 0:10:03.959 so thirty percent growth year of year. So six percent 0:10:04.040 --> 0:10:06.960 year of a year growth is from AI. Could be good, 0:10:07.040 --> 0:10:09.080 but also all of the rest of it came not 0:10:09.120 --> 0:10:11.559 from new products, just from the natural growth of the company. 0:10:13.320 --> 0:10:15.520 It's unclear how much money that really is, but six 0:10:15.600 --> 0:10:18.280 percent of the year over year growth isn't really that 0:10:18.400 --> 0:10:23.360 exciting anyway elsewhere. Amazon CEO Andy Jasse, who took over 0:10:23.400 --> 0:10:26.120 from Bezos a few years ago and was the chief 0:10:26.120 --> 0:10:29.360 of Amazon Web Services, said that generative AI revenue was 0:10:29.400 --> 0:10:32.560 still relatively small, but don't worry, you said it would 0:10:32.600 --> 0:10:34.600 drive tens of billions of dollars of revenue over the 0:10:34.640 --> 0:10:38.319 next several years, adding that virtually every consumer business Amazon 0:10:38.360 --> 0:10:42.160 operated in already had or would have generative AI offerings. 0:10:42.720 --> 0:10:46.080 Now they can just say that stuff. I really want 0:10:46.080 --> 0:10:47.520 you to know. You can say what you want on 0:10:47.559 --> 0:10:50.520 earning scores, as long as you're not just outright lying, 0:10:50.559 --> 0:10:52.679 like saying we have one hundred billion dollars in cash, 0:10:52.679 --> 0:10:54.760 but you have fifty dollars. That is a lie. You 0:10:54.800 --> 0:10:57.000 can't do that. But you can be like iah At 0:10:57.080 --> 0:10:59.200 We've got all sorts of AI and everything. Now it's 0:10:59.240 --> 0:11:02.319 bloody magically. Take you don't take a look, but it's there. 0:11:02.320 --> 0:11:05.120 I promise you. They can just say what they want. 0:11:06.320 --> 0:11:09.760 But don't worry, they're not the only ones. Salesforce chief 0:11:09.760 --> 0:11:13.080 financial officer Amy Weaver said in their most recent earning 0:11:13.120 --> 0:11:17.360 score that Salesforce was not factoring in material contribution from 0:11:17.400 --> 0:11:21.240 Salesforce's numerous AI products in its financial year twenty twenty five. 0:11:21.280 --> 0:11:26.079 Guidance software company Adobe shares slid in their last earnings. 0:11:26.120 --> 0:11:28.680 It's the company failed to generate meaningful revenue from its 0:11:28.679 --> 0:11:31.600 masses of AI products, with analysts now worried about its 0:11:31.640 --> 0:11:36.000 ability to actually monetize any of these generative products. Service 0:11:36.040 --> 0:11:38.720 now claimed to its earnings that generative AI was meaningfully 0:11:38.720 --> 0:11:40.840 contributed to its bottom line. Yet a story from The 0:11:40.840 --> 0:11:44.640 Information quotes their chief financial officer as saying that from 0:11:44.640 --> 0:11:48.760 a revenue contribution perspective, it's not going to be huge. 0:11:49.679 --> 0:11:51.400 Going to be a bit honest and feeling a little 0:11:51.440 --> 0:11:55.440 insane with this stuff. I feel crazy every time I 0:11:55.520 --> 0:11:59.920 think about these stories, because elsewhere in the media, so 0:12:00.360 --> 0:12:03.960 many people are saying how big and successful the generative 0:12:03.960 --> 0:12:07.600 AI revolution is, and it's going to be Yeah. Every 0:12:07.600 --> 0:12:09.920 time I look at the actual places where they write 0:12:09.960 --> 0:12:12.680 down how much money it makes, any of the actual 0:12:12.720 --> 0:12:19.400 signs of growth and significance and utility and adoption, it's 0:12:19.440 --> 0:12:22.640 just not there. It's just breathless hype, with this kind 0:12:22.640 --> 0:12:27.720 of whisper of stagnation and non existent adoption. And while 0:12:27.760 --> 0:12:31.200 there are startups beginning to mind usefulness out of general AI, 0:12:31.520 --> 0:12:34.360 and they do inside by automating internal queries and customer 0:12:34.360 --> 0:12:39.920 sport questions, these are integrations rather than revolutions, and they're 0:12:39.920 --> 0:12:44.640 far from the substance of a true movement. Maybe the 0:12:44.760 --> 0:12:47.280 darker truth of the generative AI boom is that it's 0:12:47.320 --> 0:12:50.480 a feature, not a product, and that these features might 0:12:50.520 --> 0:12:54.440 be built entirely off the back of large language models 0:12:54.480 --> 0:12:59.319 which are unsustainable to run, grow, or even make better. 0:13:00.480 --> 0:13:03.360 What if AI only drives a couple percentage points of 0:13:03.400 --> 0:13:06.160 real revenue growth of these companies. What if what we're 0:13:06.160 --> 0:13:10.640 seeing today is the upper limit, not the beginning. Honestly, 0:13:10.679 --> 0:13:12.760 I'm beginning to believe that a large part of the 0:13:12.760 --> 0:13:15.560 AI boom is just hot air, and it's being pumped 0:13:15.640 --> 0:13:20.319 up through a combination of executive bullshittery and very compliant 0:13:20.360 --> 0:13:23.280 media that's so happy to write stories imagining what AI 0:13:23.360 --> 0:13:27.240 can do, yet seems unable to check what it can 0:13:27.360 --> 0:13:33.160 do or what it's doing. It's so weird. Now, there's 0:13:33.200 --> 0:13:35.360 a bloke over at the Wall Street Journal called Chipcutter 0:13:35.360 --> 0:13:36.959 who should really look into if you want to know 0:13:37.000 --> 0:13:38.840 why your boss keeps asking you to go back to 0:13:38.840 --> 0:13:41.400 the office. Wall Street Journal's Chip Cutter. He loves to 0:13:41.400 --> 0:13:44.400 write things about how bosses are good and how returning 0:13:44.440 --> 0:13:46.760 to the office is good. He wrote a piece in 0:13:46.840 --> 0:13:49.600 March about how AI is being integrated into the office, 0:13:49.640 --> 0:13:51.719 and most of it was just hundreds of words of 0:13:52.000 --> 0:13:54.880 him guessing about what people might do but when he 0:13:54.920 --> 0:13:57.680 gets to the bottom and he starts talking about companies 0:13:57.760 --> 0:14:02.040 using it, it's almost entirely exact samples of people saying, yeah, 0:14:02.040 --> 0:14:04.000 it makes too many mistakes for us to rely on it, 0:14:04.040 --> 0:14:07.400 and we're just experimenting it. Elsewhere in the media, the 0:14:07.440 --> 0:14:10.000 New York Times talked with Salesforce is head of AI 0:14:10.440 --> 0:14:14.599 Clara's share and in this I think six hundred and 0:14:14.640 --> 0:14:18.120 seven hundred word article, didn't really get to say much 0:14:18.160 --> 0:14:20.880 of anything about AI or what their products do. All 0:14:20.920 --> 0:14:24.840 she said was that the Einstein Trust layer handles data. 0:14:24.960 --> 0:14:26.840 And you may think I'm being facetious here, that's all 0:14:26.880 --> 0:14:29.920 she said about that, and then she added that it 0:14:29.960 --> 0:14:33.280 would be transformational for jobs the way the internet was. 0:14:34.320 --> 0:14:38.120 What what does that mean? Why am I reading this 0:14:38.160 --> 0:14:40.680 in the newspaper? Why is this what I read in 0:14:40.720 --> 0:14:44.840 the newspaper? How is this helping? I know I rant 0:14:44.880 --> 0:14:46.480 a lot on this podcast, and I'm going to keep 0:14:46.480 --> 0:14:49.000 doing it. You're stuck with me, all right, it's free, Okay, 0:14:49.200 --> 0:14:51.360 you don't pay for this unless you do cooler zone media, 0:14:51.360 --> 0:14:54.640 which you should pay for anyway. I know I'm ranting, 0:14:54.680 --> 0:14:58.280 But the reason that this stuff really infuriates me is 0:14:58.480 --> 0:15:01.640 it's misinformation on some level. I know it's kind of 0:15:01.720 --> 0:15:04.720 dramatic to say, oh, they're misinforming people by suggesting that 0:15:04.760 --> 0:15:08.400 AI can do stuff, but it is. It is misinformation. 0:15:08.680 --> 0:15:11.720 When you're letting corporate executives go in the newspaper and 0:15:11.760 --> 0:15:14.680 talk about how amazing their products will be without asking 0:15:14.720 --> 0:15:17.200 them what they can do today, you're just giving them 0:15:17.240 --> 0:15:20.920 free press. You're not giving them credit for stuff they've done. 0:15:21.040 --> 0:15:23.200 You're giving them credit for things they're making up on 0:15:23.280 --> 0:15:26.520 the spot. And when you do that, you make the 0:15:26.680 --> 0:15:30.800 rich richer and the poor poorer. You centralize power in 0:15:30.800 --> 0:15:34.720 the hands of assholes, people who are excited, people who 0:15:34.760 --> 0:15:39.160 are borderline masturbatory, jumping around saying, oh God, I can't 0:15:39.200 --> 0:15:43.840 wait till and replace humans with fucking computers. Good news 0:15:43.920 --> 0:15:45.920 is they're not going to be able to, but that's 0:15:45.960 --> 0:15:47.920 what they're excited about, and that's what they're getting media 0:15:47.960 --> 0:15:52.280 coverage around. The media has been fooled, just like they 0:15:52.320 --> 0:15:56.400 were with the metaverus, by this specious promise train of 0:15:56.440 --> 0:16:01.120 the generative AI generation, and these worthless executives champions these 0:16:01.280 --> 0:16:05.560 half truths and this magical thinking has spread far faster 0:16:05.840 --> 0:16:08.320 due to the fact that AI actually exists and is 0:16:08.400 --> 0:16:11.760 doing something, and it's actually much easier to imagine how 0:16:11.760 --> 0:16:15.000 it might change our lives, unlike the metaverse. Even if 0:16:15.040 --> 0:16:17.640 the way it might do so is somewhere between improbable 0:16:17.680 --> 0:16:22.840 and impossible, it is easy to think about how my work. 0:16:23.320 --> 0:16:25.040 You know, you could use an AI to or make 0:16:25.120 --> 0:16:28.960 data entry or boring busy work. Surely all of this 0:16:29.040 --> 0:16:31.160 you can automate, right, And when you use chat GPT 0:16:31.520 --> 0:16:33.760 you can almost kind of, sort of somewhat see how 0:16:33.800 --> 0:16:36.840 it might happen. Even if when you open up chat 0:16:36.880 --> 0:16:39.720 GPT and try and make it do something, it's always 0:16:39.760 --> 0:16:42.560 a bit off, never seems to quite do it. I'm 0:16:42.600 --> 0:16:44.560 in a very in my day job, my PR firm. 0:16:44.720 --> 0:16:47.760 I'm in a spreadsheet and document heavy business. Of all 0:16:47.800 --> 0:16:50.000 the people who this could help, you think it would 0:16:50.000 --> 0:16:53.040 be me A lot of my work is hey, all 0:16:53.080 --> 0:16:54.760 of these things I need him in a spreadsheet. It 0:16:54.800 --> 0:16:59.480 can't bloody do it. And I'm sure you listeners will 0:16:59.520 --> 0:17:02.200 probably email me and say, oh, I've used CHATGPT for this, 0:17:02.560 --> 0:17:05.879 don't care. I really mean that this thing is not 0:17:06.040 --> 0:17:08.080 changing the world. And actually I think far more of 0:17:08.119 --> 0:17:09.920 you have already shared. Thank you. By the way, easy. 0:17:09.920 --> 0:17:12.080 At Better off Line dot com. You can email me 0:17:12.160 --> 0:17:14.520 your ideas and your angry comments, so you can go 0:17:14.560 --> 0:17:17.879 on the reddit to complain. But the thing I'm hearing 0:17:17.920 --> 0:17:19.879 for most people is, yeah, I've tried it and it 0:17:19.920 --> 0:17:22.920 didn't do enough. I tried it and there were too 0:17:22.920 --> 0:17:26.639 many mistakes. There was a Wall Street Journal article back 0:17:26.680 --> 0:17:31.680 in February about how Amazon and Google were having trouble 0:17:31.760 --> 0:17:36.080 selling AI services because well, when they went to sell 0:17:36.119 --> 0:17:39.720 them these companies, these financial services companies in particular, they 0:17:39.720 --> 0:17:42.399 were saying, yeah, but these hallucinations could actually get the 0:17:42.440 --> 0:17:47.159 sec mad at us. And the answer that they had was, yeah, 0:17:47.480 --> 0:17:49.200 what if we just made it so that the models 0:17:49.200 --> 0:17:53.639 would sometimes say they don't know stuff. Every time you 0:17:53.720 --> 0:17:56.800 get to a reckoning with AI where you want it 0:17:56.840 --> 0:18:00.800 to be better, where you're like, hey, AI executive, how 0:18:00.800 --> 0:18:04.000 will this actually be fixed these hallucination problems, for example, 0:18:04.160 --> 0:18:07.199 they come up with the most meanly mouthship. And I 0:18:07.280 --> 0:18:09.720 truly believe it's because there is no answer to these problems, 0:18:09.760 --> 0:18:13.480 as I said in the previous episode, and I think 0:18:13.520 --> 0:18:16.360 that's why I can't find any companies that have integrated 0:18:16.400 --> 0:18:18.639 generative AI in a way that's truly improved their bottom 0:18:18.720 --> 0:18:22.440 line other than Klana, which allows you to do zero 0:18:22.520 --> 0:18:25.720 percent interest free loans on almost anything. It's a very 0:18:25.760 --> 0:18:29.600 worrying company anyway. They claimed that their AI powered support 0:18:29.680 --> 0:18:34.000 bot was estimated to drive a forty million dollar amount 0:18:34.080 --> 0:18:38.159 in profit improvement in twenty twenty four, which does not, 0:18:38.280 --> 0:18:40.840 by the way, despite it being troubleted by members of 0:18:40.880 --> 0:18:43.640 the media, otherwise mean that they made forty million dollars 0:18:43.680 --> 0:18:47.040 in profit. I actually can't find what profit improvement refers to. 0:18:48.400 --> 0:18:51.840 And this is the classic AI boom story. By the way, 0:18:51.960 --> 0:18:57.120 there's always this weird verbal judo going on where they're like, yeah, sir, 0:18:57.840 --> 0:19:00.800 forty million dollars in profit improvement, upwards, downwards and side 0:19:00.800 --> 0:19:03.760 to sidemark, it's really good, And I think it's just 0:19:03.880 --> 0:19:08.240 headline grabbing. I think it's just buzz. And despite fears 0:19:08.320 --> 0:19:11.600 to the contrary, AI doesn't appear to be replacing a 0:19:11.680 --> 0:19:14.200 large amount of workers, and when it has, the results 0:19:14.240 --> 0:19:17.960 have been pretty terrible, like when Microsoft replaced MSN dot 0:19:18.000 --> 0:19:21.040 COM's editorial team with a series of AI bots that 0:19:21.080 --> 0:19:25.040 have spread misinformation and conspiracy theories, things like Joe Biden 0:19:25.119 --> 0:19:29.440 falling asleep. It's so weird. Interestingly, There was also a 0:19:29.480 --> 0:19:32.480 study in Boston Consultancy Group, and just as a note, 0:19:32.640 --> 0:19:35.520 if anyone would love the opportunity to just replace workers 0:19:35.520 --> 0:19:39.840 with robots, it's BCG, mckenzi, Accentia. All these companies were 0:19:39.880 --> 0:19:43.920 absolutely they would be giving, however much open ai wanted 0:19:43.960 --> 0:19:46.040 to do that, and then they would charge fifty million 0:19:46.080 --> 0:19:47.840 dollars for in integration that didn't work, which I guess 0:19:47.880 --> 0:19:51.000 makes AI perfect them putting that aside. In a study 0:19:51.000 --> 0:19:54.439 from BCG, they found the consultants that solved business problems 0:19:54.480 --> 0:19:57.520 with open AI's GPT four model performed twenty three percent 0:19:57.920 --> 0:20:01.560 worse than those who didn't use it, even when the 0:20:01.640 --> 0:20:04.800 consultant was warned about the limitations of generative AI and 0:20:04.800 --> 0:20:10.920 the risk of hallucinations. Yeah, really great stuff. To be clear, 0:20:11.400 --> 0:20:15.280 I am not advocating for the replacement of workers with AI. However, 0:20:15.840 --> 0:20:18.400 I'm saying that if it was actually capable of replacing 0:20:18.440 --> 0:20:22.120 human outputs, even if it was even anywhere near doing so, 0:20:22.520 --> 0:20:25.480 any number of these massive, horrifying firms would be doing 0:20:25.520 --> 0:20:27.920 so at scale, and planning to do so more as 0:20:27.920 --> 0:20:31.440 the models improve. They'd be fuddling cash right up. Open 0:20:31.480 --> 0:20:35.040 ayes ass it would be incredible, but the reality of 0:20:35.080 --> 0:20:37.920 the AI boom is kind of a little more boring. 0:20:39.000 --> 0:20:42.440 It recently came out that Amazon's cashless just walk out 0:20:42.480 --> 0:20:45.199 technology in some of their stores. You could walk in, 0:20:45.359 --> 0:20:47.320 scan a QR code, and then you could just grab 0:20:47.320 --> 0:20:51.720 your rouse tomato sauce and your condoms or whatever, your 0:20:51.760 --> 0:20:53.560 weird magazines. I don't know what they sell, and there 0:20:53.560 --> 0:20:55.399 I'm not giving any more money to Amazon than I 0:20:55.520 --> 0:20:59.159 need to. Anyways, everyone thought, oh, it's just AI. You 0:20:59.160 --> 0:21:01.399 could just walk in. It's the cameras would tell you 0:21:01.440 --> 0:21:04.320 through computer vision what you had bought. It would be great. 0:21:04.560 --> 0:21:06.760 Now it turns out that there's one thousand workers in 0:21:06.760 --> 0:21:11.680 India that were monitoring these cameras and approving transactions. Worse still, 0:21:12.040 --> 0:21:14.760 open AI used Kenyan workers who were paid less than 0:21:14.800 --> 0:21:19.159 two dollars an hour to train chet GPT's outputs, and 0:21:19.200 --> 0:21:22.280 they currently pay fifteen dollars an hour. I think for 0:21:22.320 --> 0:21:26.680 American contractors, no benefits of course, you know, fuck workers, right, 0:21:27.119 --> 0:21:30.320 that's the thing underneath this whole thing. It's just this 0:21:30.480 --> 0:21:33.680 undercurrent of disrespect for human beings, and it pisses me off. 0:21:34.200 --> 0:21:36.120 And I realized I'm pissed off about a lot of things. 0:21:36.119 --> 0:21:38.720 You'd be listening for like an hour now half an 0:21:38.760 --> 0:21:42.920 hour in this episode. Anyway, I'll keep going. But yeah, 0:21:43.000 --> 0:21:46.120 like I said, if AI was changing things, if AI 0:21:46.280 --> 0:21:50.480 was actually capable of replacing a person, it would have happened. 0:21:50.800 --> 0:21:53.360 It would be happening right now, It'd be happening at scale. 0:21:53.600 --> 0:21:57.879 It would be so much worse than things feel now, unless, 0:21:57.880 --> 0:22:01.720 of course, it just wasn't possible. What if what we're 0:22:01.720 --> 0:22:04.760 seeing today is not a glimpse of the future, but 0:22:04.840 --> 0:22:08.080 actually the new terms of the present. What if generator 0:22:08.119 --> 0:22:11.199 of AI isn't actually capable of doing much more than 0:22:11.200 --> 0:22:13.760 what we're seeing today. What if there's not really a 0:22:13.840 --> 0:22:16.000 clear timeline when it will actually be able to do more. 0:22:17.440 --> 0:22:21.080 What if this entire hype cycle has been built, goosed, 0:22:21.600 --> 0:22:24.760 and propped up by this compliant media, ready and willing 0:22:24.800 --> 0:22:29.000 to take whatever these career embellishing bullshitters have to say. 0:22:29.920 --> 0:22:32.359 What if this is just another metaverse, but with a 0:22:32.400 --> 0:22:36.360 little bit more product. Every single time I've read about 0:22:36.400 --> 0:22:40.280 the amazing things that artificial intelligence can do, I just 0:22:40.359 --> 0:22:42.800 see somebody attempting to add fuel to a fire that's 0:22:42.800 --> 0:22:45.600 close to going out. When the Wall Street Journals Joanna 0:22:45.640 --> 0:22:48.439 Stern wrote about Sora open ayes yet to be released 0:22:48.520 --> 0:22:52.359 video generating chatbot. She talked about how its photorealistic clips 0:22:52.359 --> 0:22:55.160 were good enough to freak her out. And I get 0:22:55.200 --> 0:23:00.080 it at first glance. These do look like people, these 0:23:00.119 --> 0:23:04.520 images do. They look like something approaching a video. They 0:23:04.560 --> 0:23:09.439 look almost real, kind of like text some chat GBT 0:23:09.640 --> 0:23:13.400 is almost right or it's right fully, but it doesn't 0:23:13.880 --> 0:23:18.040 feel right. But much like the rest of these outputs, 0:23:19.480 --> 0:23:21.600 you look a little closer, and they have these weird 0:23:21.680 --> 0:23:24.240