WEBVTT - I'm Mad As Hell, And I'm Not Gonna Take It Anymore 0:00:03.160 --> 0:00:07.360 Media. I'm mad as hell and I'm not gonna take 0:00:07.360 --> 0:00:23.279 it anymore. Welcome to bear Offline. I'm your host ed Zetron. Now. 0:00:23.280 --> 0:00:25.280 This is the second part of a two part series 0:00:25.280 --> 0:00:25.759 about how. 0:00:25.720 --> 0:00:29.080 Open Ay's stupid, bloody growth myth has gone the entire 0:00:29.160 --> 0:00:32.800 media ecosystem, forcing people who really should know better to 0:00:32.840 --> 0:00:37.960 repeat absolutely insane things as though they're actually perfectly reasonable. Now, 0:00:37.960 --> 0:00:40.199 if you haven't already listened to the last episode and 0:00:40.240 --> 0:00:43.639 come back, just go, just go. I don't I'll need 0:00:43.680 --> 0:00:45.479 you to listen to it otherwise, Let's talk about why 0:00:45.479 --> 0:00:47.760 this bubble actually inflated. And I realized I did a 0:00:47.760 --> 0:00:51.200 monologue about this, but it's so obvious that I needed 0:00:51.240 --> 0:00:54.360 to do a longer episode. But let's start simple. The 0:00:54.440 --> 0:00:57.840 term artificial intelligence is bastardized to the point it effectively 0:00:57.880 --> 0:01:00.880 means nothing and everything at the same time. When people 0:01:00.920 --> 0:01:03.400 hear AI, they think of an autonomous intelligence that can 0:01:03.440 --> 0:01:06.240 do things for them, and generative AI can do things 0:01:06.240 --> 0:01:08.320 for you, like generate an image or a text from 0:01:08.360 --> 0:01:11.640 a simple prompt. As a result, it's easy to manipulate 0:01:11.680 --> 0:01:14.600 people who don't know much about tech, or even try 0:01:14.640 --> 0:01:16.480 it in the first place, into believing that this will 0:01:16.560 --> 0:01:18.759 naturally progress from It can create a bunch of text 0:01:18.800 --> 0:01:20.160 for me that I have to write for my job 0:01:20.280 --> 0:01:22.440 just by me typing in a prompt. Too, it can 0:01:22.480 --> 0:01:25.759 do my job for me just by typing in a prompt. Basically, 0:01:25.840 --> 0:01:29.200 everything you've read about the future of AI extrapolates generative 0:01:29.240 --> 0:01:32.039 AI's ability to sort of generate something a human would 0:01:32.120 --> 0:01:35.280 make and turns it into a It can do whatever 0:01:35.319 --> 0:01:37.800 a human can do, or because some tech in the 0:01:37.840 --> 0:01:41.160 past has sometimes been bad at the beginning and linearly 0:01:41.200 --> 0:01:45.679 improved as time drags on. This illogical thinking underpins the 0:01:45.840 --> 0:01:48.640 entire generative AI boom because we found out exactly how 0:01:48.720 --> 0:01:50.559 many people do not know what the fuck they're talking 0:01:50.640 --> 0:01:52.920 about and are willing to believe the last semi intelligent 0:01:52.920 --> 0:01:55.680 person that they talk to. Generatif AI is kind of 0:01:55.720 --> 0:01:59.240 a remarkable karner, just good enough version of human expression 0:01:59.280 --> 0:02:01.920 to get it past the gatekeepers in finance and the media, 0:02:02.040 --> 0:02:04.520 knowing that neither will apply a second gear of critical 0:02:04.560 --> 0:02:07.080 thinking beyond hook as we'll do an AI now. The 0:02:07.160 --> 0:02:10.799 expectation that general IVAI will transform into this much more 0:02:10.840 --> 0:02:14.400 powerful version requires you to first ignore its existing limitations, 0:02:14.639 --> 0:02:16.440 believing it to be more capable than it is, and 0:02:16.480 --> 0:02:18.280 also ignore the fact that these models have yet to 0:02:18.320 --> 0:02:21.880 show meaningful improvement over the past few years unless you're looking, 0:02:21.919 --> 0:02:24.760 at course, at the benchmarks that they are built specifically 0:02:24.760 --> 0:02:28.760 to pass. They still halluciname, they're still ungodly expensive to run, 0:02:28.840 --> 0:02:31.440 they're still unreliable, and they still don't really do that much. 0:02:31.960 --> 0:02:34.720 We're still chat GPT's growth as galvanized these people into 0:02:34.760 --> 0:02:37.800 believing that this is a legitimate, meaningful movement rather than 0:02:37.840 --> 0:02:41.840 the most successful public relation campaign of all time. Think 0:02:41.880 --> 0:02:44.480 of it like this, and some of you really don't 0:02:44.560 --> 0:02:47.519 like this argument is something couldn't be this big just 0:02:47.560 --> 0:02:48.399 because of the media. 0:02:48.639 --> 0:02:49.280 It really could. 0:02:49.440 --> 0:02:52.440 If almost every single media outlet talked about one thing, 0:02:52.560 --> 0:02:54.720 this thing being generally of AI, and that one thing 0:02:54.800 --> 0:02:57.720 was available from one company, Open Ai, wouldn't it look 0:02:57.760 --> 0:03:01.079 exactly how things look today? Got open ai with hundreds 0:03:01.120 --> 0:03:03.000 of millions of monthly active users, and then a bunch 0:03:03.000 --> 0:03:05.519 of other companies, including big tech firms with multi trillion 0:03:05.520 --> 0:03:08.320 dollar market caps, with somewhere between ten and sixty nine 0:03:08.360 --> 0:03:11.240 million monthly active users, and I don't count Gemini because 0:03:11.240 --> 0:03:14.440 they're cheating. You can't slap it on Google Assistant. That's unfair. 0:03:15.040 --> 0:03:17.840 What we're seeing here is one company taking most of 0:03:17.840 --> 0:03:20.120 the users and money available and doing so because the 0:03:20.200 --> 0:03:23.440 media fucking helped them. People aren't amazed by chat GPT. 0:03:23.639 --> 0:03:24.320 They're curious. 0:03:24.360 --> 0:03:26.680 They're curious about why the media won't shut up about it, 0:03:27.280 --> 0:03:30.440 and we have to realize and recognize and accept that 0:03:30.520 --> 0:03:33.360 part of the reason this bubble was inflated was the 0:03:33.360 --> 0:03:36.800 failure of Google Search. Everyone I talk to that uses 0:03:36.840 --> 0:03:39.120 chat gpt regularly uses it as either a way to 0:03:39.160 --> 0:03:42.240 generate shitty limericks or as a replacement for Google Search, 0:03:42.280 --> 0:03:44.400 a product that Google is deliberately made worse as a 0:03:44.400 --> 0:03:47.880 means of increasing profits. Listen to our award winning episode 0:03:48.320 --> 0:03:50.520 The Man That Destroyed Google Search. I'm going to say 0:03:50.520 --> 0:03:53.600 that forever, and by the way, that's absolutely what you 0:03:53.640 --> 0:03:56.760 should listen to. It's wherebe award winning the best Business episode. 0:03:56.800 --> 0:03:57.640 We won it, folks. 0:03:57.760 --> 0:03:59.840 Anyway, I hate to say this, I really don't like 0:03:59.840 --> 0:04:01.800 saying this. I don't like to give it to them, 0:04:01.800 --> 0:04:05.160 but chat GPT, if I'm honest, is better at processing 0:04:05.200 --> 0:04:07.880 search strings than Google Search, which is not so much 0:04:07.880 --> 0:04:10.640 a sign that chatchpt is good at something as it 0:04:10.720 --> 0:04:13.280 is that Google has stopped innovating in any meaningful way. 0:04:13.720 --> 0:04:15.800 Over time, Google search should have become something that was 0:04:15.840 --> 0:04:18.640 able to interpret searches into kind of like a perfect result, 0:04:18.640 --> 0:04:21.120 which would require the company to improve how it processes 0:04:21.160 --> 0:04:25.400 your requests. It should be how Google used to feel magical. Instead, 0:04:25.400 --> 0:04:28.120 Google search has become dramatically worse, mostly because the company's 0:04:28.120 --> 0:04:30.479 incentives change from help people find something on the web 0:04:30.520 --> 0:04:33.000 to funnel as much traffic and show as many add 0:04:33.080 --> 0:04:37.560 impressions as possible on Google dot Com. It does little 0:04:37.560 --> 0:04:40.119 annoying things like ignoring random words in the search string, 0:04:40.240 --> 0:04:42.120 even though those words were there for a reason. And 0:04:42.160 --> 0:04:46.120 Google doesn't know better than you what you want ideally, 0:04:46.160 --> 0:04:48.239 though they should when it comes to a search product. 0:04:48.279 --> 0:04:51.760 That's ideally what you get someone to search something for you, 0:04:51.760 --> 0:04:54.600 You want them to bring back the best thing. And 0:04:55.480 --> 0:04:58.640 by this point, Google search should have been far more magical, 0:04:58.720 --> 0:05:01.040 more capable of taking a de witched question and turning 0:05:01.040 --> 0:05:02.919 it into a great answer. Would said answer being a 0:05:02.920 --> 0:05:06.400 result from the internet right. Note that nothing I'm saying 0:05:06.440 --> 0:05:09.039 here is actually about generating a results. It's about processing 0:05:09.080 --> 0:05:12.240 a user's query and presenting an answer. The very foundation 0:05:12.320 --> 0:05:14.239 of computing and the thing that Google at one point 0:05:14.279 --> 0:05:16.400 was the best in the world at doing. Thanks to 0:05:16.440 --> 0:05:18.599 Propagar Ragavan, the former head of Ads that led a 0:05:18.640 --> 0:05:20.719 coup to become the head of Search, Google is pulled 0:05:20.720 --> 0:05:23.720 away from being the meaningful source of information on the Internet. 0:05:24.000 --> 0:05:26.960 It really is actually quite sad when you think about it. 0:05:27.839 --> 0:05:30.920 And I'd argue that chat GPT filled the void left 0:05:30.920 --> 0:05:33.279 by Google Search by doing the thing that people wanted 0:05:33.320 --> 0:05:36.839 Google Search to do, answer a question even if the 0:05:36.920 --> 0:05:40.160 user isn't really sure how to ask it. Google Search 0:05:40.160 --> 0:05:42.520 has become clunky and messy, putting the burden of using 0:05:42.560 --> 0:05:44.719 the service on the user rather than helping fill the 0:05:44.720 --> 0:05:47.480 gap between a query and an answer in any meaningful way. 0:05:47.880 --> 0:05:50.320 Google's AI summaries don't even try to do what chat 0:05:50.320 --> 0:05:53.320 GPT does. They just generate summaries based on search results 0:05:53.320 --> 0:05:56.000 and say, uh, mate, is this what you want? I 0:05:56.000 --> 0:05:58.920 don't fucking know? Only may I only mate latsambillion does 0:05:58.960 --> 0:06:01.680 a core give me a fucking break? But one though 0:06:01.720 --> 0:06:04.840 on Google's AI summaries, they're designed to answer a question 0:06:04.960 --> 0:06:07.440 rather than provide a right answer. There's a distinction that 0:06:07.480 --> 0:06:09.520 needs to be made here because it speaks the underlying 0:06:09.640 --> 0:06:13.280 utility of search itself. One good illustration came earlier this 0:06:13.360 --> 0:06:15.320 week when someone noticed that you could ask Google to 0:06:15.320 --> 0:06:17.560 explain the meaning of a completely made up phrase and 0:06:17.600 --> 0:06:20.680 it would dutifully obey you. Two dry frogs in the situation, 0:06:20.800 --> 0:06:22.480 Google said, refer to a group of people in an 0:06:22.480 --> 0:06:26.320 awkward or difficult social situation. Not every insect has a mortgage, 0:06:26.320 --> 0:06:28.600 Google claimed, there's a humorous way of explaining that not 0:06:28.720 --> 0:06:33.040 everything is as it seems. But my favorite is, sorry, 0:06:33.040 --> 0:06:36.719 I haven't read this bags. Big winky on the skillet 0:06:36.720 --> 0:06:39.039 bowl is apparently a slang TERWN that refers to a 0:06:39.040 --> 0:06:40.440 piece of bread with an egg in the middle. 0:06:40.520 --> 0:06:44.000 Or very funny. Sure, but is it useful? 0:06:44.080 --> 0:06:46.520 No? And it's also really cool that this company just 0:06:46.600 --> 0:06:50.120 makes billions of dollars off of this. With all its 0:06:50.160 --> 0:06:52.080 data and all its talent, Google has put out the 0:06:52.160 --> 0:06:55.600 laziest version of an LLM on top of its questionably 0:06:55.680 --> 0:06:59.040 functional search product as a means of impressing shareholders. And 0:06:59.080 --> 0:07:02.000 I'd like to say the results speak for themselves. Now, 0:07:02.040 --> 0:07:03.440 it must be clear none of this is to say 0:07:03.480 --> 0:07:08.200 that chat GPT is good like it really isn't. It's 0:07:08.240 --> 0:07:12.160 just better at understanding a user's request than Google Search. Yes, 0:07:12.200 --> 0:07:14.720 I fundamentally believe, by the way, that five hundred million 0:07:14.720 --> 0:07:16.960 people a week could be using chat GPT as some 0:07:17.000 --> 0:07:19.240 sort of search replacement. And no, no, I do not 0:07:19.400 --> 0:07:22.240 believe that's a functional business model, in part because if 0:07:22.280 --> 0:07:26.680 it was chat gpt open AIE, they'd have a functional business. 0:07:26.720 --> 0:07:30.000 And to be clear, that's a load bearing kit because 0:07:30.320 --> 0:07:32.840 I know that five hundred million people aren't using chat 0:07:32.840 --> 0:07:34.200 GPT as a search replacement. 0:07:34.280 --> 0:07:35.400 It's hypothetical. 0:07:35.400 --> 0:07:38.160 We actually don't know, and we cannot trust what comes 0:07:38.160 --> 0:07:41.320 out of them, not that they share. It's not just 0:07:41.360 --> 0:07:43.360 the open aiy as a shambles of a company. 0:07:43.480 --> 0:07:44.000 It appears that. 0:07:44.000 --> 0:07:46.280 Google's ability to turn search into such a big business 0:07:46.320 --> 0:07:48.800 was because it held a monopoly on search, search advertising, 0:07:48.880 --> 0:07:51.800 and pretty much the entire online ads industry. And if 0:07:51.800 --> 0:07:54.760 it was a truly competitive market and Google wasn't allowed 0:07:54.800 --> 0:07:57.880 to be vertically integrated with the entire digital advertising apparatus 0:07:57.920 --> 0:08:00.000 of the Web, it would likely be making much less 0:08:00.040 --> 0:08:02.880 revenue per user. And that's bad if your Google replacement 0:08:02.960 --> 0:08:06.200 costs many, many, many, many many more times more money 0:08:06.240 --> 0:08:08.880 than Google to run as an a side. By the way, 0:08:08.880 --> 0:08:11.520 if you're wondering. No, open Ai cannot just create a 0:08:11.560 --> 0:08:15.400 Google Search competitor. Search GPT will be a significantly more 0:08:15.440 --> 0:08:17.960 expensive product to run at Google scale than Chat GPT, 0:08:18.160 --> 0:08:21.240 both infrastructurally and then the cost of revenue itself, with 0:08:21.360 --> 0:08:24.240 open ai being forced to create this massive advertising arm 0:08:24.240 --> 0:08:27.600 that does not exist. And by the way, another great 0:08:27.640 --> 0:08:29.680 critique I get is, like open Air, I can just 0:08:29.800 --> 0:08:32.440 hire a bunch of ads people. Why do you think 0:08:32.520 --> 0:08:36.800 there hasn't been another competitor to Search? You realize that 0:08:36.840 --> 0:08:38.839 there are lots of other companies with lots of other 0:08:38.920 --> 0:08:41.880 money that could just do this, and indeed they've tried. 0:08:42.000 --> 0:08:44.839 Why do you think that hasn't worked. Is it because 0:08:44.880 --> 0:08:47.320 of the monopoly. It's because of the monopoly. It's because 0:08:47.360 --> 0:08:48.000 of the monopoly. 0:08:48.800 --> 0:08:51.040 Look, people love the chat GPT in todays. 0:08:51.120 --> 0:08:52.559 It's the box where you can type in one thing, 0:08:52.559 --> 0:08:54.800 get another thing out because it resembles how everybody has 0:08:54.840 --> 0:08:56.400 always wanted Google Search to work. 0:08:56.840 --> 0:08:57.680 Does it work well? 0:08:57.720 --> 0:08:59.959 Who knows, but people feel like they're getting more out. 0:09:10.679 --> 0:09:12.880 But before I go any further, I'd like to talk 0:09:12.920 --> 0:09:18.760 to you about HEGI Artificial general Intelligence art official bloody 0:09:18.840 --> 0:09:21.800 general intelligence, the conscious computer that does not exist and 0:09:21.920 --> 0:09:24.719 is entirely fictional. And this two part of by the Way, 0:09:24.720 --> 0:09:28.480 has been a break from my usual onerous, tortured, punished 0:09:28.679 --> 0:09:31.800 and analysis. And it's because I needed to have a 0:09:31.840 --> 0:09:33.200 little fun. And I think you've kind of heard that 0:09:33.200 --> 0:09:35.679 in the episode. I'm like a little flexible. I'm more 0:09:36.120 --> 0:09:38.800 all like limber and excited, and it also comes from 0:09:38.800 --> 0:09:42.319 a place of frustration. None of this AI stuff has 0:09:42.320 --> 0:09:45.080 ever felt substantive or real because the actual things that 0:09:45.120 --> 0:09:47.400 you can do with Generator IVAI never seemed to come 0:09:47.440 --> 0:09:49.360 close to the things that people are Sam Ortman and 0:09:49.360 --> 0:09:51.760 Warrio Aama Day seem to be promising, nor do they 0:09:51.760 --> 0:09:54.559 come close to the bullshit like people Casey Newton and 0:09:54.640 --> 0:09:58.679 Kevin Ruster peddling. None of this ever resembled artificial general intelligence, 0:09:58.679 --> 0:10:00.760 and if I'm honest, very little it seems to even 0:10:00.800 --> 0:10:04.319 suggest it's a functional industry. And when cynical plants like 0:10:04.400 --> 0:10:07.560 Kevin rus of the Times bumble around asking theoretical questions 0:10:07.600 --> 0:10:10.000 such as, do you think that there's a fifty percent 0:10:10.120 --> 0:10:12.560 chance or higher that AGI to find this an AI 0:10:12.640 --> 0:10:15.800 system that outperforms human experts, Eventually all cognitive tasks will 0:10:15.840 --> 0:10:19.760 be built before twenty thirty, I think when we're reading 0:10:20.040 --> 0:10:22.160 people saying that, and that was Kevin rus at like 0:10:22.200 --> 0:10:25.720 some little AI panel, things surrounded, but just like venture 0:10:25.720 --> 0:10:30.080 capitalists and the senior officials of anthropic, when you've got 0:10:30.120 --> 0:10:32.400 like the lead columnist of the time saying this, we 0:10:32.440 --> 0:10:34.480 should all be terrified. We should be terrified, but not 0:10:34.520 --> 0:10:37.520 of AGI, but that the lead technic columnist that The 0:10:37.520 --> 0:10:41.400 New York Times appears to have an undiagnosed concussion. Rus's logic, 0:10:41.440 --> 0:10:43.400 as with Kate sin Newton's, is based on the idea 0:10:43.440 --> 0:10:46.160 that he's talked to a bunch of people that say, yeah, yeah, 0:10:46.160 --> 0:10:48.680 AGI is right around the corner, rather than any kind 0:10:48.679 --> 0:10:52.840 of like tangible proof for evidence. Just line going up. 0:10:53.240 --> 0:10:53.720 Do you see it? 0:10:53.720 --> 0:10:55.920 The line's going up? Which line is it? Get out 0:10:55.960 --> 0:11:01.079 of my house. Look, I will say I've been a 0:11:01.080 --> 0:11:03.080 little mean on Kevin Ruse. I've been saying that he's 0:11:03.120 --> 0:11:06.920 written the worst shit about artificial intelligence, that he is 0:11:07.000 --> 0:11:10.920 a credulous buffoon or at worse, c cynical plant, that 0:11:11.200 --> 0:11:13.640 Kevin Ruse at The New York Times is regularly pushing 0:11:13.679 --> 0:11:17.559 things like NFTs and crypto and getting basic things wrong, 0:11:17.640 --> 0:11:20.319 like the Helium piece in which Helium lied. 0:11:20.160 --> 0:11:22.200 About their customers. 0:11:23.080 --> 0:11:26.400 I've said that Kevin Ruse regularly says things that don't 0:11:26.440 --> 0:11:28.640 make any sense, and he has indeed written the worst 0:11:28.679 --> 0:11:31.959 tech criticism out there, criticism being the wrong word, take 0:11:32.000 --> 0:11:34.080 analysis out there. And I want to just say that 0:11:34.120 --> 0:11:35.960 I've been wrong. I've been wrong the entire time. What 0:11:36.040 --> 0:11:37.800 I should have said is that he's written the worst 0:11:37.840 --> 0:11:41.439 stuff yet. And I should have said that because he's 0:11:41.480 --> 0:11:44.199 recently published one of the dumbest fucking things I've read 0:11:44.200 --> 0:11:45.520 in my life, and I can't wait to tell you 0:11:45.520 --> 0:11:48.760 about it. His most egregious example of credulousness came in 0:11:48.840 --> 0:11:51.840 late April when he published this thinly veiled puff piece 0:11:52.040 --> 0:11:55.080 about what to do if AI models become conscious in 0:11:55.120 --> 0:11:58.880 the future. Now, this piece, this piece is possibly the 0:11:58.960 --> 0:12:01.439 dumbest tech piece I've read in my life. I really 0:12:01.480 --> 0:12:03.680 mean this. I've read some dog share, I've gone on 0:12:03.720 --> 0:12:06.839 telegram groups for the smallest icos you've seen in your life. 0:12:06.840 --> 0:12:09.319 From then the rutger House speech from a blade runner 0:12:09.360 --> 0:12:11.959 right now, the things I've seen with your eyes different speech. 0:12:12.000 --> 0:12:14.400 I realize this piece from the New York Times is 0:12:14.400 --> 0:12:17.640 called if AI Systems become conscious, should they have rights? 0:12:18.200 --> 0:12:21.480 And the subhead is as artificial intelligence systems become smarter, 0:12:21.559 --> 0:12:23.840 one AI company is trying to figure out what to 0:12:23.840 --> 0:12:26.520 do if they become conscious. I'm going to give you 0:12:26.920 --> 0:12:30.360 exactly one guess what company he's talking about. And here's 0:12:30.360 --> 0:12:35.160 a clue. It's the company Casey Newton's boyfriend WORKSAM. That's right, 0:12:35.200 --> 0:12:39.080 it's Anthropic. Kevin Ruse interviewed two people, both employed by Anthropic, 0:12:39.120 --> 0:12:42.440 with one holding the genuinely hilarious job description of AI 0:12:42.559 --> 0:12:47.040 welfare researcher, who said nakedly insane things like there's only 0:12:47.120 --> 0:12:50.080 a small chance, maybe fifteen percent or so, that Claude 0:12:50.160 --> 0:12:53.400 or another AI system is conscious. And it seems that 0:12:53.480 --> 0:12:55.800 if you find yourself in a situation of bringing some 0:12:55.920 --> 0:12:58.240 new class of being into existence dot dot dart, then 0:12:58.280 --> 0:13:01.000 it seems quite prudent to at least be asking questions 0:13:01.000 --> 0:13:04.720 about whether that system might have its own kinds of experiences. Well, 0:13:04.760 --> 0:13:09.199 I'm experiencing brain damage reading this article out. And what 0:13:09.280 --> 0:13:12.240 makes this article so appalling is that Ruth acknowledges that 0:13:12.280 --> 0:13:14.679 this shit is seen by most level headed people as 0:13:14.760 --> 0:13:18.800 utterly fantastical. He describes the concept of AI consciousness as 0:13:18.800 --> 0:13:21.040 a taboo subject and that many critics will see this 0:13:21.120 --> 0:13:23.680 as crazy talk, but doesn't bother to speak to any 0:13:23.720 --> 0:13:26.559 actual critics. He does, however, speculate on the motives that 0:13:26.640 --> 0:13:29.400 said critics, saying that they might object to an AI 0:13:29.440 --> 0:13:32.800 company's studying consciousness in the first place because it might 0:13:32.840 --> 0:13:35.600 create incentives to train their systems to act more sentient 0:13:35.800 --> 0:13:38.360 than they actually are, Eh, Kevin, Wouldn't it be really 0:13:38.440 --> 0:13:41.040 bad if a generative AI company deliberately did something to 0:13:41.040 --> 0:13:43.840 convince someone that their systems were sentient. You know, someone 0:13:43.880 --> 0:13:47.360 really credulous would accept whatever narrative said AI company was pushing, 0:13:47.520 --> 0:13:49.840 such as that they created a special job just in 0:13:49.880 --> 0:13:53.920 case AI was sentient. Genuine question, Kevin, when you walk 0:13:53.960 --> 0:13:55.960 past the mirror, do you start barking at it because 0:13:55.960 --> 0:14:00.240 you see another guy? Anyway, nothing about anything that open 0:14:00.280 --> 0:14:02.640 AI or anthropic is building or shipping suggests that we 0:14:02.679 --> 0:14:05.920 are anywhere near any kind of autonomous computing. They've used 0:14:05.920 --> 0:14:08.840 the concept of AI safety and now AI welfare as 0:14:08.880 --> 0:14:11.960 a marketing term to convince people that they're expensive, wasteful 0:14:12.000 --> 0:14:15.440 software will somehow become conscious because they're having discussions about 0:14:15.440 --> 0:14:18.120 what to do if it does so, and anyone literally 0:14:18.160 --> 0:14:20.360 any report are accepting this that face Value is doing. 0:14:20.440 --> 0:14:23.840 Their readers are disservice and embarrassing themselves in the process. 0:14:24.400 --> 0:14:27.240 If AI safety advocates cared about I don't know safety 0:14:27.320 --> 0:14:29.800 or AI, they'd have cared about the environmental impact or 0:14:29.800 --> 0:14:32.480 the fact that these models train using stolen material, or 0:14:32.480 --> 0:14:36.120 the fact that these models don't really deliver on their promises. 0:14:36.240 --> 0:14:39.160 And if they did, it would shock the labor market 0:14:39.480 --> 0:14:41.560 and it would hurt millions, if not billions of people. 0:14:41.600 --> 0:14:43.760 And we don't have anywhere near the social safety network 0:14:43.800 --> 0:14:45.760 to support the ones we have right now, let alone 0:14:45.800 --> 0:14:49.480 in this completely fictional future. These companies do not care 0:14:49.520 --> 0:14:51.960 about your safety, and they don't have any way to 0:14:52.000 --> 0:14:54.480 get to AGI. They're full of shit, and it's time 0:14:54.520 --> 0:14:56.400 to stop being honest that you don't have any proof 0:14:56.400 --> 0:14:59.880 that they do anything that they say they will. Also, 0:15:00.080 --> 0:15:03.560 by the way, I think the bubble might actually be bursting. Yeah, 0:15:03.880 --> 0:15:06.000 remember in August of last year I was talking about 0:15:06.040 --> 0:15:08.680 the pale horses of the AI apocalypse. Well, one of 0:15:08.680 --> 0:15:11.760 the major warning signs that the bubble was bursting was 0:15:11.840 --> 0:15:14.680 big tech firms would be reducing their capital expenditures and 0:15:14.720 --> 0:15:16.880 This is a call I've made quite a few times, 0:15:16.880 --> 0:15:19.480 and there was one in alarming Clarity from April fourth, 0:15:19.520 --> 0:15:22.080 twenty twenty four. And I'm going to quote myself here. Well, 0:15:22.120 --> 0:15:24.720 I hope no, that's the wrong voice. Well, I hope 0:15:24.760 --> 0:15:26.720 I'm wrong. The calamity I fear is one where the 0:15:26.760 --> 0:15:29.200 massive overinvestment in data centers is met with a lack 0:15:29.240 --> 0:15:31.800 of meaningful growth or profit, leading to the markets turning 0:15:31.840 --> 0:15:33.840 on the major cloud players that state their future and 0:15:33.920 --> 0:15:38.640 unproven GENERATIVEAI. If businesses don't adopt AI at scale, not experimentally, 0:15:38.640 --> 0:15:40.680 but at the core of their operations, the revenue is 0:15:40.720 --> 0:15:42.760 simply not there to sustain the hype, and once the 0:15:42.760 --> 0:15:45.480 market turns, it will turn hard, demanding efficiency and cutbacks 0:15:45.520 --> 0:15:48.840 that will lead to tens of thousands of job carts. Well, 0:15:48.880 --> 0:15:51.080 we're about to find out if I'm right. On April 0:15:51.080 --> 0:15:53.800 twenty first year, who Finance reported that analyst Josh Beck 0:15:54.040 --> 0:15:57.240 said that Amazon's generative AI revenue for Amazon Web services 0:15:57.280 --> 0:16:02.160 would be five billion dollars five billion dollars this year, 0:16:02.840 --> 0:16:05.880 a remarkably small sum that is not profit, by the way, 0:16:05.920 --> 0:16:08.400 and also are dropping the bucket compared to Amazon's projected 0:16:08.400 --> 0:16:10.840 one hundred and five billion dollars in capital expenditures in 0:16:10.880 --> 0:16:13.600 twenty twenty five, it's seventy eight point two billion dollars 0:16:13.640 --> 0:16:16.080 in expenditures in twenty twenty four, or it's forty eight 0:16:16.080 --> 0:16:17.880 point four billion dollars in twenty twenty three. 0:16:17.880 --> 0:16:20.360 And is that really fucking Are you fucking kidding me? 0:16:21.360 --> 0:16:24.360 Are you fucking kidding me? I have had Jack Coles 0:16:24.360 --> 0:16:26.520 for years telling me this was the next big money driver. 0:16:26.720 --> 0:16:28.560 Five billion dollars, five billion. 0:16:28.600 --> 0:16:29.200 Goddamn dog? 0:16:29.280 --> 0:16:31.760 Are you fucking kid? You'd make more money auctioning dogs. 0:16:31.800 --> 0:16:34.040 This is a disgrace. And if you're wondering, yes, all 0:16:34.080 --> 0:16:37.560 of this is for AI, all of that capex from 0:16:37.600 --> 0:16:40.760 the Yahoo Finance article, the one for Laura Bratton wrote, well, 0:16:40.800 --> 0:16:42.480 this is a quote, by the way, from the article. 0:16:42.720 --> 0:16:45.240 CEO Andy Jassey said in February that the vast majority 0:16:45.240 --> 0:16:47.640 of it's this year's one hundred billion dollars in capital 0:16:47.960 --> 0:16:50.240 investments from the tech giant will go towards building out 0:16:50.280 --> 0:16:55.680 artificial intelligence capacity for its cloud segment Amazon Web Services AWS. Well, shit, 0:16:55.760 --> 0:16:57.680 I bet investors are going to love this. Better save 0:16:57.760 --> 0:16:59.280 some money, Andy, what's that? 0:17:00.400 --> 0:17:01.040 What's that? Handy? 0:17:01.920 --> 0:17:06.400 What Oh, you're saving money already? How oh shit? 0:17:06.600 --> 0:17:07.600 Oh uh oh. 0:17:07.720 --> 0:17:09.960 A report from Wells Fargo analysts published at the end 0:17:10.000 --> 0:17:13.240 of April, called Data Centers Aws goes on Pause says 0:17:13.280 --> 0:17:15.720 that Amazon has paused a portion of its leasing discussions 0:17:15.720 --> 0:17:17.960 on the co location side, and while it's not clear 0:17:18.000 --> 0:17:20.119 the magnitude of the pause, the positioning is similar to 0:17:20.160 --> 0:17:23.480 what analysts have heard recently from Microsoft that they are 0:17:23.520 --> 0:17:26.840 digesting aggressive recent lease ubdeals, pulling back from a pipeline 0:17:26.840 --> 0:17:30.359 of loys or sqq's. Now, some asshole is going to 0:17:30.400 --> 0:17:32.600 say lwys and sqqs aren't a big deal, But they are. 0:17:33.160 --> 0:17:36.040 They are. Go listen to the Power Cut episodes. Go listen, 0:17:36.080 --> 0:17:38.040 Go back and listen to them right now, Get off 0:17:38.080 --> 0:17:39.639 my how'd you get in my house? 0:17:39.960 --> 0:17:40.360 Anyway? 0:17:40.840 --> 0:17:43.480 Digesting in this case refers to when hyperscalers sit with 0:17:43.480 --> 0:17:45.560 their current capacity for a minute, and Wells Fargo adds 0:17:45.600 --> 0:17:47.960 that these periods typically last six to twelve months, that 0:17:48.040 --> 0:17:50.560 they can be much shorter. It's not obvious how much 0:17:50.560 --> 0:17:53.000 capacity Amazon is walking away from, but they're walking away 0:17:53.040 --> 0:18:10.560 from capacity. It's happening. But what if it wasn't just Amazon. 0:18:11.640 --> 0:18:13.760 Another report from a friend of the podcast read people 0:18:13.760 --> 0:18:16.560 I email, occasionally asking for a PDF, of course, referring 0:18:16.560 --> 0:18:19.240 to analyst td CO and put our report last week that, 0:18:19.359 --> 0:18:21.240 while titled in a way that suggested there wasn't a 0:18:21.240 --> 0:18:23.320 pullback at all, actually said that there was. It was 0:18:23.320 --> 0:18:25.000 called like hyperscaled demand is up. 0:18:25.280 --> 0:18:25.880 Let's take a. 0:18:25.840 --> 0:18:29.600 Look at one damning quote a relative to the hyperscalar 0:18:29.680 --> 0:18:32.480 demand backdrop at PCC, which is a conference hyperscal of 0:18:32.520 --> 0:18:35.120 demand has moderated a bit, driven by the Microsoft pullback 0:18:35.160 --> 0:18:39.080 and to a lesser extent Amazon discussed later, particularly in Europe, 0:18:39.119 --> 0:18:41.240 there has been a broader moderation in the urgency and 0:18:41.320 --> 0:18:44.320 speed with which the hyperscalers are looking to take down capacity, 0:18:44.680 --> 0:18:47.520 and the number of large deals a four hundred megwat 0:18:47.520 --> 0:18:50.440 deals in the market appears to have moderated. In plain English, 0:18:50.520 --> 0:18:52.720 this means demand has come down, there's less urgency to 0:18:52.720 --> 0:18:56.000 build stuff, and the market is also slowing down. Cohen 0:18:56.040 --> 0:18:59.080 also added that it observed a moderation in the exuberance 0:18:59.280 --> 0:19:02.480 around the outlook for hyperscale at demand, which characterized the 0:19:02.520 --> 0:19:07.439 market this time last year. Brother, my brother in Christ, 0:19:07.520 --> 0:19:09.000 isn't this meant to be the next big thing? 0:19:09.160 --> 0:19:09.520 We need? 0:19:09.560 --> 0:19:12.800 More exuberants, not less. We're still a Microsoft appears to 0:19:12.840 --> 0:19:15.280 have pulled back even further, with TD Cohen noting that 0:19:15.359 --> 0:19:18.040 there has been a slow in demand, slow down in demand, 0:19:18.200 --> 0:19:20.359 and that it saw very little third party leasing from 0:19:20.400 --> 0:19:23.359 Microsoft this quarter and most damningly in our boat and 0:19:23.400 --> 0:19:25.240 I'll be can't I really want you to take this? 0:19:25.320 --> 0:19:28.560