WEBVTT - Monologue: AI Is Making The Tech Industry Weak 0:00:02.800 --> 0:00:08.040 Zone Media, Hello and welcome to this week's Better Offline Monologue. 0:00:08.080 --> 0:00:19.320 I'm your host ed Zetron. Sorry about no monologue last week. 0:00:19.320 --> 0:00:21.279 I've had a very tough few months and needed a break, 0:00:21.280 --> 0:00:23.680 and honestly I'm still flagging a bit, but there's too 0:00:23.760 --> 0:00:25.919 much shit going on for me to stop talking, and 0:00:25.960 --> 0:00:29.880 I love doing this job too much anyway. Earlier in 0:00:29.920 --> 0:00:31.880 this week, I ran a newsletter that I'll likely be 0:00:31.920 --> 0:00:34.440 turning into a few episodes about how few data centers 0:00:34.479 --> 0:00:37.320 are actually getting built, and found that the vast majority 0:00:37.360 --> 0:00:39.400 of those that broke ground in twenty twenty three and 0:00:39.440 --> 0:00:42.920 twenty twenty four are yet to be completed. My working 0:00:43.040 --> 0:00:46.720 theory is now that barely any capacity is coming online, 0:00:46.760 --> 0:00:49.279 and that's why Anthropic had to buy up all three 0:00:49.360 --> 0:00:53.080 hundred megawats of Elon Musk's Colusus one data center capacity, 0:00:53.320 --> 0:00:55.480 because there are no other options out there other than 0:00:55.520 --> 0:00:58.280 the guy who said that Anthropic was misanthroopic and evil, 0:00:58.400 --> 0:01:01.080 an embarrassing insult when you could just say that Dario 0:01:01.120 --> 0:01:04.480 Amma Day sounds like he's doing Elizabeth Holmes's Steve Jobs voice. 0:01:05.880 --> 0:01:08.119 In any case, I think we're in the good old 0:01:08.120 --> 0:01:11.000 fashioned bullshit phase of the bubble. I know some of 0:01:11.000 --> 0:01:13.200 you will say, but data centers are getting built, But 0:01:13.280 --> 0:01:15.839 what you mean is that data centers are under construction. 0:01:16.319 --> 0:01:20.400 They are not, for the most part, being finished. Across 0:01:20.480 --> 0:01:23.679 every Microsoft data center announced since twenty twenty three, I 0:01:23.880 --> 0:01:28.199 found two Fairwater, Wisconsin and Atlanta, that were anything close 0:01:28.240 --> 0:01:31.399 to complete, and despite Microsoft saying that they're both operational, 0:01:31.440 --> 0:01:34.120 they're actually both stuck on the initial phases. I think 0:01:34.200 --> 0:01:37.520 there's some capacity online, but even then I'm not really sure. 0:01:37.560 --> 0:01:40.200 I saw an article in the local newspaper saying that Fairwall, 0:01:40.240 --> 0:01:44.360 Wisconsin wasn't even working. Now. Microsoft has also claim that 0:01:44.400 --> 0:01:46.560 it brought on a gigawatt of capacity in each of 0:01:46.560 --> 0:01:48.920 the last two quarters, and I think it's using a 0:01:48.960 --> 0:01:53.800 different definition of capacity than operational data centers. It's weird. 0:01:54.200 --> 0:01:57.360 It's fucking strange. I'm sorry. I know I meant to 0:01:57.360 --> 0:02:00.400 be articulate and talk about economics and some shape, but 0:02:00.440 --> 0:02:02.880 I'm just I'm kind of tired of it. We're four 0:02:02.960 --> 0:02:05.160 years in and even the largest companies in the world 0:02:05.200 --> 0:02:08.080 still describe their return on investment like the the fucking Riddler. 0:02:08.480 --> 0:02:11.320 How much capacity do you have, Satya? Why don't we know, 0:02:12.000 --> 0:02:14.440 why don't we know how much you actually made on ai? 0:02:14.680 --> 0:02:18.040 Why are you still using run rate? This is strange. 0:02:18.360 --> 0:02:21.680 What's going on? About to fall into Tucker Carlson voice there, 0:02:21.800 --> 0:02:24.680 But nonetheless, for my article, I actually reached out to 0:02:24.760 --> 0:02:27.679 Microsoft to confirm these things, and they said they get 0:02:27.720 --> 0:02:31.200 back to me last Monday, and then they just stopped replying. 0:02:31.840 --> 0:02:36.440 That's generally what people do when they're hiding something, and 0:02:36.480 --> 0:02:38.600 I'm tired of that too. I'm tired of the fact 0:02:38.639 --> 0:02:41.080 that these companies won't just give me a straight answer, 0:02:41.200 --> 0:02:43.800 or you a straight answer, or their investors a straight answer. 0:02:44.960 --> 0:02:49.120 It's frustrating. And I know you're frustrated too, dear listener, 0:02:49.800 --> 0:02:51.959 and I know you keep asking when this shit ends. 0:02:52.040 --> 0:02:54.320 I truly don't know at this point, and I'll explain. 0:02:54.800 --> 0:02:58.600 The obfuscative finances of these firms are so thoroughly mangled 0:02:58.600 --> 0:03:01.239 that it's hard to tell a fanthropic thirty billion dollars 0:03:01.240 --> 0:03:04.840 in the bank or thirty dollars. It's fundamentally impossible to 0:03:04.919 --> 0:03:08.360 understand how much money open ai actually makes, because both 0:03:08.440 --> 0:03:11.720 I and the Information have now published very different numbers. 0:03:12.240 --> 0:03:15.320 Mine are much lower, billions lower. For twenty twenty five. 0:03:15.560 --> 0:03:17.880 I mean I don't have the final quarter, and I 0:03:17.919 --> 0:03:20.519 will say that I can speak to mine with complete certainty. 0:03:20.600 --> 0:03:22.560 Is these are the kind of financials that are borderline 0:03:22.560 --> 0:03:26.840 impossible to lie about, unlike investor documents, which usually well 0:03:27.360 --> 0:03:30.200 are full of them. Now, I want to be clear, 0:03:30.720 --> 0:03:34.520 I am not even accusing open ai of lying. I'm 0:03:34.560 --> 0:03:38.080 sure whatever it's handed over to investors has more asterisks 0:03:38.120 --> 0:03:41.120 on it than Cubert reading mind comp I'm sure it 0:03:41.200 --> 0:03:45.400 says that these are non gap numbers generally accepted accounting principles, 0:03:45.560 --> 0:03:47.520 which means that they could be any number of different 0:03:47.600 --> 0:03:51.720 versions of Ebittenstein's monster. It's hard to tell how much 0:03:51.800 --> 0:03:54.040 open ai is spending, which makes it harder to tell 0:03:54.080 --> 0:03:56.080 how much money it needs or when it might run 0:03:56.080 --> 0:03:59.280 out a bit. If open ai closes this entire one 0:03:59.360 --> 0:04:01.360 hundred and twenty two two billion dollar round in VideA 0:04:01.400 --> 0:04:03.320 and soft Bank are paying in three monthly chunks like 0:04:03.360 --> 0:04:06.920 a fucking Klana plan, and Amazon's thirty five billion dollars 0:04:06.920 --> 0:04:10.480 contingent on either IPO or Agi. Open air will have 0:04:10.600 --> 0:04:12.840 if they successfully do this, would have raised over one 0:04:12.880 --> 0:04:15.119 hundred and sixty billion dollars in the space of a year. 0:04:15.600 --> 0:04:17.480 And even if it doesn't, it will have raised then 0:04:17.520 --> 0:04:20.839 I seem spent most of around seventy billion dollars or 0:04:20.880 --> 0:04:23.560 more in the space of a year. That's so much money. 0:04:23.839 --> 0:04:27.080 Where's it gone? Wherever has it gone? Sammy? Do you 0:04:27.120 --> 0:04:31.039 think they know? Do you think any of these people know? 0:04:31.480 --> 0:04:33.880 Do you think they really know how their finances are doing? 0:04:34.279 --> 0:04:37.840 Because according to open Ai CFO Sarah Fisher, they are 0:04:37.880 --> 0:04:40.599 and I quote the Wall Street Journal here, not ready 0:04:40.640 --> 0:04:43.320 for the scrutiny that comes with being a public company. 0:04:44.000 --> 0:04:46.560 It's pretty worrying to me considering how much money mister 0:04:46.640 --> 0:04:48.440 Ortman is now on the hook for and how much 0:04:48.440 --> 0:04:52.960 money he's taken in. Similarly, his equally damp friend, mister 0:04:53.040 --> 0:04:55.960 Warrio amade of Anthropic is now raising at a scale 0:04:56.040 --> 0:04:58.440 unheard of other than when it just happened with open Ai, 0:04:58.839 --> 0:05:01.040 by which I mean that if it's rounds with Google 0:05:01.040 --> 0:05:03.839 and Amazon fully close, Anthropic will have raised over one 0:05:03.920 --> 0:05:06.719 hundred billion dollars in the space of a year. And 0:05:06.800 --> 0:05:10.880 again it doesn't seem to want to go public. Why not, 0:05:10.960 --> 0:05:15.680 mister Amiday? What do you have to hide? What's going on? 0:05:17.279 --> 0:05:20.039 And now mister Amaday is out raising another fifty billion 0:05:20.080 --> 0:05:22.960 dollars that must close in twenty four hours. No wait, sorry, 0:05:23.120 --> 0:05:25.880 that was a few weeks ago. Now, I mean it's 0:05:25.960 --> 0:05:30.000 thirty billion to fifty billion dollars that might close by 0:05:30.040 --> 0:05:31.520 the end of the month. And yes, that is the 0:05:31.640 --> 0:05:34.480 order of the events. And again, no one, no one 0:05:34.480 --> 0:05:38.320 seems to think this is strange. No one in the 0:05:38.360 --> 0:05:41.000 media seems to seem to think this is strange. No 0:05:41.040 --> 0:05:43.400 one seems to be concerned about this at all. I'm 0:05:43.440 --> 0:05:47.240 actually kind of deeply worried about all of this because 0:05:47.760 --> 0:05:50.760 these companies may go public, and if they go public, 0:05:51.080 --> 0:05:53.400 well they're going to lose everyone's money if they invest. 0:05:53.440 --> 0:05:55.919 But if they don't go public, it's still a bunch 0:05:55.920 --> 0:06:02.880 of money that's gone into them that's just potentially gone nowhere. Anyway, 0:06:03.080 --> 0:06:06.640 this is kind of exciting, I'll admit it's kind of exciting. 0:06:07.160 --> 0:06:09.960 That's a bizarre thing to say. I know, it's strange 0:06:10.360 --> 0:06:14.560 to suggest that this might be an exciting thing because 0:06:14.720 --> 0:06:17.159 the stakes have been raised so much that eventually one 0:06:17.680 --> 0:06:19.440 or more of us is going to be proven correct. 0:06:19.760 --> 0:06:21.680 It's going to be me or the boosters. And if 0:06:21.720 --> 0:06:25.200 the boosters are right here, this is crazy. I don't 0:06:25.200 --> 0:06:27.120 think they're right at all. Gone to my head. I 0:06:27.120 --> 0:06:30.240 think Anthropic is doing some sort of very bizarre accounting, 0:06:30.560 --> 0:06:34.320 likely taking in API payments paid in advance. To explain, 0:06:34.520 --> 0:06:37.080 I think it's possible they are counting an organization paying 0:06:37.080 --> 0:06:39.760 its tokens upfront, say fifty million dollars that's likely to 0:06:39.839 --> 0:06:42.279 last twelve months, and acting is if that's a month 0:06:42.320 --> 0:06:45.440 of actual revenue, versus accounting for it spread out over 0:06:45.480 --> 0:06:48.440 twelve months, artificially inflating the run rate by a factor 0:06:48.480 --> 0:06:51.800 of over one thousand percent. To explain, AR is just 0:06:51.960 --> 0:06:55.800 month times twelve. So if they say, if fifty million 0:06:55.800 --> 0:06:58.600 dollars is meant to be spread over twelve months, counting 0:06:58.600 --> 0:07:00.920 it in a single month and multiplying, that's kind of bullshit. 0:07:01.920 --> 0:07:03.880 I don't know for certain if this is what's happening, 0:07:03.920 --> 0:07:05.800 but pre selling tokens and using it to drive up 0:07:05.800 --> 0:07:08.160 AURR is a way It makes sense. And also it 0:07:08.200 --> 0:07:09.800 is not the first time this has happened. There are 0:07:09.840 --> 0:07:12.160 plenty of startups that have done this bullshit in the past, 0:07:12.600 --> 0:07:15.840 some of them in the dot com bubble. I find 0:07:15.840 --> 0:07:20.120 the alternative a little ridiculous. Anthropic CFO Krishna Rao said, 0:07:20.160 --> 0:07:23.520 on March sixth, twenty twenty six, that Anthropic had made 0:07:23.560 --> 0:07:26.680 five billion dollars in lifetime revenue. He said, exceeding, but 0:07:27.600 --> 0:07:29.120 if it was much higher than that, he would have 0:07:29.160 --> 0:07:32.320 said a different number. If that's the case, Anthropics suddenly 0:07:32.320 --> 0:07:34.480 grew its business were a few hundred million dollars in 0:07:34.560 --> 0:07:37.840 January to a few billion dollars a month in April. 0:07:38.280 --> 0:07:41.600 This is, on an accounting basis, a step change in complexity. 0:07:42.240 --> 0:07:46.040 It's a step change in customers scale. Anthropic must have 0:07:46.080 --> 0:07:47.960 had to hire and scale at a speed that defies 0:07:48.000 --> 0:07:50.480 all logic to keep up. They've been handling so much 0:07:50.480 --> 0:07:55.160 fucking cash in their brand new at this Also, Krishna 0:07:55.240 --> 0:07:57.920 raw the CFO, got interviewed recently on one of the 0:07:57.920 --> 0:07:59.960 many Booster podcasts where they just sit there and let 0:08:00.080 --> 0:08:02.160 you say what you want for an hour, and he 0:08:02.320 --> 0:08:05.680 said that the largest user of tokens in Anthropic is 0:08:05.720 --> 0:08:10.360 their tax person. That's the one person I would never 0:08:10.440 --> 0:08:12.560 let use. In all of them, I would slap their 0:08:12.560 --> 0:08:15.960 hands if I saw them using claud I'd be very worried. 0:08:16.960 --> 0:08:20.200 But Krishna seems very excited about this. He's talking about 0:08:20.200 --> 0:08:24.120 how all of this is speeding up all of their 0:08:25.040 --> 0:08:27.360 all of our accounting. And I must be clear, when 0:08:27.360 --> 0:08:30.680 you're handling billions of dollars very suddenly coming in, you 0:08:30.720 --> 0:08:34.679 don't want to rush the fucking accounting, Krishna. You need 0:08:34.720 --> 0:08:39.080 to calm your fucking jaws a little bit. Fucking just 0:08:40.559 --> 0:08:43.439 it's so ridiculous, and it's ridiculous that I am somehow 0:08:43.559 --> 0:08:45.839 unique in the way I cover this, that I am 0:08:45.920 --> 0:08:48.960 unique in my scrutiny. I don't know why, am I mean? 0:08:49.080 --> 0:08:51.720 I do. It's much easier to not scrutinize these companies 0:08:51.720 --> 0:08:54.240 and just applaud whenever you see a big number. But 0:08:54.360 --> 0:08:57.240 I think it's letting down listeners and readers across the 0:08:57.280 --> 0:08:59.960 board to not approach this and replace of scrutiny, even 0:09:00.000 --> 0:09:01.880 if you think this will work out, even if you 0:09:01.920 --> 0:09:04.160 believe AI will be the biggest most you just thing. Ever, 0:09:04.520 --> 0:09:08.880 I think it's a detriment to your listeners and readers 0:09:08.880 --> 0:09:12.120 to not approach this with a high degree of skepticism 0:09:12.200 --> 0:09:25.960 and scrutiny because this is so much money. And here's 0:09:26.000 --> 0:09:30.480 the thing, maybe Christna's being asked, maybe there are at 0:09:30.559 --> 0:09:34.199 thirty billion, fifty billion, whoever knows there are, And it's 0:09:34.240 --> 0:09:37.560 just this amazing business, or maybe it just isn't that big, 0:09:37.679 --> 0:09:40.640 or the business isn't that good. Maybe that's what it is. 0:09:41.400 --> 0:09:45.640 Do you think do you think that that I would 0:09:45.640 --> 0:09:49.720 be very shocked if Anthropic grew from making a couple 0:09:49.760 --> 0:09:54.120 hundred million dollars in one November last year to four 0:09:54.120 --> 0:09:57.679 billion dollars in April. I'd be really fucking shocked. And 0:09:57.720 --> 0:09:59.920 I think that there is a if that is how. 0:10:00.640 --> 0:10:03.560 It's because of the largest papig ai psychosis in the world, 0:10:03.559 --> 0:10:06.760 which you'll get to in a bit. But I've had 0:10:06.760 --> 0:10:10.520 some people recently say that I am not talking about 0:10:10.520 --> 0:10:13.720 the tech enough, and I want to explain why. And 0:10:13.760 --> 0:10:15.800 it's the fact that four years into this, I'm just 0:10:16.040 --> 0:10:20.760 not sure what it is everybody's excited about. What are 0:10:20.800 --> 0:10:24.880 the big AI applications that they're changing my life? Every 0:10:24.960 --> 0:10:28.520 two days Anthropic releases some sort of new harness product 0:10:28.800 --> 0:10:31.839 for finance, law or medicine or accountancy, leading to someone 0:10:31.920 --> 0:10:35.680 stopped crashing based on the theoretical damage that Anthropics product 0:10:35.679 --> 0:10:38.520 could cause without anybody actually trying it or verifying what 0:10:38.600 --> 0:10:42.240 it does, or even explaining why it fucking maps? What 0:10:42.480 --> 0:10:44.920 is it that I am meant to be excited about. 0:10:45.880 --> 0:10:48.000 What is it that I'm meant to be impressed by? 0:10:49.400 --> 0:10:53.160 But ultimately, the reason that anthropics releasing so much stuff 0:10:53.240 --> 0:10:55.640 is it's flak. It's meant to be an impossible to 0:10:55.679 --> 0:10:58.719 pass blob of different things that create the gestalt of 0:10:58.760 --> 0:11:02.160 an indomitable company that can do anything because the markets 0:11:02.200 --> 0:11:04.920 and the media don't bother check. We hear about everybody 0:11:05.040 --> 0:11:08.160 using this shit day in day out. Why has nothing changed? 0:11:08.600 --> 0:11:11.000 What a business is doing differently? How are they different? 0:11:11.200 --> 0:11:15.520 What is different? What is it that llms are actually doing? 0:11:16.960 --> 0:11:19.760 These should not be difficult to complex questions to answer, 0:11:19.960 --> 0:11:21.600 but nobody seems to be able to give me an 0:11:21.600 --> 0:11:25.360 answer with any real clarity or specificity. Every time they do, 0:11:25.440 --> 0:11:27.880 it's a weird dance involving someone saying I now get 0:11:27.920 --> 0:11:30.640 a summary of some stuff from my open claw, or 0:11:30.720 --> 0:11:33.080 of course some sort of multi agent coding fiasco that 0:11:33.160 --> 0:11:36.040 never actually ends with functional or impressive software or anything 0:11:36.080 --> 0:11:41.520 being shipped faster and look on the coding stuff. I'm sorry, 0:11:41.520 --> 0:11:44.920 I'm just not that impressed. I speak regularly with very 0:11:45.000 --> 0:11:47.880 talented software engineers, and their reaction is it's kind of cool. 0:11:47.920 --> 0:11:50.520 I guess it's not anywhere here as good as people say, 0:11:50.760 --> 0:11:53.560 or This allows bad software engineers to do impressions of 0:11:53.600 --> 0:11:58.640 functional software engineers, which creates material risk to the stability, functionality, 0:11:58.720 --> 0:12:01.840 and security of software because people trust these people who 0:12:01.880 --> 0:12:05.400 cannot do their jobs, which has begun to decay across 0:12:05.440 --> 0:12:08.040 the board across the entire tech industry with the rise 0:12:08.080 --> 0:12:11.959 of AI coding tools. It sounds like LLM coding is 0:12:12.000 --> 0:12:14.320 a very inefficient way to sometimes speed up some things 0:12:14.320 --> 0:12:16.640 in a way that also creates a culture of overwork 0:12:17.120 --> 0:12:20.200 and either constant bullshit detection or something I call the 0:12:20.200 --> 0:12:23.000 engineer's torpor, a state where your senses are dulled and 0:12:23.040 --> 0:12:24.719 you stop giving as much of a shit into a 0:12:24.800 --> 0:12:27.720 ship breaks and you realize your eight concurrent agents are 0:12:27.720 --> 0:12:30.120 burning three hundred dollars a day of tokens and have 0:12:30.160 --> 0:12:32.840 written two and a half million lines of functional, yet 0:12:32.840 --> 0:12:36.320 brittle software designed with no intent or purpose. Like an 0:12:36.320 --> 0:12:38.720 apartment where the rooms vary in size and the number 0:12:38.720 --> 0:12:42.239 of bedrooms and the toilets are different, with varying finishings 0:12:42.280 --> 0:12:45.800 and the plumbings different room to room. I guess people 0:12:45.880 --> 0:12:47.520 could live there, but it would be a paid in 0:12:47.559 --> 0:12:49.800 the ass to run and get worse the more people 0:12:49.800 --> 0:12:53.640 that moved in. The more agency you give a coding model, 0:12:53.679 --> 0:12:56.199 the less you understand what it is you're building, no 0:12:56.320 --> 0:12:58.720 matter how intently you prompted, or how many agents you 0:12:58.760 --> 0:13:00.839 throw it in, and how weird little tweaks you make 0:13:00.880 --> 0:13:03.800 to your clawed MD file. The more that you don't build, 0:13:03.840 --> 0:13:06.880 the less that it makes sense to make. Matters worse. 0:13:06.920 --> 0:13:09.040 The more you hand over, the more rusty you're going 0:13:09.040 --> 0:13:11.960 to be at writing code. The shit isn't good enough, 0:13:11.960 --> 0:13:15.240 it's not getting good enough, and it cannot be trusted. 0:13:15.440 --> 0:13:17.520 And my evidence is that a lot of software is 0:13:17.559 --> 0:13:19.400 just kind of fucking broken all over the shop in 0:13:19.440 --> 0:13:22.439 ways that's so common that we're entirely numb to it. 0:13:23.000 --> 0:13:25.800 We are tolerant of our ship being broken or terribly designed. 0:13:26.000 --> 0:13:29.240 It was a problem before LLMS, and it's got worse since. 0:13:30.080 --> 0:13:33.400 Let me give you some examples. Sometimes my iPhone will 0:13:33.480 --> 0:13:36.600 randomly show a black scream of the pinwheel. It just resets. 0:13:36.720 --> 0:13:39.200 It's fine, it'll even keep playing my music. It's happened 0:13:39.200 --> 0:13:43.439 every iPhone for years. It sometimes it doesn't happen for months. 0:13:43.480 --> 0:13:46.800 Sometimes it happens multiple times a week. Sometimes apps just 0:13:47.000 --> 0:13:49.520 arbitrarily slow down my phone isn't hot or anything. They 0:13:49.640 --> 0:13:52.560 just slow to a crawl on Chrome when I load 0:13:52.640 --> 0:13:55.800 certain websites like tech Crunch. For some reason, my two 0:13:55.920 --> 0:13:58.839 year old MacBook Pro sixteen inch with Ultra M three 0:13:58.960 --> 0:14:01.439 core or what have you. And I said, well, look, actually, 0:14:01.480 --> 0:14:05.800 while we're live on airs. See I have sixty four 0:14:05.800 --> 0:14:09.120 gigabytes of RAM and this shit slows down because some 0:14:09.160 --> 0:14:12.280 ad cookies it doesn't like. Insane. A lot of popular 0:14:12.360 --> 0:14:15.160 video games crash regularly. The Maden franchises looked like total 0:14:15.200 --> 0:14:17.360 shit for years and is getting worse each time. The 0:14:17.440 --> 0:14:21.360 menus are incredibly slow, the Google docs, iPhone and iPad 0:14:21.360 --> 0:14:23.560 apps are so unstable I want to take away their 0:14:23.600 --> 0:14:25.440 car keys, on top of having some of the most 0:14:25.520 --> 0:14:28.480 vexing design decisions across the board, almost as if they 0:14:28.520 --> 0:14:32.800 don't know how to make things right anymore. And let's 0:14:32.840 --> 0:14:36.760 talk about all those organizations burning what ten, twenty, thirty 0:14:36.760 --> 0:14:40.000 percent or more of their headcount on tokens. This is 0:14:40.040 --> 0:14:45.520 a sickness. It is a psychosis. The amounts of money 0:14:45.680 --> 0:14:50.680 being spent are genuinely disgusting. Any company that spends anything 0:14:50.720 --> 0:14:53.480 more than five percent on their headcount on tokens is 0:14:53.520 --> 0:14:57.880 a fucking Mark, Your businesses are not getting better. Your 0:14:57.960 --> 0:15:01.920 CEOs have ai psycho. The more they speak to the 0:15:01.960 --> 0:15:04.440 machine that tells them they're the smartest, sexiest boy in 0:15:04.440 --> 0:15:07.120 the room, the more they'll be convinced that you must 0:15:07.200 --> 0:15:10.840 join them. They do not see this as annihilating money 0:15:10.840 --> 0:15:13.800 for effectively no reason. They see it as praying to 0:15:13.960 --> 0:15:19.480 a soulless, non existent god, a non spiritual god made 0:15:19.560 --> 0:15:24.200 up of more annoying math, requiring the largest amount of 0:15:24.240 --> 0:15:28.200 infrastructure spent on anything ever. It fucking sick. It is 0:15:28.320 --> 0:15:30.640 just fucking sucks. I'm not correcting that, but you'll just 0:15:30.720 --> 0:15:34.880 have to deal with that. Okay, this is what happens 0:15:35.120 --> 0:15:38.520 when every boss is a fucking business idiot who doesn't 0:15:38.560 --> 0:15:41.680 actually do any work. What do you think Mark Zuckerberg 0:15:41.680 --> 0:15:44.200 does all day? But he walks into it, walks up 0:15:44.200 --> 0:15:45.840 to the mirror. Sometimes she looks down at his willy 0:15:45.880 --> 0:15:48.000 and goes, what are you? What do you do? What 0:15:48.080 --> 0:15:51.080 are you for? What is your concept? Or he walks 0:15:51.080 --> 0:15:53.360 into rooms and bumbles in there and he says we 0:15:53.480 --> 0:15:55.800 need to do more on the exponential, and then some 0:15:56.080 --> 0:15:59.240 wretched worm tongue like figure like boss or Adam Mursky 0:15:59.360 --> 0:16:01.960 oinks in Skar walks at one of two thousand different 0:16:02.040 --> 0:16:05.680 Vice presidents to spread his nasty, vague ideas, whatever they 0:16:05.800 --> 0:16:09.520 might actually be. If you're lucky, you might even get 0:16:09.560 --> 0:16:12.640 to deal with one of five different dysfunctional AI divisions 0:16:12.640 --> 0:16:16.000 with different ultraovs that you have to placate as Mark 0:16:16.080 --> 0:16:19.000 Zuckerberg does something with AI that seems to change every 0:16:19.040 --> 0:16:21.560 fucking week. Not to worry the Wall Street Jonal or 0:16:21.640 --> 0:16:24.800 cover it anyway. And I think that this is how 0:16:24.840 --> 0:16:27.200 a lot of Silicon Valley and a lot of businesses 0:16:27.200 --> 0:16:30.120 are being run. I think that's how Jack Dawsey acts 0:16:30.120 --> 0:16:33.120 about It's how Uber's run. Think of an app you're 0:16:33.240 --> 0:16:36.640 using right now that's anything other than okay? Do you 0:16:36.760 --> 0:16:40.400 love any app? Does anything work great for you? I 0:16:40.480 --> 0:16:43.280 like Flighty Flight, He's a good app. That's one. Even 0:16:43.360 --> 0:16:46.480 the Apple Notes app I use constantly doesn't sink right 0:16:46.760 --> 0:16:49.080 at random for no reason and has a bunch of 0:16:49.200 --> 0:16:52.160 random shit come up whenever I choose the menu. Everything's 0:16:52.160 --> 0:16:55.240 a fucking mess. Nothing is smooth or easy. I have 0:16:55.280 --> 0:16:57.600 to log into something every four minutes. I have to 0:16:57.600 --> 0:17:00.880 give somebody a two FA token every six minute. Sometimes 0:17:00.960 --> 0:17:03.000 Google just logs me out of shit for no reason. 0:17:03.440 --> 0:17:07.359 Sometimes my email client loses its access to Outlook or Gmail, 0:17:07.480 --> 0:17:11.520 and sometimes my password system hasn't saved a password and 0:17:11.560 --> 0:17:13.119 I have to reset my password. I have to go 0:17:13.200 --> 0:17:15.240 to my email and there's three fucking emails from a 0:17:15.280 --> 0:17:17.840 pizza place I visited in New Jersey fifteen years ago. 0:17:18.000 --> 0:17:20.639 And by the way, if you fucking email me and 0:17:20.680 --> 0:17:23.320 tell me, get one password, get bit worn, and get 0:17:23.640 --> 0:17:28.119 fuck you fuck off. It's so sickening that everyone thinks 0:17:28.160 --> 0:17:31.879 that there is some panacea. Oh, my app works for me. 0:17:32.119 --> 0:17:34.840 Apps do not work consistently across the board. That is 0:17:34.840 --> 0:17:38.679 a unilateral truth. One password works well for one hundred 0:17:38.680 --> 0:17:41.040 people and doesn't work well for fifty others. I've seen 0:17:41.040 --> 0:17:43.640 it with my own fucking eyes. Same with last password, 0:17:43.760 --> 0:17:46.720 same with bitworn. The state of these fucking companies is 0:17:46.800 --> 0:17:48.679 dog shit. And I know I'm going to get emails 0:17:48.680 --> 0:17:51.320 from people saying Bitwarden works, I pay for bitworn, don't 0:17:51.400 --> 0:17:53.760 like it. Bay fucking works. It's a bit better than 0:17:53.760 --> 0:17:56.640 one password. But it's like, which of these do I trust? Now? 0:17:57.160 --> 0:18:00.840 What's being saved? Where I'm sick of it. I'm absolutely 0:18:00.880 --> 0:18:03.720 sick of it. Apologies for getting heated. I know it's 0:18:03.800 --> 0:18:07.000 atypical of this show for me to rant. Now, I 0:18:07.040 --> 0:18:09.439 want to speak to if any LM coding people. The 0:18:09.520 --> 0:18:11.359 real code jock is that if you use it a 0:18:11.359 --> 0:18:12.679 bit of work. I don't care. But the people who 0:18:12.760 --> 0:18:15.320 think that this is truly the most amazing thing to 0:18:15.440 --> 0:18:20.680 any LLLM coding folks actually use software, it stinks. It's bloated, 0:18:21.320 --> 0:18:24.880 it's counterintuitive, it's slow, it's inefficient. The ten or more 0:18:24.920 --> 0:18:27.400 tracking cookies and one hundred APIs you connect to every 0:18:27.400 --> 0:18:30.639 two seconds, for some reason, have created a sludge that 0:18:30.800 --> 0:18:33.760 is making every single experience on the computer worse, doing 0:18:33.880 --> 0:18:37.240 varying degrees of k fame to impress another dickhead with 0:18:37.320 --> 0:18:40.000 two million dollars in RSUs doesn't make up for the 0:18:40.040 --> 0:18:42.320 fact that the current state of software is fucking sad 0:18:42.320 --> 0:18:45.359 and ugly across the board. Do you think the modern 0:18:45.400 --> 0:18:48.600 state of software is good? And do you think the insecure, 0:18:48.680 --> 0:18:51.960 bug ridden slop fest of openclaw is a good representative 0:18:51.960 --> 0:18:54.840 of your industry's future. You should be burning it to 0:18:54.880 --> 0:18:57.840 the ground. You should be killing open Claw. It's destroying 0:18:57.880 --> 0:19:01.480 the fabric of software. Lllms make shitty code. I don't 0:19:01.480 --> 0:19:04.240 care how it makes you feel. And I think it's 0:19:04.280 --> 0:19:07.480 made a select few people genuinely mentally ill in a 0:19:07.480 --> 0:19:09.360 way that we're yet to account for. And I don't 0:19:09.400 --> 0:19:12.840 mean random people. I mean prominent vcs and CEOs and 0:19:12.920 --> 0:19:15.439 a lot of software engineers were obsessed with this stuff 0:19:15.480 --> 0:19:20.320 who have now been irredeemably proven to be insane gullible marks. 0:19:20.840 --> 0:19:23.600 And we do not talk enough about how insane this 0:19:23.760 --> 0:19:27.800 all is. Eight hundred billion dollars more than that spent 0:19:27.840 --> 0:19:31.360 by Microsoft, Google, Amazon and Meta on AI so far, 0:19:31.640 --> 0:19:34.680 and for what a tech industry that has unilaterally increased 0:19:34.680 --> 0:19:37.240 its expenses across the board to varying degrees for a 0:19:37.240 --> 0:19:40.879 return on investment that's nearly impossible to actually explain, and 0:19:40.960 --> 0:19:43.480 impossible to explain in clear terms like does it make 0:19:43.520 --> 0:19:46.520 more money than ispent? It's not profitable. The state of 0:19:46.520 --> 0:19:49.280 software is arguably worse. Most products try and make you 0:19:49.359 --> 0:19:52.399 use an AI feature with questionable utility or efficacy, and 0:19:52.440 --> 0:19:55.960 some organizations actively ostracize you if you fail to use 0:19:56.000 --> 0:19:59.280 it because their CEOs have AI psychosis because their models 0:19:59.320 --> 0:20:03.600 told them they were brilliant. Every time somebody explains AI's 0:20:03.640 --> 0:20:06.359 economic value, they sound like they just got caught committing 0:20:06.400 --> 0:20:08.080 a crime and they're trying to lie their way out 0:20:08.160 --> 0:20:10.480 of it. Why does open a I need all this money. 0:20:11.119 --> 0:20:14.040 Why does Anthropic need all this money? What justification is 0:20:14.040 --> 0:20:16.960 there for any of this expenditure. I know some people 0:20:16.960 --> 0:20:19.520 want to come up with a conspiratorial reason that all 0:20:19.520 --> 0:20:21.920 of this is. Oh, it's a super secret government military 0:20:21.920 --> 0:20:23.399 thing and Trump's going to bail it out. It's all 0:20:23.400 --> 0:20:25.840 a secret government. It's all built for government, government, government, government, 0:20:25.880 --> 0:20:30.160 It's all a secret Trump conspiracy. No, stow it stop. 0:20:30.560 --> 0:20:33.560 That's not what this is. That is a comfortable way 0:20:33.600 --> 0:20:36.080 to feel smart or in control of a situation you 0:20:36.160 --> 0:20:39.800 have no control over. It is a doom saying philosophy 0:20:40.040 --> 0:20:42.480 that allows you to avoid the hard intellectual work of 0:20:42.560 --> 0:20:45.840 actually thinking of what happens when bad things happen. And 0:20:45.880 --> 0:20:50.200 it also avoids the intellectual exercise of accepting that these 0:20:50.200 --> 0:20:52.119 are a bunch of men who do not really know 0:20:52.160 --> 0:20:55.119 how to do business and are self involved cowards that 0:20:55.320 --> 0:20:58.240 only know how to follow each other. I realize it's 0:20:58.280 --> 0:21:01.920 easier to look at the world as dark, scary and say, wow, 0:21:02.240 --> 0:21:04.600 it's all a big secret plan and this is all 0:21:04.640 --> 0:21:08.119 government spying. Already happening way before LMS. Oh, it's all 0:21:08.119 --> 0:21:10.800 a big data gathering. It's all a big surveillance thing 0:21:10.880 --> 0:21:14.000 already happening way before LMS. LMS aren't even particularly useful 0:21:14.000 --> 0:21:16.240 for it. I'm sure there's some uses, but come on, 0:21:17.119 --> 0:21:19.640 Doing that makes you feel like you're smart, It makes 0:21:19.680 --> 0:21:22.159 you feel like you're ahead of the game, when in reality, 0:21:22.160 --> 0:21:25.040