Dreaming Against the Machine - Episode 10： Better Off Regulated, with Ed Zitron

00:00:00.000 Welcome back to Dreaming Against the Machine.

00:00:04.040 I'm your host, Adam Becker.

00:00:05.940 This week's guest is Ed Zitron,

00:00:08.500 who I'm tempted to say needs no introduction.

00:00:11.120 He is perhaps the most prominent critic of the AI industry.

00:00:15.820 He is the host of the podcast Better Offline,

00:00:18.880 where I've been a guest a couple of times.

00:00:21.240 He is a very outspoken guy

00:00:24.760 and a bit of a curmudgeon.

00:00:29.320 And so on this week's episode, I ask him, because this is Dreaming Against the Machine and not Better Offline, I ask him what he would like to see from the AI industry and from the tech industry, rather than just criticize it.

00:00:46.540 Not that criticism is a bad thing, I engage in it all the time, but we're trying to think

00:00:53.340 about what a good future might look like, and Ed's answer was pretty interesting, and

00:00:58.420 I think you're going to enjoy this conversation.

00:01:00.540 Let's get to it.

00:01:05.740 Ed, welcome to Dreaming Against the Machine.

00:01:08.940 Hello.

00:01:09.760 Good to have you here.

00:01:10.920 Good to be here.

00:01:12.180 I was about to say good to have you here, but I guess it is.

00:01:15.160 Yeah, sure.

00:01:15.760 It is good?

00:01:16.540 no it's good it's good to have the we love adam jesus

00:01:21.340 hit him with the trump impression but five seconds oh god that's that's what we're gonna do

00:01:28.920 here i'll have it i have it again before the end of the time as well i'll hit the musk yeah it's

00:01:33.660 gonna set the tone for the show is that your cat yes the that my cat that randomly will sleep for

00:01:40.240 most of the day but also sometimes join for an entire video which is more more fun than it looks

00:01:47.040 but also far more annoying than it looks too okay hold on before we jump into it then we we got to

00:01:51.800 know like what's your cat's name what's your cat's deal his name's howl howl like howl's moving

00:01:57.280 yes good fantastic and he is a big white and ginger ragdoll and he is a big and he's big

00:02:05.780 beautiful and annoying just like daddy oh he's now trying to drink my diet coke just like me 0.99

00:02:12.660 incredible okay well i'm glad you're here this is going to be interesting because like i've been on

00:02:19.300 your show a couple of times we've talked before we've met but on on your show i feel like we

00:02:24.260 complain about a lot of things and on this show we try to talk about what a good future could look

00:02:29.660 like but that also often ends up involving complaining about a lot of things yeah so i was

00:02:34.900 thinking that today we could talk a little bit about something that we both hate and see if we

00:02:46.860 can turn that into something that we like uh or see if we can turn that into having a conversation

00:02:52.500 about things that we would like to have happen so let's start by talking about this upcoming

00:02:58.520 spacex ipo right because that looks to me like nonsense it does and i think the fact we still

00:03:08.060 haven't seen an s1 is completely insane okay can you explain to me what an s1 is so it's the

00:03:14.200 paperwork you file when you go to go public and usually we see it months in advance we have yet 0.75

00:03:21.780 to see it we've seen various reports from various outlets that i assume are run by cowards who

00:03:28.500 refuse to publish these things but it's very weird because those reports i've seen about it

00:03:34.800 have said yeah the ai stuff is losing billions of dollars and now elon has handed off the data

00:03:40.460 center colossus one to anthropic who he referred to as musk referred to as misanthropic and evil 0.98

00:03:47.900 which is just his insults are so fucking lame or misanthropic and evil god dario amadei sounds 0.96

00:03:56.400 like elizabeth he's doing elizabeth holmes's ceo voice that's a good insult he really does 0.97

00:04:03.280 we're closely exponential it's really weird but now that that deal is happening people are saying

00:04:10.720 oh well musk is going to make a bunch of profit because he's renting out his data center yeah as

00:04:15.720 long as you don't look at how most ai data centers lose money it's great but yeah it's a very bizarre

00:04:21.020 it's a really bizarre company now because it's a social network a rocket company an internet

00:04:27.320 service provider a neo cloud for ai compute so just a company that sells ai compute and also

00:04:33.720 an ai lab and of those businesses i think starlink and spacex i think starlink is profitable ish i

00:04:42.340 don't know if spacex is xai loses so much money um x the everything amp loses tons of money and

00:04:51.660 is like the number one place to get lead poisoning digitally in a functioning regulatory body

00:04:58.800 this would not have been a company that was possible yeah this would not have been a company

00:05:03.800 that could be pulled together that's correct so it's it's bizarre it's just the whole thing's

00:05:09.200 weird isn't even clear when they're going public yeah like anyone heard no i mean the date i keep

00:05:14.340 hearing is june 28th because that's musk's birthday and he's an egomaniac um but whether

00:05:21.200 that's actually when it happens i don't know i mean originally i think it was supposed to be like

00:05:26.040 q1 this year then it got pushed to this it might get pushed the second half of the year i mean it's

00:05:31.160 happening but um but no we don't at the as of this recording which is what mid-may uh i don't

00:05:38.340 believe that there's been a date set it's just very weird because let's see cerebrus went public

00:05:43.160 when we're recording this their s1 was at least several yeah over yeah we got it a month in

00:05:50.240 advance so it's just very weird it's very what the whole thing's very strange i'm sure that musk

00:05:56.680 will use it to make himself richer on paper he'll probably be able to take out a bunch of margin

00:06:01.860 loans on that it's just it's more of the elon con but the thing is if we're talking about

00:06:08.020 happy things spacex is a very bad company like as a as a business they could generally you don't

00:06:14.860 want your business to do everything and if you do you want every part of the business to be at

00:06:19.040 least functional yep the ai part is just it's attaching a fire to your house it is very strange

00:06:25.980 you know i saw this breakdown from i think it was reuters uh trying to figure out what the valuation

00:06:33.260 like what the what the proper valuation of this company would be and you know they said okay we're

00:06:39.000 gonna give there's there's basically three things that spacex does it you know operates starlink

00:06:44.480 it launches payloads into space using the reusable rockets and then there's xai and if we give really

00:06:52.080 really really really generous assumptions about like future growth potential for those three

00:06:59.280 things that it does, we get a number that's no more than half of the proposed, you know,

00:07:07.440 IPO value that Musk has put out there. And so the other half has to be like fantasy Elon Musk

00:07:14.160 bullshit. And then the other reporting on this, and this is just absolutely wild to me, is that 0.99

00:07:21.400 the compensation package that's part of these, you know, pre-IPO documents that's been proposed

00:07:26.820 for musk says that you know he doesn't get any additional shares of the company unless they put

00:07:33.360 a million people on mars which is insane honestly that rocks i kind of love that i kind of love it 0.99

00:07:39.900 like he finally made it he made up an idea that is so stupid that even he can't do it 0.77

00:07:44.780 like i thought it was going to be that if he just if he if spacex makes a million dollars he gets a 0.99

00:07:51.000 trillion dollars i thought it was going to be something far more no no it's it's like it's this

00:07:56.020 And the only other condition is if they put, what was it?

00:07:58.980 I think 100 terawatts of data centers into space.

00:08:05.180 And yeah, for comparison, total worldwide data center capacity as of, I think, 2024.

00:08:12.220 I want to say that was somewhere around 100 gigawatts.

00:08:16.540 I don't even think it's that much.

00:08:17.940 No, it's probably not.

00:08:18.500 Like, I have it on good authority that there is a major hyperscaler

00:08:22.620 that only has one and a half gigawatts of actual capacity.

00:08:25.000 so yeah that's not good uh it's i think i want him to try and put data centers in space just

00:08:32.020 because it'll be such a funny waste of money i don't want him to do it because i don't want him

00:08:38.000 putting even more oh he's not going to do what no but i don't even want him to well i guess it

00:08:43.160 depends on where he fails right because you're right he's not going to do it question is where

00:08:47.620 does he fail as long as the failure doesn't lead to more debris and orbit i suppose i'm fine with

00:08:54.940 him trying that's pretty much for certain yeah i know that's the problem because yeah there's

00:09:01.900 already so many satellites up there and it's it's a problem like this is this is the better future

00:09:07.660 that i want is one where musk just no longer has permits to put unlimited numbers of satellites

00:09:14.360 into space it's completely ridiculous that the future i want is where you can't randomly smash

00:09:21.220 your companies together and you can't do circular financing that's like post ai bubble i am actually

00:09:27.440 probably going to push on a political level for the end of round tripping the end of circular

00:09:32.880 finance i think like spacex bought half a billion dollars worth of cyber trucks that shouldn't wait

00:09:41.140 wait wait wait wait spacex bought half a half a let's see uh spacex cyber truck i thought they

00:09:48.220 bought 500 million of them seriously maybe i'm wrong on the number let me look they bought nearly

00:09:53.240 20 of cyber trucks i mean that's a lot of money sorry it's 1 279 cyber trucks uh likely exceeding

00:10:01.740 100 million in value my bad not half a billion but still that's still a lot of money oh god

00:10:07.480 that should be illegal no more of that or there should be if you buy above a certain amount you

00:10:13.760 have to pay a certain like an insane 50 75 percent tax something that makes it so that if you can

00:10:20.300 find a way where this is necessary fine but you'll pay through the nose for it like it has to be

00:10:26.160 something that defies pretty much it defies any kind of like you would have to have a really good 0.99

00:10:33.720 reason and really be dedicated to it which he is musk is dedicated to nothing he's a a twat yeah 0.82

00:10:40.580 i think is the technical term i think that's right i think that's that's what they call him 0.75

00:10:43.520 in finance it's just really bizarre because i don't know i i thought we had financial regulation

00:10:49.760 yeah i thought that existed and it's just it's never applied to him and you wonder if it could

00:10:55.460 ever stop and i think it could just because musk's empire has been built on debt and built on this

00:11:01.600 kind of extraction of debt from his stock he's able to do these margin loans and everything's

00:11:06.660 fine but he has generally not fucked up his companies to such an extent that it's irredeemable

00:11:12.880 he has built this messy gigawatt scale and that would be if it was actually a gigawatt i think

00:11:19.400 colossus 2 his data center is like 350 megawatts he built this thing as fast as possible which

00:11:24.620 means it's as inefficient as possible which means on top of that it's just he has attached this vast

00:11:32.040 billion or more burning object to his otherwise not quite as burning businesses and there is a

00:11:40.700 level of at some point someone's going to need dollars he doesn't have now they put these gpus

00:11:45.700 in an spv so he could do some legal work there but it's still to a point where this materially

00:11:53.560 threatens spacex this materially threatens tesla there are elements of this i but musk is this 0.95

00:12:00.060 weird reality defining creature so it's hard to say what would do it but he is taking the piss

00:12:06.480 with this tesla with this um data center stuff i go back and forth on whether or not he actually

00:12:13.740 believes this stuff is possible god no you don't think so yeah i mean i don't think he believes in

00:12:18.780 anything or he believes in absolutely everything everything he says he believes in it either means 0.91

00:12:25.340 he's a cretin or uh or a psycho like it's like he is i tend to go with he believes everything he's 0.81

00:12:32.720 saying as he's saying it and then you know and then immediately forget exactly yeah well my whole 0.99

00:12:39.980 thing is at some point he will run out of money despite all the money he has because he relies

00:12:45.080 on that and the one thing that ai bubble is doing is draining the world's debt it's absolutely and

00:12:52.580 if he is levered up with private credit that is not going to go well for him long term the question

00:12:58.800 is when i think it is when rather than if but there is a certain degree of what is the event

00:13:04.700 that triggers it i think it's these data centers i think they cost too much money i think i hope 0.97

00:13:09.240 he's buying more gpus i really hope so because those things are gonna be worth fucking nothing 0.66

00:13:13.020 in a few years okay so normally we do this at the end of the episode but this is a little bit 0.64

00:13:17.340 too perfect of a segue for me not to do it um we have patreon questions or well today one patreon

00:13:24.040 question because you know this is this is still an early podcast and we don't have that many but

00:13:27.680 um the question that we have from patreon is from a subscriber named toronto will and he wants to

00:13:34.940 know what you think an optimistic scenario for how the ai bubble could pop would be and he says

00:13:40.600 optimistic meaning a sooner rather than later and be wiping out chatbots toxic presence from our

00:13:48.240 daily lives as much as possible so what it would have to be is it would have to be open ai or

00:13:54.840 anthropic runs out of money or they can't by which i mean they can't raise more i think that the most

00:14:00.900 realistic trigger i actually have a premium thing i'm meant to be writing after this about this i

00:14:05.640 think the most realistic thing is that data center financing stops we're already getting stories in

00:14:10.500 the ft that are saying that uh banks are worried they're going to choke on ai data center debt

00:14:15.920 data centers have taken 18 to 36 months to build so there's a degree of if these things aren't

00:14:21.320 getting built and the debt's not coming they're going to stop being built possibly entirely

00:14:26.620 most construction projects there's this book called uh how big stuff gets done i think it's

00:14:31.420 called by these two economists and there's a point it's like 99.5 percent of construction projects

00:14:36.720 run over budget um and over time so these are all large construction projects so

00:14:44.940 i my other thing is i don't think many of these data centers are capitalized well enough

00:14:50.560 i don't think they have enough money to start with so we're gonna the slowdown of data center

00:14:56.320 construction will be the first trigger and at some point open ai and anthropic are not going

00:15:00.800 to be able to raise more money they are taking 30 40 50 120 billion dollars every year every time

00:15:06.540 they raise money it's like we are the fucking world but anthropics last funding round they had

00:15:11.420 37 different investors one of them was microsoft and then microsoft didn't mention it in the 0.97

00:15:15.240 quarterly earnings which suggests that microsoft sent no money nevertheless at some point something

00:15:21.500 is going to slow down it's not going to just be something dying it's going to be the money stops

00:15:27.700 coming and the money stops coming first from the data centers so the slowdown happens once the

00:15:35.080 markets ingest that it's a problem that will really fuck with everything and really it's just

00:15:41.320 going to be then i i think adding a time to it is foolish at this point yeah like it's it's difficult

00:15:47.080 to say because and i say this actually in a common monologue because right now it's not clear how much

00:15:54.940 money these companies even have because open air didn't raise 122 billion dollars they got maybe 0.95

00:16:00.160 30 we truly don't know uh we don't know how much these companies make because they all dick around

00:16:05.380 with their revenues i have published lower revenue figures than the information no one seems to want

00:16:10.040 to talk about that so we don't know if they've got 30 billion or three dollars we truly don't

00:16:15.340 so it may be very sudden they go ah shit we've run out of money anthropic raised 30 billion 0.88

00:16:21.340 dollars in february they raised five billion dollars from amazon and 10 billion dollars from 0.97

00:16:25.760 google in uh april late april and now they want to raise another 30 or 50 billion dollars somebody 0.73

00:16:32.000 going dario where'd you fucking put it and i think what's happening is these are poorly run 0.80

00:16:37.640 businesses their cfo was just doing an interview with someone where he was like oh yeah uh we our 0.97

00:16:45.060 largest user of llm tokens in the organization in the entirety of anthropic is our tax person

00:16:51.840 that's a bad sign man seriously yeah that's like finding out that your builder does lsd 0.93

00:16:58.740 i i'm just like what the fuck are you talking that's the one person i do not want touching 0.99

00:17:05.340 yeah yeah our compliance person they do they do a line of coke they crush up some advil and snort 0.98

00:17:13.720 it they hit themselves in the head with a hammer and then they hit the books and it's just so i

00:17:19.060 think that there is some sort of accounting scandal coming um but i just think fundamentally

00:17:24.100 it's going to be a melt it's going to be a slowdown and then a meltdown it's going to be 0.98

00:17:28.920 okay data center funding falls through yeah shit we can't raise the money oracle can't raise the 0.89

00:17:34.560 money open ai is going to need more than 122 billion dollars they need at least another 300 0.96

00:17:40.740 billion dollars i think probably more okay what happens then open ai is apparently slowing down

00:17:45.920 it's already happening right it's just a question of when and we really don't know like it's tough

00:17:51.340 to say because the markets are insane yep it's completely bonkers out there um the debt is still

00:17:57.240 coming but will stop coming at some point and it's not just the debt for big tech it's the debt for

00:18:03.260 every data center you've heard about and once this narrative takes hold it's going to be very

00:18:08.360 difficult to start convincing people to give you more money because they're going to say

00:18:12.120 you sure you're ever going to build this thing do you know where this thing go do you have the 0.99

00:18:16.780 like mr kevin o'leary mr dog shit people call him mr wonderful his natural name is mr dog shit 0.98

00:18:22.200 he's claiming he's doing a nine gigawatt data center in i in utah even and there are people 1.00

00:18:28.900 at the meeting about just screaming at him just being like fuck you kevin o'leary oh we hate this 0.96

00:18:35.320 and he's like no actually it's a good thing everyone loves this it's actually great for 1.00

00:18:39.140 all of you mr dog shit isn't going to build anything but like that is an example of something 0.93

00:18:44.780 going wrong like the this is very clearly speculation and i've heard from several people 0.99

00:18:50.400 that most of these data center deals are land speculation nothing's getting built they're just

00:18:54.540 being like oh we'll sell it to someone else these do there's enough bits and pieces here that it's

00:19:00.160 just the first domino needs to fall and if i'm wrong just to be clear for me to be wrong these

00:19:07.460 things are going to have to get built there's going to have to be revenue that comes out these

00:19:13.040 companies that doesn't exist open ai and anthropic are going to have to become profitable no one has

00:19:17.340 any explanation as to how uh also every ai startup will have to become profitable also uh well um

00:19:24.820 every data center will have to become profitable and have clients for five to six years so they

00:19:29.920 can pay off their debt and at that point they might start making a profit so as long as all

00:19:34.840 that happens i'll be terribly wrong so long story short look at the data center down watch that if

00:19:42.820 that starts falling apart that is very very likely to be the first tale and it needs to be no one

00:19:50.320 gives them the money like it can't just be another person gave them the money like with oracle it has

00:19:55.740 to be we can't raise the cash right okay so once the slowdown happens people will realize that at

00:20:03.920 some point the music will stop which they should have realized already and then once that happens

00:20:09.480 then the bubble pops but the thing is the bubble is not going to be one pop it's going to be

00:20:14.480 probably several years worth of events trickling and trickling and trickling and trickling it's

00:20:20.760 going to be we're going to have months if not years of endless data center collapses

00:20:26.480 it's really it's going to be quite horrible yeah and quite funny well yeah but quite horrible

00:20:33.900 But, yeah, it's also going to be horrible.

00:20:35.660 I mean, this is one of the things that really gets me.

00:20:38.240 Like, I've been working on an afterword for the paperback of More Everything Forever, which is going to be out at the end of the year, available wherever fine books are sold.

00:20:48.180 And one of the things I say in there is, like, you know, this is a bubble, and it would be good if the bubble pops sooner rather than later because that's generally the thing with bubbles.

00:20:59.780 It's always painful, and it's better for it to happen sooner.

00:21:02.660 i mean at this point though even if it i don't know if it popped tomorrow the damage would be

00:21:09.040 huge yeah and i mean the longer it takes the worst right gets exactly the more times that

00:21:15.320 nvidia beats and raises the bigger that pile of potential financial problems grows because

00:21:22.600 nvidia is clearly pre-selling millions of gpus just sitting in warehouses that is going to be

00:21:27.940 an sec investigation one day also the more time that goes on the more capex that gets put into it

00:21:33.720 the more the construction industry gets inflated the more that the chips industry gets inflated

00:21:38.320 because none of this is going to end with the ai industry they want i know that because um try and

00:21:46.140 ask them what they think it is try and ask any of them what are you thinking what what is it that

00:21:51.200 you think where does this stop when do we stop building data centers and if their answer is

00:21:56.220 we never stop it's like okay well limited amount of earth and if they say uh well uh what about

00:22:03.600 orbital data centers you you gas them yeah you you hit them with a cudgel you you yeah i mean

00:22:09.440 throw them in there you put them in a dumpster because they are no longer sensible but yeah the

00:22:13.520 sooner the better the wetter here because if we let this continue more speculative data centers

00:22:20.020 are going to get put into play and a bunch of private credit money is going to flow into this

00:22:24.060 and that money is coming from pensions and insurance funds.

00:22:27.680 And then people are going to lose their shirts.

00:22:29.480 Yeah.

00:22:31.280 Okay, but say that this happens, you know, sooner rather than later.

00:22:35.420 What do we want to be doing after this bubble pops?

00:22:38.980 Okay, we want SEC investigations.

00:22:41.000 We want, you know, an aggressive law enforcement response

00:22:45.220 like the kind that we didn't really see in 2008.

00:22:47.780 We don't expect to see that from this administration,

00:22:50.320 but, you know, maybe there'll be another administration in place

00:22:52.620 shortly after the bubble pops.

00:22:53.900 i don't know we'll see but um you know depending on timing and whatnot in elections but like more

00:23:00.920 broadly what what do we want like what what do you ed think would be a good future to come out

00:23:08.100 of the bubble popping so i think it starts with i think we need sec star regulation of private

00:23:15.020 companies i know people go say it's anti it's anti-business look at what we've done so far

00:23:20.120 I think that if you are a company that's private, that discloses your revenue to the media, you actually have to prove it with tax returns or significant financial documents.

00:23:30.860 And if you misrepresent it, you should be subject to the same punishments that a public company would be if they misrepresented their revenue.

00:23:37.980 That's a real easy way to stop this.

00:23:40.940 I also think that venture capital will get fundamentally destroyed in its current form, which is a great thing.

00:23:48.640 That's really good.

00:23:49.700 yeah because vc the thing i want vc to do again is to stop funding late stage because if you look

00:23:56.380 at the allocation of venture capital where they're actually investing their money they sell themselves

00:24:00.360 as this big industry that's investing in the future no they all double down on each other's

00:24:04.820 ideas the majority of money goes into mid and late stage vc it does not go into seed it doesn't go

00:24:11.180 into precede barely goes into series a that needs to stop in fact i think the regulation that pushes

00:24:18.360 that is the best move because if venture capital was predominantly making early stage bets i'd be

00:24:25.120 super into it i'd be real i would be way more supportive but that's not what they're doing

00:24:30.320 it's just it's not been what they've been doing for a while i also think that the fundament of

00:24:36.860 venture capital needs to change in the i think equity investment is not going to be the right

00:24:42.740 call i think there need to be there needs to be upside and there needs to be profit sharing there

00:24:48.020 needs to be a way in which venture capital is actually invested in building big businesses

00:24:52.680 that are good businesses i also think that perhaps there's a second stream for what they call it deep 0.71

00:24:59.120 tech i call it fucking science but there need to be incentives to investing in actual scientific 0.71

00:25:06.160 endeavors again venture capital it's one thing where i'd actually be pro-government tax subsidies 0.96

00:25:11.820 what have you i there would have to be massive controls around this yes investing in actual

00:25:16.800 scientific things actually like in moderna for companies like moderna early stage and so on

00:25:21.720 there i'd be less worried about later stage things because yeah medicines we need to invest in that

00:25:27.900 actual hardware we need to invest in that semiconductors we actually need to invest in 0.98

00:25:32.880 that some fuckwits recursive learning startup world models i oh i fucking hate fei fei lee 0.96

00:25:41.860 right i'm gonna say something controversial fefe lee clearly done a lot for machine learning 0.99

00:25:46.820 yeah grifter now huh she goes and does the thing where she goes out and goes yeah worlds you know

00:25:53.960 generating worlds on a computer if we could simulate that wouldn't that be great invest

00:25:57.880 in world labs please invest nothing seems to come out from her she has done incredible things within

00:26:03.880 the world she was all image that real alternate i forget which one yeah no she definitely did

00:26:07.600 something i don't remember the details like she she's done like i'm not discounting what she's

00:26:11.600 done. But right now, she keeps popping up talking about world models, like Yann LeCoum. And it's

00:26:17.200 like, wow, you're in an entirely theory. You were sucking up venture capital that could go into

00:26:21.480 something that even kind of exists. And I realize people will argue, oh, well, then nothing will get

00:26:26.820 invested in. Would that be such a bad thing if everything wasn't fucking speculative? I also, 0.91

00:26:34.160 I genuinely think, though, venture capital needs to be stopped or stop investing in anything past

00:26:41.480 the series c stage or at least have allocations where you cannot invest more than x percent of

00:26:48.120 a fund in that like it has to be on a legal level because if you think of it like this

00:26:52.560 i don't like that the tech industry is predominant like innovation is funded by venture capital i

00:26:58.420 think it's a bad incentive structure but i think that if it's going to exist it needs to exist in

00:27:05.140 such a way the actually benefits early stage like because that's that is why we tolerate these people

00:27:12.820 we tolerate them because they're allegedly funding the future they're not well but ed

00:27:18.880 if we change the way that venture capital is regulated and structured then mark andresen

00:27:23.520 might have to introspect and we can't have that all he does every day mark andresen every morning

00:27:28.820 he loses six hours because he walks to the bathroom he looks in the marina goes who the 0.98

00:27:33.300 fuck is that i have my same clothes they're mimicking everything i'm doing he's talking 0.99

00:27:38.960 to claude he's going claude claude who is this man who in my bathroom he looks exactly the same 0.99

00:27:44.820 as me why is he copying everything i do mark andreessen fails the mirror test that's that's

00:27:51.160 what he was admitting who is this mysterious identical man to me every time i go in the

00:27:57.200 bathroom he's there mocking me but that's the thing like andresen andresen horowitz also 0.87

00:28:02.560 something's got to be done about y combinator they're in a weird fucking place because they 0.99

00:28:07.480 are only early stage but you look at the shit coming out of yc and it's like it's chat gpt 0.98

00:28:12.240 generated it's just like the same company 11 times and i don't know i yc should be smaller 0.94

00:28:19.520 but i quite like there was a problem where when i first came to the valley several years into yc

00:28:25.240 yc was too exclusive but now yc is like a like a convention of people all making sass startups

00:28:33.260 i don't know it's just i don't think venture capital is providing the value it claims it is

00:28:40.300 i think venture capital well also venture capital's business model is falling apart

00:28:43.900 but as far as the world i want to see i want to see all vc going into early stage that's all there

00:28:48.660 should be i i really don't think i perhaps there is a you can't invest in the same company more

00:28:54.440 than twice just real real shit like unless there is like a this company will die situation but

00:29:02.420 you're probably not investing in that one that's not how venture rolls right and i think that this 0.76

00:29:06.800 would help change how silicon valley works i think it would fundamentally improve the startup ecosystem

00:29:14.040 because it would be fundamentally less about mates rates and doubling down on friends things

00:29:20.660 it would be about actually i don't know crazy idea going out and talking to early stage founders

00:29:26.840 and investing on them based on what their ideas are i just just you know i i i don't know that's

00:29:33.580 just an idea i had from the beginning of silicon valley i think i yeah that's one thing that needs

00:29:38.960 to change i think you have more faith in venture capitalists ability to critically examine startup

00:29:46.680 founders ideas than i do well i agree i'm saying that when you make it so the incentive structure

00:29:53.520 rewards doing that they will be forced to huh i'm not saying that they're going to do so

00:30:00.280 i'm saying that if you force them to if you make it so that this is the only way they can do this

00:30:07.560 they will either be really bad at it or they will learn to get good at it the reason that they keep

00:30:13.520 doubling down and tripling down is because they don't know how businesses work because they are

00:30:18.140 not incentivized to they are incentivized to double down on what's what keys are being jingled

00:30:23.940 that week or who looks and sounds like them and perhaps that problem will be there at the early

00:30:28.920 stage i don't discount that but when you take away the incentive of doubling and tripling down on

00:30:34.140 everyone's same share yeah you might actually start seeing companies that are well built because

00:30:41.780 late stage can't rely on venture capital and people will end up saying oh but then we won't 0.89

00:30:47.460 get companies like uber oh no oh i'll get companies like open ai fuck no no don't do that to me it's

00:30:57.460 like well then open ai wouldn't be able to grow that big maybe that's a good thing yeah maybe it

00:31:02.880 would change it so that businesses at the series b stage were like oh right we need to make more

00:31:08.740 money. We need to make more money than we spend. And then they'd say, yeah, this business line is

00:31:14.720 not viable. Now people will say, well, this just pushes the money into large companies,

00:31:21.340 the hyperscalers. Like that's what they'll say. That's it. At which point I say, you make it

00:31:27.560 illegal for hyperscalers to invest more than a certain amount in any given entity. And that

00:31:32.220 includes side entities, SPVs, and so on and so forth. I think you just, you make it so that

00:31:37.320 hyper you can't invest more than a billion dollars in the company like just over the course of their

00:31:42.760 entire life you and that includes building infrastructure for them open ai microsoft said

00:31:48.900 that they've spent over a hundred billion dollars in open ai between their investments and their

00:31:52.600 infrastructure jesus i think that's too much money i agree that is too much money and and people will

00:31:58.900 say well well then who would build the vast ecosystem of gpus today we shouldn't have built

00:32:04.180 that yeah shouldn't do it it's a bad idea it was never a good idea well then we wouldn't have got

00:32:09.560 llms we already had them the llms existed before we didn't melanox was acquired in 2019 the attention 0.55

00:32:16.800 is all you need paper is from 2017 like well what are we fucking doing here people will come up with

00:32:22.940 lots of little well this will happen which will be bad you gotta start somewhere and it isn't 0.85

00:32:28.780 Like, where we are today is insufficient.

00:32:32.140 Yeah.

00:32:33.620 Unfortunately, we still live in capitalism,

00:32:36.040 and that means we need to talk about our partners and sponsors.

00:32:38.860 But fortunately, our partners and sponsors are great.

00:32:41.840 And if you want to support Dreaming Against the Machine directly,

00:32:44.920 you can join us on Patreon,

00:32:46.560 where $5 a month gives you the chance to ask questions of upcoming guests

00:32:50.520 and a dedicated stream of cat pictures.

00:32:53.720 We love our patrons,

00:32:54.880 And we also love being a part of the Multitude Collective at MultitudeShows.com.

00:32:59.720 Multitude is a podcast company made up of passionate people creating shows you can count

00:33:03.480 on.

00:33:04.260 Shows like This Guy Sucked.

00:33:06.660 This Guy Sucked is a history podcast for haters by haters. 0.96

00:33:10.780 Join historian Dr. Claire Aubin and a new expert every week to pull back the scholarly

00:33:14.940 curtain on some of the world's biggest bummers.

00:33:17.520 From Charlemagne to Carl Schmitt, every episode gives listeners all the ammo they need to

00:33:22.180 win dinner table arguments over why history's main characters were actually kind of the worst.

00:33:28.220 No dead person is safe, and the show's guests prove that the best part of understanding the

00:33:32.840 past is criticizing it. New episodes every Thursday, wherever you listen to podcasts.

00:33:38.580 And here's a word from our sponsors.

00:33:40.900 One of this week's sponsors is Shaker and Spoon, a monthly cocktail and mocktail subscription.

00:33:46.040 Full disclosure, they sent me a free box this month, and I'm excited to try it on a night off

00:33:50.220 soon. Each box includes three original recipes from reward-winning bartenders with enough

00:33:55.740 ingredients to make four of each of those drinks for at least 12 drinks in total. And each box

00:34:01.100 includes everything you need to make the drink except for the alcohol. And they're built around

00:34:05.620 one spirit per box, so you really just need one bottle of something that you probably already

00:34:10.180 have on hand if you drink. And if you don't drink, every box includes zero proof options for every

00:34:15.940 recipe, and some of the drinks are designed to be completely spirit-free from the start.

00:34:20.860 The boxes come with recipe cards that include guidance on mixing techniques and glossary of

00:34:26.180 terms that teach you how to shake, stir, build, and garnish these unusual cocktails with eclectic

00:34:31.860 ingredients. And they also include everything else that you need to try out these drinks,

00:34:36.040 like custom syrups, bitters, extracts, garnishes, and aromatics. And so you can try the drinks

00:34:42.680 without going off and buying full bottles of niche ingredients that you'll never use again.

00:34:47.080 It's a great idea for date nights or hosting a party or bringing a gift to someone else's party.

00:34:53.760 And you can always skip or swap boxes or even just buy individual kits or ingredients

00:34:59.280 without subscribing. So you can try this stuff out at their website, shakerandspoon.com.

00:35:05.220 Use code DATM for 10% off your purchase.

00:35:09.300 there's a cultural problem in the vc world as well like and maybe maybe it's coming out of

00:35:16.240 these incentives maybe maybe that's right i went to this conference maybe about a year ago um where

00:35:23.740 the details are not important and i don't want to like name and shame people um and i also don't

00:35:29.660 even remember the name of this particular vc who was talking about this but it was a group of vcs

00:35:34.640 up on a stage talking about this industry

00:35:36.920 that they'd invested in that was

00:35:38.320 it was like a part of the 0.99

00:35:40.560 energy industry that I thought was bullshit 0.94

00:35:42.640 and I was writing an article about it 0.95

00:35:44.420 actually this I can say it's an article

00:35:46.820 that I ended up writing for Fortune magazine

00:35:48.780 about nuclear

00:35:50.760 fusion startups which are three

00:35:52.660 words that do not belong together

00:35:54.320 yeah no there should not be

00:35:56.420 helium baby

00:35:57.480 it's not a thing but

00:35:59.860 well I mean it is a thing it's just you know

00:36:02.340 there's no good business case for it as far as i can tell um and and to my shock and uh delight

00:36:09.040 fortune was very willing to have me write about that for them which is great and so i was doing

00:36:13.900 research research for this thing and i was at this conference and i saw some vcs up on stage

00:36:18.400 talking about you know the investments that they had made in the nuclear fusion sector uh and

00:36:26.020 and one of them there was this striking moment it was almost like well i'll tell you what he said

00:36:35.000 and i'll tell you what my reaction to it was he said yeah you know uh uh there's this joke that i

00:36:40.680 like to tell that you should never listen to a venture capitalist when they talk to you about

00:36:46.560 any company that they invested in because they're just going to try to you know convince you to

00:36:50.560 invest in it whether or not it's a good idea and i i was like wow it's almost like i watched this

00:36:55.540 guy's conscience like you know long atrophied suppressed conscience temporarily actually

00:37:03.160 jailbreak and like escape out of his mouth into the world for a moment and then get reeled back

00:37:08.800 in it's the garfield me yeah wait with garfield on the tv station where he's like huh wonder who

00:37:14.900 that's about and it's a picture of garfield with a big red yeah yeah yeah exactly yeah i just like

00:37:21.500 in what sense was that a joke dude yeah it's not really a joke it's a it's a truth yeah but here's

00:37:27.300 the thing as well we have credited venture capitalists with more intelligence than they have

00:37:31.380 yes yeah exactly i have talked to most of the leading venture capital firms over my life and

00:37:38.120 i have fundamentally found that these people are just rich they just they just got money that's

00:37:43.600 all they got yeah they don't have nothing special going on up there that's for sure there's nothing

00:37:47.900 there's no grand secret there is no grand secret and i mean that's with the ai bubble as well i

00:37:53.160 saw that weird freak uh the professor ethan mollett oh yeah on msnbc the other day he would

00:37:59.400 drink sam maltman's bath water he's a freak and he was saying well if you talk to the people at

00:38:06.080 the labs they say whoa ag is coming it's like you're being lied to ethan ethan and you are 1.00

00:38:12.760 lying by extension you're a fucking liar he blocked me on twitter by the way and it's one 0.99

00:38:17.720 of the few cases where i'm like no that's like i i i've made fun of you so many times i've said 1.00

00:38:23.980 you would drink sandwich i was about to say you you just uh you just said like that he was going

00:38:27.740 to reenact salt burn but with sam altman like i yeah that's a movie that nobody wants to see 0.96

00:38:32.500 altburn no altburn oh shit and now this is gonna go on social oh no but here's the thing 0.88

00:38:40.840 if we're thinking about good things that could come out of this i actually think if venture 0.97

00:38:45.140 capital was forced yeah forced to focus on early stage there would be difficulties i'm not saying

00:38:51.800 it's a perfect plan there would have to be ways we deal with hyperscaler investments but

00:38:55.480 young founders are in my experience far more yeah yeah there are cynics everywhere there's always

00:39:02.860 cynics but if all of the money was going to the earliest stage people it would at the very least

00:39:08.500 incentivize people to build more new things rather than iterations of old things which is what we've

00:39:14.800 got right now because everyone they invest in the company they hope it gets the series c it gets big

00:39:19.340 and fat then they hope it gets acquired or it gets a certain amount of arr so at some point it will

00:39:24.960 be sold except that stopped happening in 2018 for like barely any ipos if we started in and there

00:39:32.940 are going to still be a lot of failures that's fine that's what happens with venture capital

00:39:37.640 that's meant to be the risk that's meant to be the risk the risk is meant to be you you take a risk

00:39:44.220 on someone. It's a venture. So if you stop, and also, I think that there should be timers on

00:39:50.500 venture capital investment. I think that they should not be able to hold a fund for more than

00:39:55.160 two years. I think that they should be forced to actually deploy capital, maybe even in a year.

00:40:02.120 And they'll say, oh, but we want to wait for the right things. No, you should be actively

00:40:08.100 incentivized to invest in early stage it will be a lot of spray and pray it'll happen but it's

00:40:14.680 better than what we have today it's fucking better i also think we should limit check sizes

00:40:19.900 i think we should i think that there should be real hard limits but the big thing is going back 0.81

00:40:24.900 to what i said earlier i also think private companies should be required to do uh revenue

00:40:29.640 disclosures and cost disclosures i think that if p we need far more transparency in this they're

00:40:35.300 getting too much money and they're being too big a prick about it no that i certainly agree with

00:40:39.700 i mean i'll be perfectly honest a lot of the specific things that you're mentioning are things

00:40:44.720 that i just don't know enough about but they all sound pretty reasonable to me as someone who you

00:40:49.940 know is is not well versed in finance but certainly greater transparency for private companies i mean

00:40:55.940 that just seems like a no-brainer transparency and regulatory demands of honesty because

00:41:02.600 anthropic and open ai are both misleading people about their revenues but they're both doing it

00:41:07.260 and yeah and it's certainly true going back to something that you said uh a minute ago that

00:41:13.820 these venture capitalists are not particularly smart guys they're just rich i mean this is

00:41:18.520 this is going to be one of the running themes on this show is dunking on mark andreason

00:41:23.400 i mean not only did he say that he doesn't introspect but one of my favorite things that

00:41:28.760 he said is that one of the jobs that ai will not be able to replace is venture capital yeah that

00:41:37.260 you know he's a very special boy and his job is one of the most intellectually demanding jobs it's

00:41:43.520 like uh i don't know buddy based on hearing his intellect i don't think that there is

00:41:48.760 intellectual demands of him other than stop talking right and and also given that he doesn't

00:41:54.080 introspect well you know what else doesn't introspect it's an llm he's just walking down

00:42:01.480 the street he sees a passes by a window he starts barking yes there's another guy in there there's

00:42:07.480 another venture capitalist um it's really sad as well because there are i've met a few decent vcs

00:42:14.480 they're not making they're wealthy whatever like it like there should be better wealth distribution

00:42:20.040 there should be tax on this stuff like i think that regular people deserve a hundred times better

00:42:25.580 quality of life i think a utopia is one where everyone thrives but there are venture capitalists

00:42:30.360 were like yeah i bet early on some companies that did well and i made money and when you talk to

00:42:34.840 them they're like yeah i invested in them because i thought it was good business and it turns out

00:42:38.760 it was a good business it sold someone i made some money they made some money it didn't make

00:42:43.720 the torment nexus it was great but our system is built currently for grifters yeah and my next big

00:42:52.980 thing my next big recommendation is i think circular financing and i mentioned it earlier

00:42:58.000 should be entirely illegal nvidia has propped up iron nebius core weave and lambda they are both

00:43:07.640 selling them gpus renting back the gpus and back in there and and investing in the company as well

00:43:13.960 that should be illegal i don't think it should be legal at all i also don't think we should i think

00:43:21.580 we should cap so this is going to sound complex so it's pretty simple we should cap counterparty

00:43:26.560 credit so when you sign up for a student loan your parents sign on it that's counterparty credit

00:43:33.340 risk they well that's kind of it's a counterparty that is on the loan sorry counterparty credit

00:43:38.440 risk something else my bad but nevertheless for example google backstopped two loans one to cipher

00:43:45.280 mining which is a crypto company becoming an ai company and the terror which is a crypto company

00:43:51.020 becoming an an ai company uh they backstop them for over a billion dollars each should be illegal

00:43:56.940 shouldn't be possible you shouldn't be able to raise a loan by going hey can we use this guy's

00:44:00.880 credit open ai was able to open ai got a line of credit i assume they use somebody else's to do

00:44:06.180 that the reason that by the way open ai and anthropic haven't built their own infrastructure

00:44:09.640 is they cannot raise debt yeah but the you just can't do it but core weave has been able to raise

00:44:15.280 tons of debt by using the credit of meta and google and uh nvidia and by using these contracts

00:44:21.700 to raise debt that should be illegal or it should be capped up you can raise i don't know 50 million 0.56

00:44:28.980 bugs that's it that's you able to you have to do all the shit in the world other than gpus you

00:44:34.800 wouldn't be able to do gpus and i think that'd be a good thing i think the nvidia ceo jensen

00:44:39.820 huang's non-profit should not be allowed to buy compute from core we yeah i think that's a pretty

00:44:44.780 easy one to stop i think just like all the circular financing like if you are a massive

00:44:48.840 vendor to a company uh you should not be able to become its customer like that's real simple

00:44:55.660 you shouldn't be able to backstop their IPO. And people say, well, they'll just use SPVs.

00:45:00.760 Well, you look at, yes, you keep an eye on the SPVs and you make it clear in the legislation

00:45:05.140 that special purpose vehicles and attempts to use special purpose vehicles to get around this

00:45:09.500 will be considered fraud. But you just make it very harsh. And the counter argument to this would

00:45:15.740 be, well, then we wouldn't have companies like Corweave. And the answer is, sign me up. Let me

00:45:21.120 know when i can hit this up because that sounds just grand to me thank you very much can you

00:45:27.000 explain for our listeners what core weave is because i don't know of a great way to answer

00:45:32.660 that question it's real simple man super simple core weave is a company that buys gpus and puts

00:45:39.900 them in data centers you may be thinking i mean builds data centers wrongo a company called core

00:45:45.360 scientific that is a completely different company builds the data centers lends them to core weave

00:45:50.440 who then puts the gpus in them core weave is effectively a holding company for buying gpus

00:45:57.520 they raise debt to buy gpus they then rent those gpus to nvidia who and then they use the contract

00:46:05.300 from nvidia to raise debt to build more data centers using core scientific to fill with gpus

00:46:10.320 bought from nvidia core weave is just an ai server company it is not a complex business it is a very

00:46:16.980 bad business but it is very complex they call these things neo clouds they call them token

00:46:22.120 factories they call them ai factories these are all deliberate terms used to make them sound

00:46:27.340 complex when they are big old they are companies that buy gpus and put them in other people's

00:46:32.600 buildings huh it's it is it's a terrible deal it's it's an awful they're terrible companies

00:46:39.520 and a lot of them are former crypto miners core weave former crypto miner iron former and current

00:46:47.460 crypto miner so uh terra wolf cipher mining um christ there are like seven others lambda lambda

00:46:55.400 i don't know if they did crypto maybe they did crypto kind of hard to tell it's like these

00:47:00.960 companies are all shrouded in secrecy but yeah they're companies that buy gpus and put them

00:47:04.720 places but like core weave is is that's what core weave ostensibly does but core weave is also like

00:47:10.580 a swirling nexus of circular finance yes here's how core weave raises money core weave signs a

00:47:16.680 big deal with meta for example they then go to bankers and go look we've got a big contract for

00:47:21.880 meta can we borrow four billion dollars or eight billion dollars i mean to buy the stuff so that

00:47:29.340 we can make the money off of meta and the banks go yeah sure okay sounds good to me even though

00:47:37.700 that data center probably won't be built for three years or more if it ever gets built and so that's

00:47:43.400 that's how this whole thing works the circular financing aspect is if you look at it like this

00:47:48.140 the actual majority of ai compute is bought by open ai anthropic microsoft meta google or amazon

00:47:56.360 Here's CoreWeave's customers.

00:47:57.920 Their largest is, of course, Microsoft for OpenAI.

00:48:00.660 Their second largest is NVIDIA.

00:48:02.420 Their third largest is OpenAI.

00:48:04.780 Their next one is Meta.

00:48:06.180 The next one is Google, who is renting capacity to OpenAI.

00:48:09.720 Then it's Anthropic.

00:48:10.980 Pretty good, right?

00:48:12.160 Pretty good business.

00:48:13.960 We love it.

00:48:15.160 It's really bad.

00:48:16.780 It's really, really bad.

00:48:18.180 It's so bad, and it's also so brittle.

00:48:20.780 And CoreWeave has, like, tens of billions of dollars of debt.

00:48:23.360 They have negative margins.

00:48:25.760 they are just a mess of a company and they also constantly to start they need to keep building

00:48:31.360 another idea for a better future is to actually have regulations around capex i think we need to

00:48:39.000 make it so that companies have then i don't know how one would limit this but i don't think we

00:48:45.000 should allow companies to go into four five times their revenue in debt and also run a massive loss

00:48:54.380 like i feel like we are meant to have and regulations around financial vehicles and

00:49:01.320 people can say again well this will stop innovation where who what innovation yeah what who else is

00:49:07.720 doing this there's no other business that does this airlines i guess but airlines at least they

00:49:12.940 move people places and they don't do anywhere near this spend i always thought the dot-com bubble 0.95

00:49:17.680 read your fucking history you are wrong it's a different amount of money it's a different kind

00:49:22.220 of structure if they don't finish these data centers they'll cost just as much in 10 years 0.99

00:49:26.400 to finish as they did today if they still have the gpus they'll still cost just as much to run

00:49:31.240 in 10 years people this is actually the thing that i think a better future would look like

00:49:37.140 if people read if people read stuff if people actually learned why they believe in things

00:49:43.160 i have heard it's just like the dot-com bubble it's just like uber it's going to be a bailout

00:49:48.520 I've heard the same shit a hundred times now, and I have published again and again, 0.97

00:49:54.020 direct proof that it's not like it. And people still say the same thing. It's also inherently 0.99

00:49:59.740 insane that this entire era is justified by the past. I also think we need to vastly pull back

00:50:06.920 how many retail investors can participate in the markets. I don't think regular people should be

00:50:12.100 able to touch options. I don't think anything good comes from that. They will say, oh, well,

00:50:17.980 will keep people out of the financial system. Do you think an average person with the Robin

00:50:22.580 Hood app should be able to take out hundreds of thousands of dollars in options? I don't.

00:50:27.520 I think it's a terrible idea. All of this would require us to actually regulate the tech industry

00:50:33.080 and the financial industry. And that's not going to happen. Which we will not be doing. Yeah,

00:50:37.120 we will not be doing until there is some very different administration and power in this

00:50:42.820 country. So let's hope that happens soon. I hope so. Okay. I can't think of a better place to stop

00:50:48.620 it than that. But Ed, thank you so much for your time. This was great. Thanks for having me. My

00:50:54.240 pleasure. Thanks again to this week's guest, Ed Zitron. Next week, I'll be talking with Corey

00:51:00.800 Doctorow about his new book, The Reverse Centaur's Guide to Life After AI. To submit questions for

00:51:09.480 future guests and to suggest other guests, and to see more pictures of Babka, join the conversation

00:51:15.840 on Patreon. You can also find us on YouTube, on Instagram at DATMPod, on the web, and on Blue Sky

00:51:23.020 at DreamingAgainstTheMachine.com, or just find us wherever you get your podcasts.

00:51:28.660 Dreaming Against the Machine is a proud member of Multitude Productions.

00:51:32.780 Our executive producer is Nick Karisimi. Our associate producer is Rosie Thomas.

00:51:37.340 Our theme music is by Jared Emerson Johnson

00:51:40.420 Our show logo is by Nick James

00:51:43.200 And our fearless leader is Babka

00:51:45.540 The greatest cat in the observable universe

00:51:47.920 I'm Adam Becker and I'll see you next week

Dreaming Against the Machine - June 16, 2026

Episode 10： Better Off Regulated, with Ed Zitron

Episode Stats

Transcript