Let’s Lose Lots of Money feat. Ed Zitron | Chapo Trap House

[00:00:01]Howdy folks, it's Amber.

[00:00:03]You may have heard that a while back I headed out to California to start a production company Cold Feet Films so that we, the good people at Chapo Trap House, could go Hollywood and buy a solid gold boat.

[00:00:16]Well, we started the boat fund and then we spent it all on a movie instead.

[00:00:21]The movie is called "Chunks" and you could get tickets to the worldwide streaming premiere on July 19th at patreon.com/chunkstv.

[00:00:31]It'll be available on demand from there on out.

[00:00:34]"Chunks" is an anthology of What the [ __ ] six films by our most talented friends.

[00:00:47]Chris, Chris, this was supposed to have names. Who the [ __ ] is Nick Mullen?

[00:00:56]The holy original never-before-seen original. This was an IP cash grab. I wanted callbacks to all the Chapo favorites.

[00:01:07]These hogs hate it when we do anything new.

[00:01:12]The real dystopia I told them to make the baseball crank rom-com.

[00:01:19]I was on set. I dressed up as a big baseball and danced around in the back like the sun from the [ __ ] Teletubbies. You think I did that for fun?

[00:01:30]No, no, no, no, no, no. This must be fixed. I am the producer.

[00:01:36]I went bald. I have hair on my forearms like Tom Cruise in Tropic Thunder. I have the big hands. I hired Bill Hader.

[00:01:47]Not his character, the guy. He's available because the Barry finale left viewers disappointed.

[00:01:57][ __ ] >> [sighs] >> I need to think.

[00:02:02]Spitzer La Musica >> Hello friends. It's Monday June 22nd and we've got some Chapo coming at you. On today's show Felix and I are joined by our senior tech and Silicon Valley correspondent Ed Zitron is back again.

[00:02:19]Ed, hello.

[00:02:20]>> Hi, it's a very dark day. RIP Keir Starmer.

[00:02:24]Goodbye England's rose.

[00:02:27]>> You know, I you you beat me to it. We are obviously going to talk about AI, Open AI, Claude, Anthropic, things of that nature and where's it where's all the money going? But first we have to begin today with some rather sobering news. Our boy Keir our Keir announced his resignation from number 10 today. Um and you know, I think headline here a great man, a visionary leader um just a a decent man above all else uh done in by a savage and ungrateful electorate. So like I like I said, we need him now in America more than ever. So hopefully he has some free time to maybe hop across the pond and come to a more receptive electorate that I think will embrace him.

[00:03:11]>> I can name all the things that Keir Starmer has done. I know all of them. He definitely did things. You know, remember that. I was just talking to Felix offline about this. He bravely brought data centers into the UK.

[00:03:25]Beautiful data center with Core Weave out in Lauter. Now, all right, the Guardian woke left to the Guardian. They went out there and they found out there wasn't a data center. There was just a pile of scaffolding.

[00:03:36]Now, I know some people might say and Felix made this point to me is uh yeah, can't build data center without scaffolding, can you mate? Yeah. QED.

[00:03:44]>> Yeah. Can you imagine going to work at a data center and um the Three Stooges are dropping bathtubs and safes on your head. Birds are crapping on you. I just like when Keir Starmer came in there, it was like you thought that [ __ ] with the Knicks winning was like a big, you know, big night.

[00:04:06]>> Yeah.

[00:04:07]>> Every town in America that Every town in the world was like that times 10.

[00:04:11]>> Yeah.

[00:04:12]>> Keir A Knicks start with K.

[00:04:14]>> Huh?

[00:04:15]>> Right? Basically the same thing.

[00:04:17]>> You couldn't walk 10 ft without being engaged in group sex after Keir Starmer's landslide win, Labour Party's landslide win in 2024.

[00:04:27]And it's just like anyone else would have gotten 15 to 30 years to implement their plans.

[00:04:32]>> Yeah.

[00:04:33]>> But I guess it's just, you know, 2 years and you're done. Oh, just because just because everything is horrible now, just because when they asked him, "Why do you want to be Prime Minister?" Um he doesn't have an answer still even after being Prime Minister. It's just kind of like, um look, we really like try not to be nihilistic on this show. I really think it's irresponsible for people with like any type of audience to be nihilistic, but it is if you have kids, you should probably euthanize them.

[00:05:04]>> [laughter] >> It's over, folks.

[00:05:06]>> Yeah, it's pretty much over.

[00:05:08]>> I hurt my cure today to see if I still feel I focus on the tools. My parents work [music] there. The only thing that's real.

[00:05:17]>> I saw a wonderful tweet about this as well where someone was take took that wonderful picture of Keir Starmer in the back of a cab.

[00:05:24]>> Oh, yeah.

[00:05:24]>> He's just just like I don't know what >> Above Doom?

[00:05:27]>> Yeah, what he just looks like he's kind of retracted into himself and they just said and this would make a beautiful addition to my collection like General Grievous and they added it to a folder that's called British people and it was like Margaret Thatcher, Keir Starmer, Prince Andrew. [laughter] It's just like it's like we don't really add much to the world, but we do have some of the best worst pictures in the back of a cab with the worst prime ministers.

[00:05:51]>> People looking like they just [ __ ] themselves in the back of a car.

[00:05:53]>> I like the So, like that there's that famous picture of Prince Philip in the cab looking like >> getting top >> he's astral projecting.

[00:06:02]>> [laughter] >> There's It could like a Prince Andrew was in it.

[00:06:07]Like every all of the 35 prime ministers the UK has had in the last like 40 months.

[00:06:16]>> Yeah.

[00:06:17]>> The the last day before they resigned. I like to think it's the same cab.

[00:06:22]>> It's the same photographer.

[00:06:24]>> Yeah.

[00:06:24]>> Get me my shittiest photographer, Grange Calvech over at the Sun. [laughter] >> We're going We have a good photographer with a glass eye who's going to take a picture of you in the cab that gives you treatment resistant depression.

[00:06:38]>> [laughter] >> It's I say this as a white guy from England as well. Like it's real I'm probably one of the whitest people, Polish British blood. And then you get these photos that's like they turn so white they turn blue.

[00:06:50]It's >> [laughter] >> It's just they become one of the nastiest dressed.

[00:06:55]>> [laughter] >> Oh, poor poor kiss snuffed out in his prime.

[00:07:00]>> Hey up lovely [clears throat] sends his regards.

[00:07:02]>> [laughter] >> We got to get We got to get those boys back on the show.

[00:07:05]Um Oh boy. But yeah, just I guess like in all seriousness here, the lesson seems to be that you know, here's a guy who was overwhelmingly popular among basically the top 50 opinion columnists in the UK and no one else.

[00:07:22]And like he became a chip on his promise was always with labor to completely get rid of the left and thus destroy the labor party.

[00:07:30]Uh basically with what time he had hand over what's left of the country to Palantir and Israel and then just get out just in time to hand hand the keys off to the uh reform party. And yeah, I think you can see very similar parallels to the Democratic Party here in this country. Yeah.

[00:07:47]>> It's except we just change our leader a lot more. That really appears to be it.

[00:07:52]Like it I've I have disconnected from a lot of British politics. Actually, I'll be honest, it was the uh it was the Lib Dem conservative coalition that did it for me. I was just when that happened, I'm just like, "Fuck this, man. I can't." Grew up here it like have like liking the Lib Dems and liking what they were fighting for something and they just you watch that happen and just [ __ ] this [ __ ] I was always already kind of on the way out in the country anyway. Just I wasn't expecting much from Keir Starmer and wouldn't you know, I didn't get much from Keir Starmer. Which I mean probably say that about most Labour leaders for a while.

[00:08:23]>> Can you tell us anything about the uh the the new the new guy at number 10, Bo Burnham?

[00:08:28]>> I truly have not looked at Bo Burnham. I think it's nice that he came out as a he was a good singer throughout COVID. I think he made a lot of people made a lot of people happy. Didn't know he moved to Manchester. It's nice for him, I guess.

[00:08:41]Haven't really followed, honestly. You know, just it's just my my editor, Matt Hughes, said it's like this never getting [ __ ] better.

[00:08:49]And it sucks because I always had some optimism towards England and I think a lot of British people, we've had our time being smug about stuff. We can't do that anymore. After we just accept our our reform going to get in, I don't I truly don't know at this point. What it's like if the Labour Party comes in, they're going to do the same [ __ ] they've done it so far, which is [ __ ] all. I don't know. And this is very thematic for me as well. Seeing the Labour Party kiss the ass of the AI people, just that just that really just proved to me it's like, "This isn't about workers anymore. I don't give a [ __ ] about anyone."

[00:09:21]>> I mean, I I I think the most prime ministers like a real country. You know what I mean? Like a real country, like a country that um has multiple airports, has had in like a five-year period.

[00:09:36]And it's kind of shocking because 10 years ago the sense was that like, yeah, the British economy is like um fake and shitty in the same way that the American one is, right? The same type of like jobless recovery where they at least got a jump start on deindustrialization before we did. And um the the the popular knowledge was, oh, they weathered the pains of that uh before we did. So, they already the shocks are already priced into the system.

[00:10:10]And it seemed like it would be, you know, like America. There would be a combination of tech and traditional like fire sector, City of London stuff, and they would just happily be this major, if not the most important player in the EU, one of them.

[00:10:28]Uh and that they would just um happily live off the dividends of that for eternity.

[00:10:36]And uh lo and behold, a lot of people were left behind by that system. And it is it does seem like a country that is at least in the ways that the people who are capable of attaining power, at least for them, it is ungovernable.

[00:10:54]Of course, I don't think it is it's like the Lord of the Flies over there and just anyone will get thrown out in 60 days, but Starmer in particular, as well as Sunak, and I would say half of the prime ministers that got thrown out on their asses and had to resign in shame, um they all had the same delusion, which is that they could they could turn back the clock to 2014, no matter what had happened.

[00:11:21]>> Yeah, I mean, Burnham lost to Ed Miliband and Corbin, so it's like, if you like those guys and you also do well, now you well, great, you got the fourth bloke you wanted and the I did Problem is cuz I moved away from England 2000 I've been back quite a few times.

[00:11:38]So, I can't judge the country as much, even though being British gives you the lifetime right to do so. I do feel that Manchester is a lot nicer than it was growing up for me. I hope I want I want to believe that things can get better, and on some level, I know it's particularly difficult with how [ __ ] [ __ ] things are going. I do try and be like, "All right, it's a new change, but at least we rotate them quite quickly."

[00:12:00]Like you said, for like 40 of them in the last 2 months or so. I mean, Australia is competitive as well. Who knows what they're doing over there?

[00:12:08]>> Yeah, they've >> I can't even I can't even put that on my docket. Their shit's too weird for me.

[00:12:13]Like, no offense, Australians.

[00:12:15]>> Yeah, I I want to believe I want it to get better. I just don't know. The fact that Starmer had what, the equivalent of a supermajority, and just kind of was like, "Well, what we need to do is bring in the nice fellows from McKinsey." And let the McKinsey people have a go. And yes, we Do we need a national health service? And they've been toying with that idea for a while. I don't think Starmer was really on the way out with that, but it's just frustrating because I don't know. Feel like you have a shot, nothing happens. Feel like a shot, nothing happens. I don't know. Don't think there's really a comparison to that in American politics, but maybe.

[00:12:47]>> Well, I think it is. America is in a similar place, right? I mean, I think there are a lot of parallels between the Brandon administration and Starmer. We, of course, when we came up with the the idea of Brandons as a measurement unit, we actually thought that Starmer was the most Brandon leader on the planet planet.

[00:13:07]>> Medichlorians.

[00:13:08]>> Oh my god, get He was like two Brandons, and Brandon was only one Brandon. I mean, we're prevented from having as many leaders in such a short window of time by virtue of having a presidential system, but I mean, three one-term presidents in a row is as close as you get to that.

[00:13:27]Okay, it it is a different situation in that two of those terms are Donald Trump's terms and it's not like it's not quite a Kier Starmer thing where I don't think Donald Trump got in there and thought, "Okay, let's return to good old-fashioned Obama-era technocrat stuff."

[00:13:45]Well, that clearly wasn't the case. With Biden, it was even weirder because he thought he could make this weird Frankenstein monster that combined a lot of modern Democratic Party elements like feverish loyalty to Israel and NATO fetishism with a sort of half-remembered version of New Deal Democrat politics.

[00:14:06]And it winded up just being this thing that everyone [ __ ] hated.

[00:14:10]>> Well, I think that the problem with a lot like British politics, American politics, the tech industry, business in general, everyone is trying to turn the clock back to 2014. They're all trying to get back to a time when like tech built new [ __ ] when governance governance did things, there was hope, we could we had a chance to change the systems we're part of versus reinforcing them. I think everyone wants to do that other than changing the systems back to how they were or making any kind of systemic changes that might actually do that. That might, you know, some regulation of tech to make the tech industry smaller, to push private enterprises out of government. But no, no, no, they love that. They love that part. It's just the complaining part that they don't like. They don't want people to be so Oh, oh, what? You don't get anything. Oh, your life sucks. Oh, your money doesn't go anywhere near as far and every time you open up your device, it tortures you. Oh, what? Why can't you just be [ __ ] happy with how much worse [ __ ] has got?

[00:15:08]With the only thing we still have from back then is Gary [ __ ] Vaynerchuk.

[00:15:13]I thought he was long gone, but we still have He's [ __ ] talking about AI now.

[00:15:17]It's just They want to bring back all the [ __ ] things in 2014 and then still keep things in the kind of hyper capitalist reality where nothing really connects properly to anything, nothing makes any sense other than the fact that number go up. It's just very but Starmer kind of stood for that. Stood for the kind of like, oh, what if we just got back to a government that kind of shook hands with people and took photos, wouldn't that be nice?

[00:15:42][ __ ] Remember that was the first time Snapchat tried snap. Oh my god, we got denied the chance to see Keir Starmer in Snapchat spectacles. [ __ ] this world.

[00:15:52][laughter] [ __ ] this world entirely. He would we if he'd have had another two months, Keir Starmer would have been at number 10 with those [ __ ] royal orb looking things.

[00:16:01]>> [laughter] >> Just [ __ ] [ __ ] trying to like rotate a burger in three dimensions.

[00:16:08]>> [laughter] >> Now he's headed to to Blue Bayou.

[00:16:12]>> We missed Keir Starmer going on one of those nightmare kick streams where like Adrian Broner [laughter] and like just abuses like 20 women while gambling.

[00:16:23]We missed [laughter] so much.

[00:16:25]>> Yeah, they we they've missed the polymarketization of England. But I think we can still have it though.

[00:16:31]>> That is what what you said about Starmer it seemed like he came in there and set up a bunch of things for Palantir, you know, kept the ball rolling on the privatization and gutting of the NHS certainly as you pointed out something that predates him of course. AF Lovely his theory was always been Starmer's role whether he knows it or not has been to just kind of be like the fall guy to set up Britain to be Russia in the 90s where whoever whatever next three guys are in there, it is going to be like the Yeltsin administration where all these great national assets and things like the NHS are gutted largely by American companies or public private partnerships and everything is just ripped apart and stripped for pieces.

[00:17:21]And it's pretty [ __ ] stultifying to think about not just that, because that's obviously horrifying, how much worse things would get uh with all those things ripped away.

[00:17:34]But also, considering how right-wing uh the UK is now, how much more right-wing would get after that, too? After even the most threadbare of things they can still point to and say, "Well, we have that.

[00:17:49]We have that We have this thing that America doesn't have." Once you rip those away, how bad can it get?

[00:17:56]It It It's really [ __ ] awful to think about. And looking at um Starmer's tenure, it's hard to I cannot say I disagree with Mr. Lovely's um appraisal of him as a figure. I- I- I- I- If you were If you were Philip, you know, Philip Green from Casino, our favorite guy to compare people to, he's Philip Green from Casino and the mafia and the teamsters their Palantir.

[00:18:19]>> Mhm.

[00:18:20]>> If he is that, what would he have done differently?

[00:18:25]He did the I'm the exact same things you would do if you were setting that up, you know?

[00:18:28]>> Yeah, and I mean, it's just the the slow roll of nothing. It's It's I can't we joke about nihilism and all that.

[00:18:36]It's just kind of hard to look at anything that's happening in England as something that's an expression of anything other than just either the right wing or the city of London.

[00:18:46]Just about the wealthiest people or the most evil people, some of which share the same I don't know dialogue conference membership, I guess.

[00:18:54]They all get to go on the same Peter Thiel The Peter Thiel cruise that he does.

[00:18:59]It's just a just a thought I had for going on here, but no, it's really rough to watch, because the right wing would like to bring it way back, like 50, 100 years, maybe more even. And whatever you call It's not the the centrist centrist power circle in England. They desperately want to just go back to Tony Blair. They wish they could [ __ ] put Blair in there. It's pretty Well, we can we invade Iraq again?

[00:19:24]Well, why not? Well, we don't want to invade Iraq. That looks difficult. But we could do we could why why don't we just we could relaunch the iPhone what?

[00:19:32]10? And we can just kind of do whatever the [ __ ] we want for a while. Yeah, [snorts] just like nothing changes other than number go up.

[00:19:40]>> It's just It's very very very frustrating cuz it is everything right now. They want to return to how things were in the sense they want us to calm the [ __ ] down despite things getting worse for everyone other than them. Keir Starmer will be fine. He'll go and walk into a McKinsey.

[00:19:54]Whatever it is. I truly don't know how one gets a job at McKinsey. I think you get chosen. It's like a stonemason's thing.

[00:20:01]>> [laughter] >> And I've been I've been watching you uh, you know, appear on the business press as sort of like the anointed AI skeptic. And like this is in a world where like we've talked about politically, culturally, there seems to be this longing to return to a bygone era where technology and politics still offered something hopeful about the future or the possibility that things things would change. That something could change.

[00:20:26]And I I've mentioned to you before that I think AI has been marketed quite successfully as this kind of deus ex machina, this magical technology that will change everything. That will change politics. That will change culture. And will make everyone rich or it just offers a a hope of a new horizon. And a time when people are feel more and more like that horizon of possibility has closed.

[00:20:50]But like so in light of that, I would like to ask you about your recent reporting on OpenAI and that company's books. So could you tell us what is actually going on at OpenAI and the question of are [clears throat] Are making money or not?

[00:21:06]>> So let let's start with the thing they're not. They're losing billions of dollars. How many billions is a good question. So, last year 2025, so I got their audited financials and validated by the FT.

[00:21:17]So, they burned about $21 billion to make $13 billion or so. Now, the the funny thing is is their net loss was higher than that because they did some weird accountancy [ __ ] They also claimed that they It was funny. The FT and I reported this out and someone, I assume from OpenAI, I truly don't know, was like, "Actually, when you look at the accounting, we only lost $8 billion." This This is partly because they moved like $18 billion worth of costs onto another balance sheet that they don't name. But when you actually look at the underlying cost of this, this company spent $19.18 billion just on R&D, which is training the models and probably some staff, too.

[00:21:55]And then there's the funny one. They spent $5.73 billion on sales and marketing costs. That is more than the Coca-Cola Corporation.

[00:22:04]Now, the reason I bring this up first is they also have a number called There's a $7.5 billion cost of revenue number.

[00:22:11]Now, people are going to say, "Well, $13.07 billion with the cost of revenue of $7.5 billion, isn't that profitable?" Absolutely not.

[00:22:18]OpenAI is clearly moving inference costs for its free users or its They give away a bunch of tokens as well to people.

[00:22:26]They do it through a startups program and everything into their sales and marketing vertical. $5.73 billion, OpenAI does stuff on ads with Facebook and Reddit and all of them, but they don't spend that much. That'd be more than Temu Temu? Temu? And Shein combined, and they're both those.

[00:22:43]Those are the two of the largest Facebook ad spenders in the world. So, what's actually happening here is this company's doing some really fun accountancy [ __ ] to try and get away from the fact that they spent $34 billion to make $13.07 billion. These companies [ __ ] suck. They're all terrible businesses.

[00:23:00]But I also think it ties into what we're all talking about, which is everyone is pushing these [ __ ] heap companies upper hill. They're pushing AI for everything cuz they want to go back to the time when you could just read TechCrunch and be like, "Oh, Facebook added a new emoji. How lovely. I'd like to invest in that." We can make new We make They want to get back to the era when you actually believed Silicon Valley could build another Facebook. When all Silicon Valley can build is OpenAI, a company that raised $122 billion this this year in one round. And uh yeah, it's They're projecting to lose $852 billion by the end of 2030.

[00:23:37]It's just It's kayfabe. It's all pretending that this is this is yet another Facebook or Apple or smartphone or cloud computing or App Store when it's just a cobbled together mess of a business with a product best known for its hallucinations.

[00:23:52]>> Uh uh another thing that that I've seen you mention is this idea of ROI. Now, correct me if I'm wrong, but that that refers to return on investment. And everyone is looking for is there ROI in a company like OpenAI?

[00:24:06]>> Mhm.

[00:24:06]>> But but like okay, so in the interview like when you say like in in your in your writing that you cannot find any return on investment for a company like OpenAI. What does What do their boosters or defenders like what is the argument against this? Or like what what what sort of when your skepticism like is it met with some sort of I don't know, like a counterexample?

[00:24:26]>> So, the counterexample people give is they go, "Well, it's writing a lot of code." Now, to be clear, even at the most complex level, these models are probabilistic. People are going to love that when they hear it. Cloud code is a series of deterministic scripts bolted on top of an LLM to make them do more stuff, usually burning more tokens in the process, but long story short, code is the one that everyone gives. That Spotify claims that their best coders are now writing no code.

[00:24:51]They're using LLMs, but no one can seem to actually give you an example of what this means in the end or whether the code is good or functional.

[00:24:58]You can, however, look at basically any app these days and see how shitty it's got. But, putting all that aside, the ROI conversation on its face is a sign that there isn't ROI because why would we be debating it otherwise?

[00:25:12]We wouldn't be sitting there being like, "Is it" You wouldn't have people emphatically saying, "AI is here and it's real."

[00:25:18]if they didn't feel some compulsion to prove that. Versus, I don't remember anyone going around being like, don't use If you don't have an iPhone. They're They're going to take your They're going to You don't If you don't use Google Docs, you're [ __ ] If you don't have a Kindle."

[00:25:33]>> Is Coca-Cola really a delicious beverage?

[00:25:36]>> [laughter] >> It's just But, but the fundamental ROI thing is it's actually very difficult to measure the cost of a task with AI. It's very difficult because for many years these companies would sell you a subscription of 20 bucks, 100 bucks, 200 bucks a month. And within that, you'd have rate limits. And no one really knew how much those rate limits were. They were just a lot.

[00:25:58]Well, it turns out according to SemiAnalysis that because when you use those services, you don't pay the actual cost of the tokens, the AI tokens that run the models or the models use.

[00:26:08]And uh it turns out that SemiAnalysis found that on the 200 buck a month plan, on with Anthropic, you can burn $8,000 worth of tokens. And on OpenAI's, you can burn $14,000 worth.

[00:26:19]Now, those numbers are both I know I know that this isn't a an economics podcast, but those numbers are both much higher than $200.

[00:26:27]>> Um oh oh holy [ __ ] Hold on. I have to call my retirement account guy.

[00:26:32]>> Yeah, you You really shouldn't have really shouldn't have levered yourself in this, Felix. But, the point I'm making is >> out I got the order of numbers wrong.

[00:26:40]>> Oh. Oh.

[00:26:42]>> no. Yeah, it turns out I my retirement account was predicated on the idea that uh $300 is more than $14,000.

[00:26:50]>> [ __ ] Well, it turns out that's actually Sam Altman's account too. So, what happened? The reason I bring this up is most people experience AI in this way.

[00:26:59]Most people you pay 20 bucks, 100 bucks, 200 bucks a month to use an AI service.

[00:27:04]Earlier this year, around March, both OpenAI and Anthropic went to their largest accounts, their enterprise accounts, and said, "Okay, guess what?

[00:27:11]Great news. Instead of paying us a single monthly fee, you're actually just going to pay for the cost of tokens served at a per million token rate."

[00:27:20]Uber through their entire AI token budget for the year in a single quarter.

[00:27:25]Uh what what Axios reported that a company Madison Mills over Axios reported that a company, because they didn't up set up spending controls, spent half a billion dollars on Anthropic in a single month.

[00:27:38]And um yeah, so all of these enterprises went from this kind of subsidized thing where they could kind of all you can eat to suddenly having engineers spending thousands of dollars a month each. I Zillow ran through their entire Cursor AI budget it for the year by the end of May. Uh Brex had to limit engineers, I think, to like 1,500, 2,000 dollars a month or a week, I forget. Uber's limiting them to 1,500 dollars a month.

[00:28:03]But, the point is that these companies were like, "Oh, [ __ ] Wait.

[00:28:07]We didn't make sure that this was useful. We just told everyone to use as much AI as possible, and then everyone did that, cost them a bunch of money, and now everyone's suddenly going, 'Wait, is there an ROI for this [ __ ] Wait, wait, wait, wait. Is this is this good? I didn't check before.'"

[00:28:21]So, Sam >> Ed, you you um you brought up Uber, and this entire pattern of like new or newish technology gets introduced, the pricing is much lower than the actual cost, the company incurs a loss. Uh this is a familiar story in tech, right? With a lot of streaming services, with Uber, with Grubhub, with most of these types of businesses that popped up around you know, 15, 10 years ago. It was a similar All these companies lost billions of dollars. So, of course, not nearly as much as OpenAI has in this last year. But, those were under the assumption that okay, we are going to basically subsidize consumers to use these services.

[00:29:10]And once our service, once, you know, like Uber gets enough market share over Lyft or whatever other competitors, we will have we we we will we'll be Coca-Cola, and we can jack the prices up to whatever we need to actually make a profit. With AI, the weird thing about it is is you can't actually say out loud what what the the pay what hitting pay dirt will be, >> Right. Yeah.

[00:29:38]>> It costs this exorbitant exorbitant amount of money, and the only way that you could ever really make it like cost-effective would be if we move, you know, we achieve like full-scale nuclear fusion energy-positive nuclear fusion, convert the whole grid in like 5 years. Or, you know, um I guess build more existing nuclear power plants. But, my point is like even if we did that, it still wouldn't it wouldn't make up for the cost of the actual reason that investors were excited about AI, which is the idea that it could replace labor costs.

[00:30:19]>> Yeah.

[00:30:19]>> And as it seems >> it seems as it seems it seems like much more expensive to have to pay for all these [ __ ] tokens than it would be to just hire someone.

[00:30:32]>> Yes.

[00:30:33]>> And that's pretty much it. But, on top of that, it doesn't do any of the things that a regular person would do with any reliability.

[00:30:40]>> Exactly. Exactly. And of course, you can't say that part out loud. It's too like ghoulish. Like you would Too many people would go well well what the [ __ ] This is going to replace me?

[00:30:52]Um but also I mean it going back for our previous conversation about the the desire of all political parties, all viable political parties in the West seeking to return to either 100 years ago or 15 years ago.

[00:31:08]It's very apt that the dominant definitely not economic engine, but you know, the thing that everyone is going to remember when they talk about this time. The big defining technology of this age is essentially something that just slams together pre-existing images and texts and thoughts and ideas into like a shitty undigestible mess that just gets further and further deformed the more times you print it out.

[00:31:37]>> Yeah, and what's funny is >> the perfect thing for our time.

[00:31:40]>> Well, but what's great about it as well is open AI the history of open AI is like in 20 I 15 2017 Sam Altman and a bunch of other rich idiots got together sent emails to each other being like if the mind was a computer we could have a billion minds and other genius they like him and Elon Musk sending each other like just like actual hot couch guy [ __ ] just like stoner level stuff being like wow, the mind is a powerful thing and we can't let Google have every mind with DeepMind. That Sam Altman was very much part of the elder guard of the valley is one of the first people in Y Combinator. He's invested in but he was like part of Stripe cuz Paul Graham of Y Combinator he is very much of the 2010-2014 era.

[00:32:26]And he's trying to build the company in it's cargo cult [ __ ] Trying to do what well Amazon Web Services cost a lot of money, right? Now that belief has allowed people to justify like a trillion dollars worth of GP capex.

[00:32:39]Truth is, Amazon Web Services cost about $53 billion. This is all of Amazon's combined capex, including AWS and the the the store and everything was about $53 billion between 2003 and 2017.

[00:32:53]Amazon Web Services became cash flow positive positive in nine. OpenAI raised $122 billion this year. Anthropic, if they close every part of it, will have raised $95 billion this year. These are not comparable, but because the entirety of Silicon Valley is just management consultants and like ayahuasca victims, they are all just sitting [laughter] around being like, "What do we do? We throw a bunch of money at it, and if people say they don't like it, we scream at them. We scream at them with all our might. We we don't have really any other ideas. What It was You've kind kind of intimated this earlier. It's when no one has any other ideas, they just think, 'What if we made a machine that does ideas for us?'"

[00:33:35]They're all talking about recursive self-improvement suddenly because they've actually run out of ideas.

[00:33:39]They're like, "Our next big idea is that it will do the ideas for us. It will research the AI for us." Because all of the people that build stuff, all the actual engineers, they got chased out a long time ago. What's left are just software engineers being tortured in every company, being like, "Use AI more." "I don't want to. It makes the code The code isn't very good." "Use more, but also don't use too much. Don't use too much of it. It's too expensive, but use more. Use the perfect amount, which I can't tell you."

[00:34:09]>> Yeah. I I need to go back to like a a more basic question cuz I It It seems from like It seems to me that like whether you're a software engineer or people like in in all kinds of email jobs now are under incredible pressure to use AI as part of their workflow. And and to use it often and more. But But at the same time, it gets back to this question of tokens. And here's the thing I don't fully understand. Is a token like essentially just a subscription to use the AI like cuz I don't use AI in anything I okay like my equivalent would be like I use like things I use for to do this job is like Google, right?

[00:34:44]And when I go to Google and I type in Stannis Baratheon gifts and like you know I'm description of my work day here. I'm paying for Google through I pay for the internet like my internet service provider. I'm paying for that when I go to Google it's essentially free at the point of service. I'm paying to use the internet.

[00:35:02]I thought for like for a while these large language models they were free but now they're being metered by a kind of subscription fee which is essentially what what what's known as these tokens, right? So like are you paying a monthly fee or you paying per use? Like how do I how does a token work and why are they called tokens?

[00:35:20]>> Happily happy to explain. So tokens is just the unit of measurement of a like three quarters of a word word even jeez.

[00:35:27]Three quarters of a word a token. So you they charge on a per million token basis in input and output tokens. So when you feed in I don't know your 900 word Kier Starmer Naruto fanfic into Claude that that would however if that were you could use a certain amount of tokens based on how how long it was and then the output it creates being like make it sexier and it would be like your it would then in the background do some it would build a plan of how to make your Naruto fanfic with Kier Starmer sexier and then it would spit it out. When it spits that stuff out those are the output tokens.

[00:36:00]Somewhat of a simplification. So the majority of people so when you use Google search and it spits out an AI overview that does not make any sense or what have you that is burning tokens on some model we don't know which. Now when you use chat GPT or Claude it will be using a model from one of those two companies from Open AI or Anthropic and you as a user paying on a monthly basis don't experience the tokenization part. You're not you don't know. Like you could work it out. There are means of working it out, but you don't. You don't see. But because enterprises because they they did the thing the Coca-Cola thing Felix mentioned where it's like, "Yeah, we're going to jack up the prices now." Except they did it and within 2 months everyone was like, "Fuck this [ __ ] What the [ __ ] do you mean?"

[00:36:45]So they went from being like cultists, they went from saying like, "This is the literal future of everything. It's going to change everything." to just like full-scale austerity measures just like, "I I don't know.

[00:36:56]We couldn't possibly looking into open source models. No."

[00:37:00]And it's really funny because the very same people at Cisco's is his president GT Patel was saying, "Yeah, the cost of tokens vastly outweigh it outweighs the cost of human beings is the return on investment isn't there."

[00:37:15]And what's funny is like these are all companies that 4-5 months ago were saying that AI was keeping them rock hard all day long. Like they were like, "This is the best thing ever. This is changing everything." And it's because most people using AI are having it subsidized and the amount they're having it subsidized is like 20 to 40 times its actual value. And just to be clear, the only people will say, "Oh, they're doing that because right now they're just in growth stage."

[00:37:40]No, the reason they're doing that the reason they're subsidizing so heavily is they know the average person would not pay anywhere near how much it actually costs. And we can kind of see that when we have enterprises already it took less than 3 months for them to start flipping out and taking hostages.

[00:37:58]>> You brought up the example also of um you know, how people I never heard this one before by the way of how people talk about how AWS costs a lot. Which as you pointed out, the cost of AWS is you know, minuscule compared to this stuff. But most crucially AWS does the thing that you pay it to do.

[00:38:22]>> Yes.

[00:38:23]>> The thing the and the entire thing that it is marketed as being capable of doing. Uh nothing to my knowledge there isn't something where like half the time when you try to upload something host something on AWS, it's just it doesn't make it, you know?

[00:38:41]>> Yeah.

[00:38:41]>> Whereas with these companies they are promising something that probably is impossible, but if it was would cost trillions of dollars and make the world immeasurably worse.

[00:38:54]>> Yeah.

[00:38:56]And it's very frustrating as well because on doing this and the way they're raising this money, the way they're Anthropic Open AI raising over $200 billion this year, the reason they've done why they've done that even is just being like, "Yeah, this thing which does not do this right now, we can't tell you how it will, will definitely become something completely different. And man, when that happens, you're [ __ ] You should be terrified of us. You should be terrified. Don't ban us, but be terrified."

[00:39:23]Like it's just it's it's it is cargo cult [ __ ] It's a bunch of guys who do not really know they don't know how to build a real company. They don't they know how to build hype cuz that's all Silicon Valley is anymore. It's just hype and speculation. They only know how to do the stuff around the stuff. They know how to get excited about nothing.

[00:39:42]They know how to explain away stuff.

[00:39:44]Like there are people that heard Open AI from my reporting lost $20.92 billion and they were like, "Actually, that's fine.

[00:39:51]It's good." And that's because this is Silicon Valley is just the cargo cult now. It's no longer run by the people who built the things that matter. It's no longer stewarded or invested in by people that do real jobs. I mean, I say this is someone with the email job of all email jobs and that I mostly send emails. And it's it's genuinely horrifying to watch both what they're doing, but also how [ __ ] easily everyone's falling for it. I mean, these but the wallet inspector has been rolling through the media for the best part of 3 years now, and it's like you try and explain it to them, and I'm I'm serious. Journalists, you get journalists who do this, you get investors who do this. I've had this argument a lot, and they say they just go, "Well, Uber lost a lot of money, $32 billion between 2019 and 2022, by the way. Uber lost a lot of money. AWS lost a They repeat these things like they're in a cult. Because when you actually Well, still abstractions one, when you have to admit I'm going to say it is you have to admit the tech industry is in a very bad way, but you also have to accept that man, there are a lot of people who can be fooled into just about anything. And fooled into defending it, too.

[00:41:00]>> That is very funny, by the way, how part of the marketing for it is talking about how scary it is. Like nothing else is like that. Like again, to bring the example back to Coca-Cola.

[00:41:11]If they were like, "Okay, we had a lot of success with like the polar bears.

[00:41:16]People think they're really cute.

[00:41:17]>> [laughter] >> But we're going to do this ad where it's a family, right? And the cute little boy with like a missing front tooth drinks less Coke.

[00:41:29]And Coke is so good that it's terrifying, right?

[00:41:32]So he kills his entire family because they're out of Coke.

[00:41:36]>> [laughter] >> The family annihilates a era of Coke.

[00:41:40]>> Yeah, we're going to anonymously leak leak things to the press where we're like we're we invented a new cherry Coke.

[00:41:46]It's so good. I think people are going to become a thousand pounds. It's not [laughter] going to work anymore.

[00:41:52]>> You better stop working out now.

[00:41:54]>> Yeah.

[00:41:56]Yeah, there's nothing That that is such a sign of like you just don't really believe in what you're doing at all. But we've talked about this every time, also, that we've had you on the show to talk about AI, that there are things that AI can do that are genuinely very exciting.

[00:42:15]I I've used the example of that my friend who knows infinitely more about the subject than I do about how AI could be used to research, you know, something like uh anti drug-resistant bacteria that uh has evolved to you know, resist existing antibiotics.

[00:42:34]>> Which wouldn't necessarily But that isn't necessarily large language models.

[00:42:37]That's something else. That's other kinds of machine learning. Yeah.

[00:42:39]>> Ex- Exactly. Exactly. Or like uh the example I've been thinking of lately is how, you know, a much less important idea than that, but I could very easily see it being used to end this current problem we have in games of like the 12-year-long development cycle. AI could do a lot of the incredibly dull, soul-crushing work that all games must have now where every bubble in a cup of soda in the game has to have individually individually rendered bubbles.

[00:43:15]>> But even that though is like that's also something that isn't a generative thing.

[00:43:19]Like they you reuse assets all the time in games. It's just It's like the promise is always wouldn't it be nice if and then some other stuff has to happen that makes it useful in that way and then we can be excited about it. And it's frustrating because it's the same [ __ ] bad faith [ __ ] every time. Oh, it's a can't Not I'm not saying you're doing this. It's just like oh, it scans for cancers. Oh, you don't like robots?

[00:43:43]It's like not what I'm talking about.

[00:43:45]>> Yeah, exactly. Exactly. Like if this was a more If we If we had a more functional economy and a more clear-eyed way of viewing things, uh this would already be like how it was marketed.

[00:44:00]It would They They would go, "Oh my god, it can make It has all these advantages for researchers. It can make the task of programming so much easier, but instead it's marketed as this is horrifying and it's also going to make uh it's going to make it so that Sydney Sweeney exists as a hologram in your living room.

[00:44:20]>> But the the funny thing is is it it gets back to the 2014 thing again as well.

[00:44:24]Because they can't do that. They can't do that cuz it costs too much money.

[00:44:29]They they can't they can't do that because Open AI they raised 40 40 something bit $42 billion in 2025 and then in March 2026 had to raise $122 billion. It's very obvious that the economics don't work just from that. And they're like, well, where is the money gone? What's it gotten us? Well, um we got some models and we fed maybe $20 billion directly to Microsoft, Apple, uh sorry, um Amazon and Google.

[00:44:58]Yeah, and uh everyone at the tech companies is [ __ ] miserable because they're constantly being screamed at to use this machine that makes mistakes all the time and then screamed at when it makes the mistakes.

[00:45:09]Yeah, but this is how this is how we did tech, right? That's how they made the iPhone, right? They just uh screamed at everyone constantly.

[00:45:17]I guess that's how Steve Jobs did did the design aspect, but >> [laughter] >> but I think he mostly left the technological stuff to the people who could actually build things. But it's just it's a bunch of people shuffling around desperate to create a new Google because it's not enough for AI to be normal.

[00:45:33]It's not enough. If like if this was an industry with a total addressable market of 50 to 100 billion dollars, which would still be way too much, but let's say it was that that is not enough.

[00:45:43]Hyperscalers, uh so Google App, Google, Amazon, Microsoft, and Meta even are going to uh by the end of 2027 spend over $2 trillion on this [ __ ] They need this to be Google Search plus iPhone plus another Google Search plus plus plus plus They cannot, this cannot just be that. Because if they have to talk about what they had today, like, yeah, we made something really unprofitable. Some people really like it. Some people it drives them to murder.

[00:46:14]Uh, also, we need more money than has ever been spent on anything for this.

[00:46:19]Does it make mistakes? Yes. Are we fixing the mistakes? Well, we can't.

[00:46:23]Open AI said it's mathematically impossible. However, absolutely it will at some point do something so good you should be terrified of it, but not today.

[00:46:34]>> Other than mythos.

[00:46:36]>> Over the next five years, it is likely that Coca-Cola will get so delicious that anyone who has been drinking Pepsi will be punished. And in fact, the like the beverage will be so good that if you don't have one with lunch, you will be fired from your job.

[00:46:52]>> Pretty much. Except, I mean, diet Pepsi's foul. But >> [laughter] >> But that's that's a separate debate.

[00:46:59]They're going to get me banned. But it's it But it really is, it's just these people who are like, how did Google search get big? Well, we pushed everyone to use it, which isn't what happened, of course. It was genuine genuinely like a college project of two guys. Like, they were it's far from a perfect company, but it came from a point where even though, by the way, Sergey and Larry in the early days, they were like, yeah, you know, if people add ads to this search engine, it will change the incentive structure of the company to work against >> against the user. And it's like, wow, really?

[00:47:32]>> But it's all of the people that build things have left. Sergey Brin has actually come back to Google, but he hasn't done any real work in 20 years. I thought I think he was had some sort of Epstein relation, right? Um, it's just very funny because it's a bunch of people that have run out of ideas who have more money and talent than anyone has ever had before. Like, every single engineer is focused on this. Every If you need money for a data center, you can [ __ ] go out the door, wave your hand, someone will give you a billion dollars, no problem. As long as you're buying GPUs, it's fine. They have all the king's horses, all the king's men. And they haven't got [ __ ] Wow, you can write some code with it. Wow, you can And mostly when it writes the code, it is copying an open-source project that it was trained upon. Like most of this is just again, recreating the past again and again with mistakes because these things don't think, they don't have a consciousness. No, they can't sell on that. They can't say, "Yeah, what you've got today is what you're going to have forever." They have to say, "Oh, it's a superhuman coder and it it's going to teach itself." Cuz if they if they stopped tap dancing for even a second, if they dropped kayfabe, they have to look at the economics and the economics are actually insane. And people say, "Oh, it's like the railroads." That was a long time ago. All the railroad companies went went bankrupt, but that was also a long time ago. What do you >> [laughter] >> If that's the best you [ __ ] got.

[00:48:54]>> [snorts and gasps] >> And to talk about, you know, all the king's horses, all the king's men. Uh what one man in particular, uh Sam Altman, just as an individual, as a guy who is sort of, I don't know, like the figurehead of a lot of what you're talking about. I'm wondering if you followed this story that uh I I I saw this week about how basically all of Hollywood has simultaneously dropped Luca Guadagnino Guadagnino's OpenAI Sam Altman biopic starring Andrew Garfield, which is basically like it's already in the can. And now basically like this movie was financed and now will like essentially will not be distributed because all of Hollywood has some sort of economic relationship with OpenAI. I mean, like I do do you have any perspective on that?

[00:49:39]>> I mean, it's I don't know if you saw the story where it was from the Haibach and Lakes where apparently Donald Trump was just making fun of Jeff Bezos for like begging him. So he's like, "He is begging me for everything." But that's because Jeff Bezos is a [ __ ] coward.

[00:49:54]Like an actual worm and a coward.

[00:49:57]And what likely happened here was because they invested $50 billion in OpenAI this year with 15 I think 15 billion of that contingent on them going public or reaching AGI, so just going public.

[00:50:09]And yeah, so what I think probably happened was Sam Altman pitched a hissy fit, which is the lamest [ __ ] in the world.

[00:50:17]He wants to be Sam Altman wants to be this cultural icon, this famed his man of history, this elder statesman. But, he's like the first critical movie he's like, "I couldn't possibly."

[00:50:27]>> Also, Andrew Garfield the end of the Well, Andrew Garfield was in The Social Network, which is like the template for this movie. And like, you know, The Social Network is not exactly like a glowing portrait of Mark Zuckerberg.

[00:50:38]But, despite the fact I think nothing has done No no one cultural item has done more to burnish and and launder the reputation of Mark Zuckerberg more than The Social Network. Cuz it's like it's like Goodfellas or people who want to be Mark Zuckerberg, you know? You could say it's like a cautionary tale, but you're like, "Oh, this guy is the man. He built this billion-dollar company to call chicks ugly who wouldn't [ __ ] him in college." Isn't that the dream that we all have every man?

[00:51:04]Uh but like, you know, I I I don't know what what the content of this movie is called Artificial, the the Luca Guadagnino movie. Like, I don't really know what to make of this, but like there are some pretty nasty allegations about Sam Altman out there.

[00:51:17]>> Yes. And the thing is with Altman, is he probably didn't even cancel it because of the allegations.

[00:51:24]Like he probably I don't think anything with his sister is in there. She alleges he like less than her. It's like a really [ __ ] horrible lawsuit, like truly awful. I I don't think that's even in the movie.

[00:51:35]I think Sam Altman is just a very thin-skinned man. There was a clip back in November of last year where Brad Gerstner, who is one of the most simpering men alive, was like, "Yeah, so I heard that you might have some growth issues." And Sam Altman just goes, "Well, you know, if I I yeah, I'll get someone to buy you a stock with my more way more than $13 billion." He just like immediately starts going >> [snorts] >> And I think what it is is that Altman is a very self-conscious guy desperate to appear like a cool and chill one. He had a period of time about month month and a half ago where he was going on Twitter and doing like 2011 era "I'm so drunk" tweeting.

[00:52:13]And it's just like [laughter] this is exactly that is he just like like [ __ ] 3:00 in the morning being like, "I don't think I should be alone with GPT right now." And it's it's just he's just a very I keep coming back to it. They're all just very boring oafs.

[00:52:30]This is I don't want a movie about it cuz you say something mean about me.

[00:52:35]Yeah, that's probably in the entirety.

[00:52:36]He probably told Jeff Bezos he needs this to happen or that you shouldn't or he like intimated that it would be bad for him or just he asked too many made it happen. It's like Coyote versus Acme except I'm sure that this movie will be just I'm sure it's going to be kind of dull because I Empire of AI, great book by Karen Hao.

[00:52:56]I've read even the Ronan Farrow Sam Altman story.

[00:53:01]The when you actually look at the life and times of these people, they're dreadfully dull.

[00:53:06]They're so [laughter] dull. Oh my you And I mentioned the kind of stoner style emails they send to each other. And it's just like cuz they're all [ __ ] boring. He drives like a $5 million car.

[00:53:19]He has a house in San Francisco. He had to sue the people he bought it from because it was falling apart. It's just like, "Oh my god, go on move to a [ __ ] HOA, Altman. You'd love it there." It's just the most dull [ __ ] ever. Dario Ammannati dull sounds like he's doing the Elizabeth Holmes CEO voice. And it's just all of these people are dreadfully dull. So, the movie I'm surprised anything happens in it. Like I don't know what what is the thing what is the plot of the movie? Sam Altman gets given he he has a company 2005 called looped that is like falling apart immediately gets bought by Green Dot. He's given percent or two in stripe by Paul Graham for no reason. Paul Graham just said, yeah, you should have you should buy into this. Paul Graham makes him the head of Y Combinator in I think 2016 says he's the only person for the job. A lot of this is just quite I hate the I hate the obvious joke. It's like a lot of this is quite literally an email. A lot of this is just guys emailing people and zooming each other. When he got kicked out, I'm sure the movie doesn't have anything about the board member that Sam Altman added to open AI in 2023, you know, Larry Summers.

[00:54:27]And [laughter] anything about Larry Summers?

[00:54:29]>> another star of the social network. Seen this movie before.

[00:54:32]>> But that's but that's the thing is just I'm sure the movie's quite dull because there's not even there's there's no one Peter Thiel shows up, I guess. You got Sam Altman, you got Dario Amodei, dull, dull, dull. You hear him speak, the dullest [ __ ] Oh god, I'm just Every time I think of the movie, I'm like, what could you possibly have happen in this [ __ ] A [ __ ] phone call between Greg Brockman and Kara Swisher? Who gives a [ __ ] [laughter] Who [ __ ] cares?

[00:55:02]>> Well, you bring up another another personality and I'm using that term very liberally here.

[00:55:09]Dario, the head of Anthropic.

[00:55:12]Yeah, yeah, and I've been seeing a lot more of him of his sparkling personality lately. And basically it was in the context of him being asked, has Claude been used in the like sort of targeting chains of the US military that were perhaps related to I don't know, fire two cruise missiles at a girl's school in Iran. And his his answer was pretty much like, >> Now, I you know, I I I I know, again, again, I would say I think overall the use of these the the use of these models is appropriate. I think it's good on net. Um you know, but military decision-makers make terrible mistakes even what even even at the best of times and I I don't know if we're in the best of times. What we've seen here is Claude assists, but a human makes the final call. So, a human made that final call, not Claude.

[00:56:01]>> Uh maybe, but like it's still it's like we had like it was human beings who made that decision, not Claude. But like I I do want to ask you about the ban of Fable and Mythos. They were it was basically the Trump administration has cited national security concerns and issued an export control directive that suspended access to Fable 5 and Mythos 5 models by uh any foreign national including Anthropic's own employees. Can you give us some background on that? And first of all, like what are Fable and Mythos? And >> Okay.

[00:56:34]So, back in April, Anthropic, because they were bored, I assume, were like, "Well, we're launching a new model, but we're not launching it. It's too powerful. It's called Mythos. We're doing a thing called Project Glasswing."

[00:56:47]And it's able to find vulnerabilities and all sorts of stuff. And the vulnerabilities are scary. We do scared of it. Can't possibly launch it other than to 15 different partners. A couple weeks later, they're like, "All right, we're going to give it to 150 partners."

[00:57:01]And you know, this whole time Dario Amodei is out there going, "It's too scary. It's too scary to launch to people, but other than the people we like." And China, China, China, just the usual. Again, dreadfully dull people just China hawk [ __ ] What if What if China can generate Kier Starmer Naruto fan fiction? The terrible national security concerns. But then a couple weeks ago, Anthropic says, "Actually, we're going to launch Fable."

[00:57:27]Now, Fable is Mythos, a Mythos class model.

[00:57:30]And this model will be exactly the same as Mythos other than you can't do cybersecurity, you can't do biology.

[00:57:37]Now, to be clear, there are a few different pieces about this. Mythos is able to find similar vulnerabilities to other to other things. It is useful in the way that a lot of automated threat analysis stuff is. It does a bit more of it for a massive extra cost. For massively more expensive. Anyhow, Anthropic puts out Fable and it's more expensive for people to go, "Oh, pretty good. It's able to burp out some more open-source projects. It's really expensive. In fact, it's way more expensive than usual." And then Howard Lutnick comes in and says, "Okay, well, turns out that we got a report from Amazon CEO Andy Jassy who said that there's a way of jailbreaking Fable so you can use all the Mythos stuff that you're not meant meant to be able to."

[00:58:21]Dario Amodei, who was apparently Access Reports, was at a wellness retreat when they called him, which is really funny.

[00:58:26]>> [laughter] >> It's [ __ ] [ __ ] off just off in a weird tent somewhere. Like, "No, you need to go and do your job, [ __ ] no."

[00:58:34]But he got on the phone with them.

[00:58:36]>> Dorsey stuff. Every time there was a huge Twitter scandal and Jack Dorsey was like his outgoing message would be like, "You have reached me while I'm at a diarrhea retreat in Bengal." [laughter] >> Yeah, what what was it one of his things getting stung by mosquitoes? Am I misremembering that? Hell yeah.

[00:58:59]>> [laughter] >> What But what I love is that Dario Amodei was probably at Oh, yeah, I'm at a wellness retreat and he was just at like a [ __ ] Hilton.

[00:59:06]Just like a regular hotel just just watching TV. But anyway, so >> Like is there is there like actually like a dispute between Anthropic and the Trump administration over whether the US military can use these AI models for their fully autonomous weapon systems?

[00:59:23]Like to create Skynet basically, yeah.

[00:59:25]>> That was the other thing. So sorry, I should have mentioned that. Back in March, just before the war in Iran kicked off, Anthropic, out of nowhere, was like, "Yeah, you can't use Claude." The government, who had been paying for Claude since June 2024, had used them tons of conflicts. It was very much part of the system. "You can't use it for autonomous weapons, which LLMs cannot do, or I think it was surveillance of Americans, specifically. And they then got No, got marked as a supply chain risk, which all in all ended up doing very little.

[01:00:01]Like, they were going to get banned, but they didn't get banned. They were going to get removed from the government systems. They didn't get banned from the government systems.

[01:00:08]But, let's get back to Fable for a second, cuz it's quite simple. The government said, "Hey, you're able to jailbreak Fable to make it like Methus."

[01:00:16]And Anthropic said, "That's not that big a deal." And the government said, "Well, actually, that's a pretty big deal to us, so we're going to have to put an export control." Which means no US non-US citizens can use this inside or outside of America. So, Anthropic pulled them both down. To be clear, this is after two or three years of Anthropic's marketing being like, "Holy [ __ ] [ __ ] this thing's going to wake up, and it's going to be so scary. It's going to do going to destroy 50% of white-collar jobs." Poor [ __ ] Jim VandeHei, or whoever his name is over at Axios, gets like a heart attack every 3 months.

[01:00:49]But, there's an Axios blog saying that like, Anthropic's going to kill us all with their new models and all that [ __ ] And then the government banned it because they said it was too powerful.

[01:00:58]It says, "Too powerful is an export control. Unless you're American, you can't use Fable or Methus." So, Anthropic went, "We're pulling this whole thing down. We have no way of doing this." They claim they're working it out. Maybe they will, maybe they won't. Be funny if they didn't. But, it's just bunch of [ __ ] people bonking their heads into each other like, "Hey, my model's too scary. It's so scary. It's going to take everything from you. It's scary." And then the government's saying, "Well, I don't understand any of this, but I'm scared. I heard the word scary. I'm going to ban this now. Also, I don't like you." Like that was the big very big part of this well. Anthropic is also pissed off the Trump admin just by existing, I think. Just by being it they think they call them the woke ones. Like they I hate [ __ ] Peak Heg Sec. I dis- dislike the man intensely, but he did say they was it ineffective altruists.

[01:01:47]Uh >> [laughter] >> hung some points on them, which is unfortunate.

[01:01:52]>> Well, I mean like at the end of the day like the girl's school is still getting blown up, you know?

[01:01:57]>> Yes.

[01:01:57]>> Yeah. Like so someone's someone's doing their job.

[01:02:00]>> Yeah, they've and we don't have any real idea of where Claude has been used by the military, but I'm pretty sure it's not for anything more than this bunch of coordinates. What do you think, Claude?

[01:02:11]It's got >> I think Pete Hegseth uses for like marital and family problems, you know?

[01:02:17]Sort of ask Claude.

[01:02:19]>> Have another drink.

[01:02:20]>> Ask ask Claude how to win an argument with your wife.

[01:02:23]>> Yeah, it's it's really unfortunate, but also yeah, when you your entire industry spends years scaring people, scaring old Bernie Sanders who had that weird conversation with Claude. Very disappointing. Made me very sad when I saw that. And did you Did you see that video of Bernie Sanders That really That really He may I think he means well, but that was embarrassing.

[01:02:46]But it's the ultimate conclusion of no one building anything for real. No one actually doing anything. Just dancing around what they think a product might look like or might be good at and then hoping it becomes that magically one day.

[01:03:00]>> Well, it's like the the last one of the the great the great figures of our era, one of the great personalities, Elon Musk. Cuz like, you know, I was wondering if you could give us some perspective on the announcement of the SpaceX IPO, which will officially make Elon Musk humanity's first trillionaire.

[01:03:19]Um And like this this is this is yeah, a public offering for a company SpaceX that like, you know, it puts satellites into orbit and then like through and then through that like Starlink is becoming a major competitor to other internet service providers.

[01:03:35]Like, what what what what's the what's the juice here? Like, what what what's the squeeze here with SpaceX? Is this company Is this company worth a trillion dollars? Are they going to start like mining asteroids anytime sooner or is this just Like, I mean this this all seems uh quite fake to me as well.

[01:03:49]>> So, the overall thing with the AI bubble is that it's just proof of how many people in positions of authority and responsibility don't know a [ __ ] thing.

[01:03:57]Elon Musk claims his company has like a $50 trillion total addressable market.

[01:04:02]It's nonsense. Goldman Sachs or JP Morgan, I can't remember which one said their AI revenue was going to 300X by 2030. It's just It's all a bunch of people lying to get equities to increase. So, SpaceX is a company that makes rockets, has Starlink in there, has XAI, the everything app with Grok attached to generate, I don't know, non-consensual porn or whatever it does or answer the questions of our our world's oaths.

[01:04:28]And >> [laughter] >> it's it's ridiculous because this company is not worth that as we're speaking. It's kind of dumping. It's within five bucks of its its listing price. Who knows what happened. But, SpaceX is just exit liquidity for the venture capitalists and Elon Musk involved. It's so that Elon Musk can have more theoretical stock that he can use for loans. He's like Larry Ellison.

[01:04:49]He's Well, he's more like himself. He loves having He takes out loans on his various stocks. He obviously wants to merge Tesla and SpaceX into one horrible big company.

[01:04:59]I would In any other administration, I'd say that would be impossible, but sadly we're in this one.

[01:05:04]But, the thing is SpaceX has an AI division and they built they rushed to build like couple hundred megawatts of data center capacity faster than anyone else did, mostly by like breaking laws and buying a bunch of gas turbines that just belch horrible air into minority communities, but he has now rented out all of after buying all this capacity being like, "I'm going to train Grok to be the most base coding model of all time." Now he is selling all that capacity to Anthropic, Google, and someone called Reflection AI.

[01:05:35]And uh which mostly sounds like the fact that Elon Musk has mostly given up on the AI stuff. He just has it attached so that he can claim he's doing AI. He claims he's building he's He claims a lot of [ __ ] so that he can just keep the plates spinning. I don't think Musk is particularly intent on doing anything other than just creating more. I just think he wants more stock that he can sell that he can use to buy things.

[01:05:57]>> Well, I mean, he can claim that he's the world's first trillionaire.

[01:06:00]>> Exactly. On paper, but nevertheless still other-worldly amounts of money. No one could have that much money or even like a small percentage point of it.

[01:06:08]SpaceX is just part of the larger the larger plan to expand out his equity empire. I think he also wants to merge them so that he can get rid of the AI bits of xAI. I think he's realizing that's a mistake. But who knows at this point? He's He'll just keep [ __ ] lying. I will just say orbital data centers are not real.

[01:06:30]They are a [ __ ] theoretical concept.

[01:06:32]They're not going to happen. They're horribly No one's actually done it, and even if they did even if they succeeded I think that they are so woefully inefficient. It's It would take so much more effort just to get the get the computer to happen up there in the vacuum of space that it would it would cost far more money.

[01:06:51]>> Yeah, well, this is well, this is a problem of of that nobody wants their data centers built anywhere near human beings live, which is becoming I think a a a growing political problem on both the right and the left. It does seem to be an issue that does seem to unite Americans across a political divide is that people don't want these [ __ ] monstrosities being built anywhere near a human being might have to like hear them or breathe in whatever they're belching out.

[01:07:16]>> Yeah, I mean that's not why he's doing it. He's doing it so that he can just say he's doing more AI. He None of these people have a plan.

[01:07:23]>> That is the striking thing to me about all this, right? That the orbital data data centers um when Elon Musk will randomly say like you know, hundreds of years now we'll have interstellar travel uh with antimatter because of what we're doing now.

[01:07:40]Um the idea that like AI is self-aware, the marketing around it all being scary.

[01:07:47]This is the ideas guy economy.

[01:07:51]There is long been For people who aren't aware of this concept, there is a long-running joke among uh people who like video games that um you know, I just have such great ideas for games. I don't know anything about designing levels or coding or or or or uh any of the numerous other things you need to know to actually make a game, but my ideas are so good and I just If someone could just hire me as an ideas guy. But that is the economy now. Um trillions of dollars are randomly assigned as value to people who just have the most extravagant ideas without any actual practicality. I mean there is there is of course like an argument for there being like utility and value in SpaceX or any of these companies. It is if you can, you know, make semi-reusable rockets or reliably the deliver payloads into orbit at a cheaper and more at a cheaper cost at a more reliable rate. Yeah, that would be that would be economically viable, of course, but um this stuff like orbital data centers and [ __ ] antimatter engines, this is why the value is being assigned. And you earlier alluded to actual engineers and programmers being tortured and let go and all the The American economy is now just one big uh game studio with a thousand ideas guys who are all getting paid like $10 billion a year and three guys who work on the code and also test for bugs at the same time.

[01:09:25]>> Yeah, and I think that but all of this is kind like it's the ideas man I've thought about a lot as well because that's what's really happened. It's all of the actual people that created the value. They're all like, "Okay, what if the data center instead of being on Earth was in space and there were a hundred Brazillian of them? Wow, wouldn't that be good?" And all the other guys who don't do anything are like, "Yeah, wow, wouldn't that be good?

[01:09:46]We could have so many tokens." Yeah, I I mean, I have 15 sub agents. Well, I got 25. It all feels like the beginning of the death of Stalin.

[01:09:55]Whenever a new model comes out, they're like, "Wow, they did Wow, I can't believe all of them are fables." Like, "Oh, I absolutely I too." And they're all a bit worried cuz none of them really know what's going on. "Oh, yeah, it's uh so impressive." And I know I'll hear from someone after this saying, "Well, I've used it a lot and it's so they've changed my life and it's the coding too." Here's the thing, in absolutely no no other technological revolution have I ever had this kind of defensiveness.

[01:10:22]Other than like on game FAQs and people that argue PlayStation versus Xbox or whatever or like a particularly uh like a particularly overly religious in Nintendo 3DS fan. Like those people will go to war with you, but I never had anyone being like, "Why you don't use this the Moonlander keyboard at work, you [ __ ] [ __ ] I get so much out of this. Why you use that weird mouse thing that looks like a hand, you piece of [ __ ] I dare you to type that." And it's because no one really knows why they're doing this. There's no Everyone makes these extrapolations because it's all ideas men now. The people the loudest people about this aren't [ __ ] doing anything. Most of my work has grown out of the fact that a lot of people just didn't see whether this stuff made any money.

[01:11:04]Like a lot of my early work is just like is that true? And it isn't. It's never [ __ ] true. None of the ephemera around this is true. It's just people saying stuff enough times in the hopes that it'll happen. Except the people with the actual money and the resources are the ideas men.

[01:11:21]And they report to other ideas men. And the ideas men tell the people with the checkbooks just spend as much money on the [ __ ] GPUs or else where.

[01:11:30]The CFO of the average company is just like a kind of a whipping boy for the CEO. They're a marketing person now.

[01:11:39]We've reached the point where they're all doing the stuff that they think happens when things become big.

[01:11:46]But without knowing why. And without knowing that something has to happen in the middle that has to actually be really useful. And because they've never had any regulation.

[01:11:56]They've never really had any pushback.

[01:11:58]The tech venture capitalist on Twitter act like they're being electrocuted whenever they get criticized in any way.

[01:12:04]Like they These people haven't really experienced real problems. Thus, they can't solve any.

[01:12:09]And so they're just sitting around being like, if I just put more money into what I think looked like the last guy, I think it'll work out. Will that be good? Do you like that? Do you pigs like that? And because the valley has become poisoned by a cultish belief in just the the startup. The noble form of the startup. It's just It's people farting in each other's mouths in the hope that something changes.

[01:12:33]>> So, where is this going? Because like this all seems to me like it has like sort of a rhyme with like the subprime mortgage market.

[01:12:42]We're like we've made this the [ __ ] too big to fail. I think the people involved in it know like know everything that you're spelling out here like they're never going to [ __ ] make money doing this [ __ ] But the cost will eventually be socialized, right? Like is there a vent Are you seeing like a govern massive government bailout of big AI?

[01:12:58]The AI bubble?

[01:13:00]>> No. So, just taking a step back for a second, too big to fail didn't just refer to TARP. It didn't just refer to the government plan. It was the Federal Reserve basically opening the money fire hose to the banks cuz the commercial paper market that basically made the economy run. Long story, but the loans that people used basically fell apart.

[01:13:20]So, banks just their liquidity mechanisms were [ __ ] So, they just got fed hundreds of billions of dollars separate to TARP.

[01:13:26]And that was because had that done to be clear. There are arguments for letting the banks fail, but we can debate that another time.

[01:13:34]Nevertheless, they did that.

[01:13:36]So, the actual cost of too big to fail was trillions and trillions. Also, there was a point to it like you keep the banks because the banks will run out of money as we all know. We don't want the banks to run out of money. Then people won't have money. Then things won't happen in the economy. Banks need to be able to issue and receive They need to be able to issue credits. If they can't do that, the economy can't run.

[01:13:53]Businesses run on that. The rental things.

[01:13:56]If OpenAI runs out of money, I mean, other than giving me a lot of material, there is not actually a giant economic problem. The actual revenues of the AI companies are very, very, very, very small. The largest company The companies making the most money off of AI are Microsoft, Google, and Amazon selling compute to OpenAI and Anthropic. OpenAI and Anthropic, I think so far this year have made a combined less than $10 billion.

[01:14:21]They've raised over $200 billion. So, a bailout exists to stop something from happening both then and again.

[01:14:29]There is no situation to stop here. If OpenAI runs out of money and dies, which they've tele which the government has actually telegraphed they would allow to happen, being bailed out would not It would need a bailout very soon after. Unless we are just going to and I do not right fundamentally do not believe this happens. Unless we're going to just connect Open AI to the Federal Reserve, which will not happen, the government will just The support would not be there unilaterally.

[01:14:54]Then, well, they're just going to die. But, there's a much bigger problem, which is all of those data centers. There's over 100 gigawatts of data centers allegedly under construction. Right? Good, bad, who knows? But, that's allegedly happening. In truth, only about 15 gigawatts.

[01:15:10]Once those data centers are built, we're going to need 400 500 billion dollars of annual AI compute revenue just to keep them going. Right now, ooh, got about 70. And that's two The two largest companies, Open AI and Anthropic, make up 90% of all of that.

[01:15:26]And they are unprofitable. We do not actually have the compute demand to back up all these data centers.

[01:15:32]Now, this is going to become a problem in the next year or two because these data centers are going to start and they're not going to have any [ __ ] customers. Now, where are the data centers being funded by?

[01:15:41]Private credit.

[01:15:42]Private credit was the thing that stepped in to fill the void when insurance and retirement funds, after the great financial crisis, needed yields or needed returns so that their investments could keep growing.

[01:15:55]And the way they've done that is basically issuing a loans to whoever they want, valuing the loans at whoever in whatever value they wanted to. And uh yeah, private credit is funneling that money now into AI data centers. And who funds private credit? Pension funds, teachers' pension funds, police funds, firemen's funds, uh state pension funds, insurance annuities, all of that. That's what's funding that.

[01:16:18]We may very well need a bailout there, but I don't think people are aware of how bad it's got.

[01:16:24]Private credit's like trillions of dollars of loans issued privately with no underwriting standards that we're aware of.

[01:16:30]They are valued at a level that is just internal to the company. On top of that, Blackstone and all them, Blue Owl, they're not required to have reserves.

[01:16:40]They're They're required to do very much of anything. So, those are the companies funding the data centers. So, right now we're in this situation where the actual places that might need the bailout would be the data centers. And even then, I don't see the hunger to that for that, either. In fact, I see if they try and bail out data center developers, they're [ __ ] But really the king, and this is the one everyone's going to love, the one that everyone has to look forward to. So, Oracle and Larry Ellison.

[01:17:09]So, Mr. Ellison has been a very naughty boy. He's taken out about $21 billion worth of loans on his on his Oracle stock. Oracle's stock is continued growth is tied to the idea that Oracle will be able to build data centers.

[01:17:23]They're building $340 billion worth or more with a data centers for one customer, OpenAI.

[01:17:30]If OpenAI is not able to pay Oracle $70 billion a year by the end of 2030, Larry Ellison could genuinely run out of money. Oracle's stock will tank because they are betting their entire future on whether Sam Altman can pay them more money than anyone has ever paid for anything ever.

[01:17:51]It's so fun. I love thinking about this.

[01:17:52][laughter] I I think I think about this just to make myself laugh because Larry and people say, "Well, Larry Ellison could get a bailout."

[01:17:59]He actually couldn't because you can't bail out the value of a stock. Oracle might get broken up, but you could bail you could stop it from going insolvent, which they might have to because they have a big government contracts, but the actual stock value of Oracle is what keeps Larry Ellison alive and my man is leveraged.

[01:18:14]>> Well, we can we can only hope that his son will turn things around at CBS and CNN and maybe turn those into trillion-dollar industries.

[01:18:22]>> Here's hoping, but he'll watch.

[01:18:24]>> And so, gentlemen, we're going to leave it there for today. I want to thank you once again for coming on as our senior tech correspondent.