In short
Podcast Summary: The 1600 - Ed Zitron Dispels the Myth of the AI Revolution
Overview In this episode of *The 1600*, host Carlo Versano engages with Ed Zitron, a writer known for his skeptical views on the AI industry. The discussion centers around the current state of artificial intelligence, the economic implications of AI technologies, and the broader cultural narratives surrounding the so-called AI revolution.
Key Themes and Discussions
The AI Boom
Reality vs. Hype
- Comparison with Historical Technological Bets: The U.S. is making unprecedented investments in AI, likening it to the Manhattan Project.
- Skepticism about AI’s Potential: Zitron questions whether AI can genuinely solve societal issues or if it's a speculative bubble fueled by investment and marketing hype.
The Economic Landscape of AI
- Financial Viability: Zitron discusses the profitability of companies like OpenAI, revealing that they have burned billions without clear revenue streams.
- Market Dynamics and Speculation:
- Heavy investments in AI infrastructure (e.g., GPUs) are not leading to substantial economic returns.
- The stock market’s reliance on tech narratives (e.g., "number go up") obscures fundamental financial realities.
The Technology Itself
- Understanding Large Language Models (LLMs):
- Zitron emphasizes that LLMs are not conscious and merely operate on probabilistic outputs, not actual reasoning or intelligence.
- The technology's limitations make it less effective for complex problem-solving compared to traditional roles (e.g., customer service, coding).
Employment and AI
- Job Displacement Debate:
- Zitron argues against the prevailing narrative that AI will significantly automate jobs. He cites flawed studies claiming job losses due to AI without robust data backing.
- The discussion highlights that many roles still require human oversight and that AI's role in job replacement is overstated.
Marketing and Consumer Perception
- Consumer Awareness:
- The episode critiques how AI technologies are marketed to consumers, often with vague promises of revolution without clear functional benefits.
- Zitron references absurd AI advertisement campaigns (e.g., AI-generated images) that fail to resonate with real-world applications.
The Future of AI and Potential Collapse
- Economic Sustainability:
- The conversation suggests that the current AI investment climate is unsustainable, with massive losses leading to potential company failures, notably OpenAI.
- Predictions of market corrections as the "AI bubble" could burst due to the lack of viable business models.
Key Takeaways
- Skepticism is Necessary: Critical examination of the AI narrative is crucial, as many narratives are driven by corporate interests rather than genuine technological advancement.
- Economic Risks: The heavy reliance on speculative investments in AI could lead to significant financial downturns if these technologies do not prove financially viable.
- Cultural Impact: The ongoing discourse surrounding AI highlights broader societal issues, including misinformation, consumer trust, and the nature of technological progress.
Conclusion This episode offers a critical analysis of the AI industry, emphasizing the need for skepticism and caution as society navigates the implications of rapid technological advancements. The conversation urges listeners to consider the realities behind the hype and to focus on the tangible impacts of AI on both the economy and daily life.
For more insights, listeners can find Ed Zitron's work at [Where's Your Ed At?](https://www.wheresyoured.at/) and follow the *The 1600* podcast for ongoing discussions on technology and culture.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Transcript
Automatic transcript. May contain errors.0:00Right now, the United States is placing the biggest all-in bet on a technology since the atom bomb since the Manhattan Project. Is the artificial intelligence boom the greatest tech revolution of our time? Will it solve most or all of our problems, as they're telling us it will? Or is this whole thing a house of cards built on sand, burnished by speculators, and a White House that's in the pocket of big tech? Or is it both or neither? This country was stone cold dead. Good day from Newsweek World Headquarters at the top of the World Trade Center on a very rainy and gross day here in New York City.
0:33Today, we're doing a deep dive here on the pod about AI, the state of artificial intelligence, both as a consumer product, as we all know it through these LLMs, chat GPT, anthropic perplexity, whatnot, but also as an industry and an economic engine, basically the thing that's keeping the stock market floating, right? It's sort of difficult to separate the wheat from the chaff on this stuff when it comes to AI, right? The media is either pretty much in the tank for big tech or kind of useless, doesn't really understand the, you know, the implications of the technology. We have an industry that is spending billions of dollars on marketing and really just trying to convince all of us that this is the next big thing.
1:14And we have in the current administration, a collection of people who are essentially invested in the success of the so-called AI revolution with the hands-off policies to match. So some of the best thinking on this subject is happening on the newsletter platforms from independent creators, independent journalists, places like Substack, places like Ghost. One of those writers is Ed Zitron, who is, I would argue, probably the writer who is leading the way from sort of a skeptical perspective on the AI boom, the AI revolution to the extent that it is a revolution. Ed writes it, whereisyouredat.ai, and we'll put a link to his.
1:51Whereisyoured.at. Where's your Ed.at? Okay. You can buy a.ai. That's a good bit, though. I think that would really upset some people. Okay, I'm sorry. No, no, no, it's fine. I think that some readers would be very mad at me, which is always funny. But I'll put a link to Ed's newsletter in the description here. But Ed, welcome. Thank you for taking the journey across the bridge. Thank you for having me. As I was saying, you're writing a lot of great stuff on AI. Before we get to that, I want to – so my personal journey on this stuff is, you know, I started out as a doomer, right? I read the Nick Bostrom book probably 10 years ago called Super Intelligence, and it freaked me out, right?
2:32Like I think anybody who sort of did a deep dive into this idea of machine learning and this possibility that we could create something that can think and make decisions on its own and potentially not need us anymore, whatever that ends up meaning, right? So I started as kind of a doomer. ChatGPT comes out three years ago this month, I believe. I moved to being a little bit inquisitive, I guess you would say. I was even a little bit bullish, right? I saw some of the applications of this, like it was useful in my job as a journalist in terms of like doing some research. You still have to check its answers.
3:04But if I needed to ask, you know, tell me the five things that Trump has said about this person, like that was an application. That was a use case for it, right? It would be a little bit harder to do that on Google, but whatever. But three years later, I've turned. I think I'm in full crank mode on this stuff. Right. I'm with you. I think that it is I've said in the newsletter that it is the without exaggeration, the work of the devil. And I mean, I'm sort of serious about that. Like, I think that if you asked Satan, right, if you asked Lucifer to build a product that was meant to exacerbate sort of like all of our social plagues, all of the, you know, the problems in contemporary life while simultaneously, you know, atrophying our brains and making our children stupider, he would probably come up with something like a large language model, at least in my sort of hyperbolic opinion.
3:56No, I'm with you on that one. I think that to broaden from there, I have this thesis, the rot economy. The markets are controlled by growth at all costs. Thinking isn't even about value creation or even anything real. It's just number go up. Right. So I think large language models are the perfect. I don't know if I'd go as far as satanic, but nevertheless, far as market hubris goes, the idea of there being this company, NVIDIA, which everyone sends money to, everyone meaning big tech companies and private equity firms, to sell GPUs that when you plug them in, immediately start losing you money.
4:28Right. Every single company, everyone plugging them in loses money. But the market says this is the best thing ever because number go up. The markets you see CNBC and the like, it's Google's AI bet pays off, despite the fact that Google has never disclosed AI revenue. Right. This is all black box. It's just pure hubris. It's when you when the watchman stop watching, when nobody watches the watchman, when you have companies that build things that are kind of abstracted away from any kind of value creation or helping human beings. And then you have a product that's really hard to describe or define and impossible to really benchmark in any reasonable way.
5:07You can look at the benchmarks they use for these models, but the benchmarks are made to make the models look good because you can't have them do things or tasks. And even I read the software benchmark, the one that they use for the coding models. Even that one is basically broken now because the models have just started looking the answer up versus trying to fix the problem. the reasoning. It is really funny because people are like, oh, that just means they're smart. No, it doesn't. It just means that they found a way to con their way. It's just because they've been trained to a test. Like anything, like our society, you train people to pass tests versus why do people go to college?
5:41Oh, it's to learn, right? No, it's to finish college. Right. Get your degree. Everything's about the end rather than the journey. But I mean, I think you make a good point. First of all, I think a lot of people don't understand. These things aren't conscious, right? ChatGPT is not conscious. It's not a It's not a thinking machine. It is a probability machine. What is it actually doing? When I type in, you know, show me a picture of Godzilla with Trump's head on it. Nice. What's it doing in the background? So there's a process called inference, which is just the process of it creating an output.
6:10I will fudge the answer about what a transform model is. It's basically extremely complex. Talk to us like we're children about this. Yeah, it's hard. I'm trying to remember it myself, but it's very good at comparing different things and parameters to come up with an answer. But nevertheless, the process of inference creates the output of Godzilla shooting a gun at Scooby-Doo. I don't know. And that process is extremely compute intensive. Weirdly enough, MIT Tech Review revealed middle of last year, it's less compute intensive to make images than texts. I was shocked by that too. But 80 to 90, again, MIT Tech Review, there were 80 to 90 % of GPUs, of GPU compute is being used for that process of inference.
6:52Now, there is also the very expensive computationally intensive training where you feed information into the models and then you post train them by saying, we like this answer. We don't like this answer. All of that stuff is happening and has been happening for three years. But really, we're at a point where I'm not sure why we're still doing this. You can look at this as a better search engine. I don't agree, but you can look at it as that. You can look at it as a companion thing. You can look at it as, I don't know, a way of brainstorming. But however many hundred billion dollars at this point, should we not have more?
7:29Because these things, to your point, don't think. They calculate. And when they say thinking, that's just instead of you saying, give me a story about Scooby-Doo with a gun, it doesn't just plop it out. It says, OK, I'm going to break the task down into this person is probably referring to beloved cartoon animal Scooby-Doo. What kind of gun would Scooby-Doo use? AK-47 perhaps, and breaks it down to steps and then spits out an answer based on those steps. That's reasoning, which they call thinking. It isn't thinking at all. Which is why if you were to do that, if you were to put in that prompt and I were to put in that prompt, we could get two different things.
8:07So look, yeah, I mean, I think that we're, to your point, we're being sort of sold a bill of goods that has no upside for those of us, those of us in the proletariat, right? Those of us who have to, you know, work for a living. All I see is that this is technology that is going to, you know, further bifurcate us into sort of like the, you know, the ultra rich people who are heavily invested in this and the rest of us who it's probably going to take a fair amount of white colored jobs, at least at the at the sort of baseline. I fundamentally disagree about jobs. So every one of those AI taking job stories are bollocks.
8:42They're nonsense. There was the one that came out a few months ago that said, oh, 10 % of AI affected jobs have seen declines in employment. One of the positions within that was accountants. Right. Now, any accountants listening will know there's been a problem with people becoming CPAs because it's a very difficult process. It's kind of a, I think it's an interesting job, but it's kind of a boring job for most people. People aren't becoming accountants because it's difficult to become an accountant. Right. The incentives aren't there. And it's been a good career path. Yeah, it's been a good one, but it's still difficult unless people are doing it.
9:12It's been a problem for years. But nevertheless, part of that study was saying, oh, these AI affected jobs. You look to the recent one where it said, oh, 10 % of jobs can already be automated by AI. Again, a PhD, this was on CNBC. It's on the television. It was 10 % of jobs. And it was something like a PhD student's paper. So not even like a peer-reviewed journal article. And on top of that, it was like, yeah, these skills can already be that. These skills. And it's got like enough asterisks to fill Barry Bonds' entire career. Like it's ridiculous. But these AI job stories are always fudging the numbers.
9:49Like there was the Oxford one from a few months ago, which said, oh, yeah, it was something like 14 % of AI of jobs. Jobs are down by 14%. I forget the exact number. But the actual statistic was a single line item. It was like, yeah, we think this is the case. because when you actually look at the practical side, it's not doing it. They can't replace jobs. Now there are ones like translators. Brian Merchant of Blood and the Machine, he's done a great job on this. Translators, so people just, jobs from like Rev.com or what have you, they're seeing their jobs cut because of AI. That's because a lot of translator jobs, people think, oh, I can just do the quickest version.
10:28Actually doesn't work that well. It doesn't work that well, but nevertheless, those people have already seen. There's some sort of customer service jobs, right? Like when you text, I had a text with the bank the other night. I assume I was talking to a robot there. You might have been talking to a bot, but like knowledge management systems, the basic bot services that you've seen, like Erica with Bank of America, for example, those aren't LLMs because you can't rely on an LLM, to your point earlier about the stories. You can't rely on them to do the same thing twice. Now, perhaps that's not so bad when it's like, give me a list of names for a cat.
10:57But when it's move money from one account to another, or how much money do I have so I can make decisions with my Merrill Lynch account, you don't want to lead that to probability. I mean, you had Knight Capital back in 2012. That was like a weird algorithmic thing with hundreds of millions of dollars and destroyed the hedge fund. That alone should scare away anything. And that wasn't even large language models. Kind of budging the answer. Nevertheless, the thing is, you can't really replace many jobs with this because while we think so little, understandably, of customer service jobs, translator jobs, perhaps, you might think, oh, these jobs are menial.
11:38We don't really need. Actually, we're kind of demanding as consumers the things we need. We need them to be done right. And we kind of need them to be done right every time. And so when you look at the AI job story, oh, coders are being replaced with LLMs. Is that right? Has anyone got any actual data on that? Because it does not exist. I've spoken to numerous software engineers, Carl Vogey, Carl Brown, Nick Suresh, like real software engineers. And Carl Brown of the Internet of Bugs actually put this way, he said it, those coding LLMs, the ones that write code, they make the easy stuff easier and the harder stuff harder.
12:12So it means that if you have some quick and dirty things, sure. What a surprise. Software engineers are using something to speed up their workflow. The story of Silicon Valley. But you can't just be like, build an entire software package. Because you can't really rely on it to do it. You can't rely on it to build stable, reliable code. There's this thing, VibeCoding. What is that? I've heard of this. It's fraud. Right. It's fraud. Anyone selling VibeCoding is a fraud. Vibe coding is the suggestion that you as a non-coder, you can't write code, me can't write code, you can use an LLM to build functional, reliable, secure software.
12:52And those are three very different things. Functional, meaning it works, and it works every time. Reliable, meaning it stays up because there's a whole infrastructural layer on every website, even the simplest website in the world. Something like Newsweek's website, for example. You've got just the CMS, so the thing that gets the stories up there. You've got the video uploads. And then you've got the masses of ad-targeting cookies and such in there on top of it. There's a bunch of stuff on the back end that you just don't know about. And it's secure as well. There are people who just dick around trying to break stuff.
13:29They're massive groups of hackers, and they sound kind of dramatic, but it's true. They just kind of f*** with a website. They're going to get in like this. it's why you have massive breaches all the time right so if you've got some vibe coded website and someone goes in there and goes i wonder if this yep it'll break something that you would know as a software engineer you would know oh that's an that's an elementary thing i'd never get that wrong not everything's elementary to an lm but also nothing is they don't know anything they don't have knowledge they're replicating it's kind of impressive what large language models can do as a scientific thing right when it comes to actual applicability and reliability can't really do much of anything i think that's sort of the problem with like how it it's being forced on us or marketed to us the consumer product of these llms when it's like i mean i it seems like there is actual value in the the technology that is machine learning i say this is a neophyte like You know, you read stories about like I just read something about a and who knows, maybe this is a scam, too.
14:34But, you know, there's some some company built an algorithm that can detect, you know, strokes earlier than a regular doctor could. I theory it. But no, but that's that's a really important point. That's not large language. Right. But the marketing within AI, they really want to combine useful. That's cool. Right. It's cool. I love. Yeah. Who's against early stroke detection? Right. I'm not an anti-AI person. I'm anti-LLMs. and I'm anti the marketing and the nonsense and the lies and the GPUs and all that. But there's cool s**t that's been happening for years in AI. Way before this, that's AI.
15:08This is something entirely, I mean, it technically is machine learning, but it's not the AI that we've been sold for years. Right. I want to come back to the marketing, because I think that actually that is a big sort of, you know, strategic mistake that the industry made. But speaking of the industry, you know, one of the reasons that, you know, My thinking is flipped on this is because of the reporting you've done. You had a great piece a month ago about OpenAI, right? OpenAI is the Sam Altman run, the big, I don't even know, is it a public company? No, it's a public benefit corporation now.
15:39Okay, whatever that means. But your piece was all about its cash burn and just about how much money this company is spending every minute. Tell me about what you found. So, forgive me for not having the numbers in front of me. So for the first three quarters of the year, they burnt$8.67 billion just on inference. So just the process of creating the outputs. The revenue story is incomplete, but based on the revenue share, I think they made 4.3 billion something. This is the revenue share they have with Microsoft. So I extrapolated through the documents I saw because Microsoft shares, sorry, they share 20%, OpenAI shares 20 % of their revenue with Microsoft.
16:20Right. So I had the 20 % from Microsoft, extrapolated from there. That's a good deal for Microsoft. It. What are they bringing to the table? So their entire infrastructure. So OpenAI and Anthropic are rich kids. They live on their rich parents' infrastructure. OpenAI hasn't had to invest anything in their infrastructure. I see. Okay. So when you hear about - OpenAI's parent in this case being Microsoft. So Microsoft invested multiple times, 2019, 2021, 2023. And I think they involved themselves in some 2024 rounds. But part of those deals were the OpenAI gives them 20%. They own all of their IP.
16:55They share all their IP, all of their research. So everything that OpenAI has, Microsoft has. And in fact, Satya Nadella was boasting the other day because OpenAI is working with a company called Broadcom to build specialist chips, which the information reported would have modest gains. That's what you want to hear. And Satya Nadella was on a podcast boasting that we, him being Microsoft, now have those designs. Sam Altman thinks he's this smooth operator, but he's being worked. But as far as what Microsoft gets from this, I mean, they're losing money on everything. The information reported middle of last year that Microsoft is providing OpenAI's compute at cost.
17:32Now, I don't know if that's still the case, but at cost here means Microsoft's taking a loss because GPU compute is not profitable at all. It's really not. Well, that's a big point in the reporting that you've made. I think maybe we should clarify that. Every time that you, the more you use chat GPT or whatever, perplexity, the more money they lose. Correct. Right. The more the more they have to reason, the more the models have to do this, the reasoning or the inference. Yes. The more energy it's taking and the more compute and the more money they lose. Correct. How is that different from, you know, say, you know, any of these sort of like tech 2.0 in Amazon or an Uber?
18:10Right. Amazon lost money, a ton of money for a while. And then they sort of stumbled upon their business model with Amazon Web Services. Right. So that's it. That's a common myth. So I actually dug into this because I've heard that so many times. So in totality, I think it was over nine or 10 years, Amazon, pardon me, spent about$69 billion to build AWS. AWS was cash flow, I think, break even or cash flow positive, I think, within three years. Amazon built AWS out because they kind of had to because they had a massive operation, which is very heavy web traffic. But nevertheless, massively cheaper.
18:44I think if you look at OpenAI, if you include Microsoft's CapEx, you assume that maybe half of that was OpenAI, because really, those GPUs, I have sources at Microsoft, those GPUs are not, Microsoft's AI services are not popular. Most of it's going to OpenAI. So, OpenAI's infrastructure is probably over$100,$150 billion just walking around. That's just, they didn't have to pay that. Amazon Web Services, about$69 billion to spin up. But the process of running a website, so if we look at Newsweek, that wouldn't be using a GPU. You'd be using CPUs, a ton of RAM. They'd be specialist enterprise CPUs.
19:21But talking meager bits of sense to run. It's not costing several dollars an hour just to keep a website. Oh, perhaps it is, but that website is making billions of dollars. The process of a user buying something on Amazon is relatively cheap. Every single thing someone does with ChatGPT is pretty expensive. And the other problem is, is the, so like an Uber, for example. Uber since 2019 burned about$33 billion. They're kind of profitable now-ish, right? They're profitable now. Yeah, but they're EBITDA, so earnings before the bad stuff, to quote Rob Smith of the FT. But it's, Uber is also very different in just the amount of money they burned, but also Uber was useful from the beginning.
20:02You could point to Uber and say it. The business model makes sense. Like it's from day one. Right. Right. It had an obvious sort of like point to it. And it was also a ton of money in R &D on their failed autonomous operation and a ton of marketing. Kind of like how Groupon, if you remember Groupon, heavily marketing based, they were just taking a loss on everything. But back to the infrastructure thing, because it is kind of simple. Facebook, for example, it's expensive to run because they have billions of users. But for one user to dick around on Facebook, tenths of a set, tiny amounts of money.
20:34A user who is a power user of Facebook or Instagram, who spends hours and hours, probably isn't costing Facebook that much, or Meta, that much money. The terrible thing about large language models, other than like the environmental damage and the stealing, is the most passionate users you have also cost you the most. Because the more someone uses it, even if they're paying you 200 bucks a month, because OpenAI loses money on that subscription. which is crazy. The more passionate they are, the more tokens they're going to burn, the more GPU hours they're going to spend. And on those more expensive packages, they're using the more expensive models, the more complex models, which are more expensive to run and more computationally expensive to run.
21:19There's a company called Augment Code. Augment Code is an AI coding platform. They had a customer, they had to change their pricing fairly recently, who paid them 250 bucks a month, and they spent around$15 ,000. Anthropic. They don't teach that in business school. Yeah, you don't know. This is the big brain business stuff that you and me just don't get. Yeah, we don't get it. Anthropic has this thing called Claude Code, another AI coding platform. So from that, you pay, I think, 200 bucks a month and you have unlimited with some rate limits. There is a, because you can extrapolate from your use of Claude Code how many tokens you're burning, people what is a token so a token is kind of a weird way of referring i should have explained that i apologize so when you one token is i think three quarters of a word okay but so basically you pay or they on a million per token basis there are input tokens so what you load into the model so if you're loading in documents and the outputs are whatever the model has to do to give you an answer okay so that means in the reasoning case an output token would be the the thought process the breaking down the task and what it gives you so the more complex the model the more output right right more expensive so tokens are just the units that they used it's the way the way it works i don't know why they designed it that way but nevertheless clawed code users there is a thing called vibranc which is a ranking of how much people are able to convince it to let them spend end of the month, they're someone who spent$50 ,000 worth, 200 bucks a month.
22:54This is the thing with large language models. They don't fit into the paradigm of how the tech industry charges for software. There are pretty much two ways they do it. They're either free and ad supported, or they are a monthly subscription. For the power users, right? Yeah, yeah. Except you can't really say, you can't stop someone from spending more than$200 worth of compute. You just can't do it. That's because a lot of these, I mean, every model, this is crazy every single ai product you use is just prompt engineering it's just they have a system prompt that specialize that they say for this operation you are a coding model you will do code or you are you are going to think through things or in the case of like a small model you're going to give quick and easy answers every single one is like that but you can't unilaterally say stop them when they've spent this much right because there's no way to stop them because that would cut an operation short and also i don't know if they know how i don't know if it's possible and this doesn't mean that they've lost control of the models this doesn't mean that just means that they don't know how to actually program them to do stuff because that's large language models baby that's probability right everything's probability it's yeah i'm so glad we built our built our markets on this it's so good well so that i mean i don't know what the statistic is But there is some statistic about how much of our GDP right now is going into basically AI CapEx.
24:18That's the reason why the S &P is up 17 % this year is essentially because of these seven or so magnificent. What happens when the music stops? Well, there are a few problems. So first of all, there was this stat that came out, I think it's reaffirmed by a professor as well, where it was AI CapEx and spend on data centers accounted for more of GDP growth in the first half of this year than all consumer spending combined. That's incredible because our entire economy is predicated on consumer spending. which is what's crazy though is nvidia claimed about a month ago that they've shipped six million blackwell gpus that's the newest generation now i went and did the maths on this about six gigawatts to 12 gigawatts of it load which just means power we haven't built that many data centers we haven't even built a quarter of that so not only is the biggest company on the stock market only getting there because of selling GPUs, but I'm suspicious as to whether those GPUs have gone anywhere, whether those GPUs are, they're not in data centers.
25:26The data centers don't exist. The power doesn't exist, but the physical data centers don't either. So 6 million of these bloody things holding up our markets, they're sitting in warehouses. Sachin Adela of Microsoft said he was sitting, they've sitting on warehouses, didn't say how many. And I think that that's the really scary thing. Cause at some point we're going to have to realize that everyone's buying these gpus and they're not going anywhere right so everything sits on this when the music stops it's going to be because nvidia's growth stops and the market has an unhealthy relationship with this company they grew i think not four trillion dollar company now right it's like 4.3 i think it was it looked yesterday it's like based on selling gpus that are likely sitting in warehouses it's like a fin dom relationship jensen one calling sachin a dayler at 3am send me three thousand dollars you pig it's so strange but what i think it is is that jensen huang ceo of nvidia has just kind of bullied everyone it's like no if you buy as many of these you little pigs right but at some point the little pigs are going to stop oinking and i don't think that we can just afford to do this forever because nvidia is on a yearly cycle meaning there will be a new GPU every year.
26:40And this new generation Blackwell is more power intensive, requires liquid cooling. So you can't just put it in a regular data center. So people have been retrofitting them and you've got data centers being built. But at some point, something more is going to have to happen because no one's making money doing this. And the other problem is that when you build a data center, it's horrendously expensive, horribly expensive. And this is before you get to like power interconnect and all that, but just the building, the GPUs, the cooling. There's just the cooling for the GPUs, and then there's the server architecture.
27:14Then there's HRAC units that go on top of all this insane. And there's no, it's not like these data centers create a whole bunch of jobs once they're up and running. It's basically, it's just a room. Yeah, right. But the idea of these hundreds of megawatts, gigawatt data centers, it's very new, very new. But on top of that, really expensive, billions in advance. And then you plug the bloody things in, And then what? Because no one's making a profit. At some point, someone's going to work that out. Lots of people are going to work that out, and they're going to stop buying these GPUs. But Jensen Huang claims that there's 20 million GPU orders for Blackwell and then the generation after that, Vera Rubin.
Read the full transcript
27:55Where are they going, Jensen? Where are they going? And when I bring this up with analysts or other journalists, they go, oh, there's plenty of power. There's plenty of power. They get very upset. Plenty of power. and it's just taking away all the fanciness where are they going where are these gpus going how are they being turned on and then who's using them because the other problem is is okay our entire market's based on selling millions of gpus doesn't that kind of make them a commodity at this point what's special why why use a gpu i mean they have you have custom silicon with like google has tpus and Amazon is trading them and all this, but everyone's buying these GPUs.
28:35At some point, isn't everyone selling the same thing because everyone bought the same thing? Right. And also the same thing loses money. So it's just this insane, illogical economy. And it seems to be based on the idea that this will turn into something one day. Everyone wants this to be, oh, it's just like dot-com boom, just like emron it's not like emron kind of like lucent i was going to actually bring up lucent because lucent was the great example the lucent big telecom company back in the 90s early 2000s right i think then they went bust but they were they didn't go bust wins so connected entities did okay right but but but the point about lucent was that they they were sort of the the godfathers of this idea of like circular financing right where these guys are all just loaning essentially loaning each other the same pile of money.
29:27So NVIDIA is better at this. So the big one with, I just was reading about this yesterday. So the big one with Lucent, they did a bunch of this, was the vendor financing thing and was with one called Winstar Communications where they lent them. Right. This big telecom build out and Lucent saying, we will lend you the money to build. $2 billion to build out your network. And then Winstar ended up going bust because they couldn't afford to pay because it was great. It was a$2 billion deal with$100 million worth of guaranteed revenue. They don't teach you this in business school. But NVIDIA is clever.
29:57NVIDIA isn't doing that. NVIDIA is not doing Enron stuff. And by the way, NVIDIA had, NVIDIA putting out a, to sell-side analysts, they sent a four-page PDF to analysts saying, we're nothing like Enron, we're nothing like Lucent, we're nothing like WorldCom. Mate, no one. Don't start putting that out in the world. Don't say Enron, please. no so what nvidia does in is they don't do okay there is one thing within this four page thing nvidia said some words which were meant to be read one way but do not they said we do not disclose any vendor financing arrangements people are meant to read that to mean we don't have any that is not what that says that says we do not disclose any right that doesn't mean they have no anyway but Nvidia, let's just say they're being honest.
30:51What they are doing is these companies called NeoClouds, CoreWeave being the largest one. CoreWeave is a company invested in by Nvidia. Nvidia is their largest customer. They paid them for, still paying them in a deal called Project Osprey, guaranteed them$1.4 billion of revenue, which they just raised to over$6 billion. They'll buy any unused compute. Nvidia was a major anchor of CoreWeave's IPO. And then here's the great part. So what CoreWeave does is they bought a bunch of GPUs and then they use those GPUs as collateral to raise them to buy more GPUs from NVIDIA. Right. This is just the snake eating its tail.
31:29Yeah, except what Jensen's real smart. Jensen hasn't had to do, I mean, they've had to buy some CoreWeave stock, 250 million, whatever. That's how much he screamed at Satya Nadella to buy last night. Like it's 250 mil is nothing to him. Okay, that's a little bit jealous. But nevertheless, they've done this again. They did it with Lambda, another company where they, I think they invested in one of their rounds. They have become one of their customers. They'll probably help them with their IPO. So NVIDIA isn't doing the we're loaning you money thing because smart. That's a bad idea because you might not get paid back.
32:05Instead, what they're doing is they're creating these unstable debt vehicles. Corweave loses hundreds of billions of dollars every quarter, I think. Do any of these companies make money? No, not a single. This is the thing that tries to be insane as well. I've been saying this is early 2024. Nobody is talking about the AI revenue. I take it back, actually. One company, Microsoft disclosed their AI revenue in the last quarter of 2024, in the first quarter of 2025, and then they stopped disclosing it. And everyone's like, that's fine. It's all right. Sorry. Must be good. Must be. Must be. Yeah, they're just shy.
32:36They don't want to show. They don't want to make anyone feel bad. But no one wants to talk about the revenues. Open AI, June of last year. sorry this year they said they have 5.5 billion annualized revenue so 5.5 divided by 12 but they're not going to do the math off the top but nevertheless it's everyone's doing this funky mathematics right and let me tell you something when things are going well you don't have to do numerical alchemy right you don't have to do like don't look at the man behind the curtain shit but no one's making money everyone's losing money every single company every single one of them and you ask people you ask journalists you ask analysts you ask tv hosts you ask them all how are they going to do it and they're like they'll work it out they're this close i in your life in my life i don't think i've ever been given this much leniency in anything right ever not as a child not as an adult and these fuck nuts they burn so much money and that's this chip like right now google stock's kind of running because of gemini 3 everyone's like wow gemini gemini 3 is really this is their latest llm i can't find a single bloody person who can tell me why I can't find one person who can go, this is the reason.
33:44And this is what Gemini makes Gemini 3 better. There is right now, there was a story of the information, the information ran, where OpenAI is on a code red because of Gemini 3. I think it's all theater. I don't think any of these people, I would love to see inside OpenAI, because I have to wonder, like they work very long hours, but I don't know what they're doing. I don't know what anyone's doing in there. because when you read this article it was great it's like yeah open ai is putting ads on the backbone now to work on making people want to use chat gpt more making chat gpt's answers better and making chat gpt able to do more to which i ask what the f have you been doing the last few months well i mean i did a stint in big tech and the dirty little secret is most people in that industry are not doing anything yeah they're just they're just sitting around i don't summer camp I say that lightly because they're working, I assume, but I think it's all kind of kayfabe.
34:41I think everyone's just going around being like, yeah, we're working on the most mysterious, powerful thing you simply wouldn't understand. I can't. OpenAI has these insane layers of security between rooms and buildings. So funny. To hide what? Right. To hide. What are you hiding, Sammy? And it's nothing. And I sound like I'm being a little bit aloof, but I don't know. I've had to prove myself quite a lot in the last year, and I'm happy doing so. I love doing it. But when you look at what they're actually doing, why are we? Well, Altman, he's a salesman. He's not a tech. I mean, he's not. Can he code?
35:18No. Right. Zuckerberg hasn't written a line of code since 2006, by the way. People think he's a code. No. It's all management consultants. They've all got MBAs other than Altman. And Altman, Amadei, so Anthropic, and Zuckerberg don't have MBAs. But Sundar Pishai, Google. Good point. Tim Cook at Apple. All these guys. Andy Jassy, Amazon. They've all got MBAs. They're just management consultants. They don't know what to do other than buy stuff and hire people or fire people. Jack Welch, baby. Well, so this actually leads into – I want to go back to what we were talking about with the marketing, right?
35:50Because the consumer-facing stuff, I think you're starting to see a backlash among the plebs, right? I don't know if you – We did a show about this a couple of months ago now, I guess. But there was this big ad campaign in the New York City subway a couple of months ago for this thing called Friend, which is like a little necklace you wear that you can talk to. I don't know. I mean, everyone is so everyone's like defacing the ads like you don't need this. you know the commercials the stuff that we are are I mean the new commercial on the subways I noticed today was a Google ad for Gemini that the the hook of the ad is imagine clowns building the New York City skyline and then it's like generates an image of clowns making Rockefeller Center like who's asking for that no it's I saw a tweet a while back which is great more is like every AI ad is like, Jim and I, what should I have for lunch?
36:39Right. And he goes, sandwich. And they're like, it's just the best one. Do you remember that this was a meta ad from a while, a couple months ago, maybe, maybe longer. It was all, it was unavoidable. And it was, the concept was this girl who had to ask the meta AI for help with topics and questions for her book club that she was hosting about Moby Dick. It's like, you can't read the book for your own book club. That's just the thing. It's always like, oh, it helped me come up with questions for, do you not have questions up here when i interview someone i usually have a few right use your brain a little bit well just beyond just what do you do just really simple do you not have question it's not even like going beyond just using your brain just if you're interviewing someone you're not curious about anything but thinking of these funny commercials so there was a gemini one during the super bowl commercial where sorry was that the one where the guy asked the gemini to to write the interview it was yeah yeah one so and it was this this man who was doing a job interview prep with gemini and it was vague what he was doing because it was clearly talking about his daughter growing up and being a father but then he went on a job interview and it was just gemini was like yeah well i think you should work on this answer more wow thank you gemini and then he went on the job interview his daughter's in her like late teens early 20s as well and he's old It's like when you are forced back into the workforce because your health insurance, because Medicare doesn't exist anymore.
38:08Gemini will be here to do something. Something for you. Because that's the thing. What does it do? Right. What does it do? You ask someone what you do with Amazon. You buy stuff. It arrives at your door. What does Uber do? Get you from place to place. Well, DoorDash do deliver food. Right. And in all three of these cases, destroy labor markets. Treat human beings like ****. But nevertheless, you can do something. You can point to the business model, right? Yeah. Business model or even just application. With this, it's like, oh, what would it look like if clowns built the new? No human being who has used something has actually thought in those ways.
38:46Like, I guess there are people who sat around and be like, what if this happened? It's like they're looking for a use case, but why not? I guess because these companies are consumer facings. But I was going to say, like, you know, why not advertise the ability of these large language models or the machine learning to do early stroke detection or some of the applications in the health and pharma space? That's because consumers can't buy that. Right. Also, the reason they can't explain what it does is they don't know either. Because this is the thing. This is one of the largest f*** ups in history.
39:17That's what I actually think this is. It's completely new. Not quite.com, not quite Lucent, not quite Cisco, not quite this. it's just i have this theory called the rock-com bubble which is tech has run out of hyper growth ideas we had smartphones we had tablets to an extent and within this we've had smaller things like mdm and so but nevertheless you've had little industries cloud computing is the big one and we got so much out of these massive companies built on the back of not just running the cloud but running the bits of the cloud right software running the cloud huge thing i think we're at of ideas but we're out of ideas there's a plateau yeah and also i think we're out of ideas that are worth hundreds of billions of dollars i just think we're out of them so tech kind of tried they tried smartphones that kind of worked really worked for apple worked pretty well for google you had some success elsewhere you had microsoft kind of up so just right shove that in the they made a phone once yeah smartphones smartphones were good we had tried wearables didn't ever do it for me yeah it's just fine i i thank you for like it's fun i i get i for sports there are useful things i like my aura ring but is that a multi-trillion dollar industry no metaverse we didn't really try that one no we really did whatever happened with that that was zuckerberg's big thing like five years ago not even so funny that we just memory just like that nope not doing that from the beginning i called that 2021 we tried that we tried clubhouse tried social audio we even tried more social media didn't really work and you see the current state of social media where like going on facebook is like an escher painting of ads and like you have to click you have to solve a riddle to make it show you the chronological feed instagram is 99 just ad insane direct consumer things and so you see them kind of google search barely works barely works and that's because of the rot economy everything must grow forever they're just squeezing everything they can out of it but they got no more ideas and they've chased out all the engineers all run by mba so all these are basically just private equity firms the only thing they know how to do is buy lay off sell buy more squeeze yep and of course squeeze so they all chat gpt comes along surprises everyone including the board of open ai because sam allman just didn't tell them and microsoft is sitting there satya nadella true story sitting there goes i need that in bing buy all the gpus and i'm deadly serious.
41:49That's why this happened. And because AI is this magical thing, this marketing term where you could justify everything. It's great for the hyperscalers. They have something to do. The CEOs have something to buy. They now have an AI strategy. This trickles down to middle managers who don't do anything. They can now talk about AI. They can now pretend to do stuff. AI superficially looks like work, just like every great middle manager. And these companies are chock full of them. It gives them tons of stuff to do. You've now got product teams who can build AI thinks. You've got investors. Where do investors invest?
42:19AI. They can make money buying and theoretically selling. Problem with investing in AI companies, though, is they all kind of do the same thing and they all are running out of money and they don't seem to have any path to profitability. Can't take them public, can't sell them, don't know what to do. Putting that aside, it gave freaks on LinkedIn something to talk about. It gave CEOs outside of hyperscalers something to talk about, something to do. It gave everyone the ability to pretend that there was something futuristic happening. And because so much money was going into it, no one really had to worry about justifying its existence.
42:54So the tech industry, and I mean, my evidence here is they would do anything else. If they could do something else, you think they want to spend$30 billion in CapEx, that's what Microsoft's doing, a quarter? They don't make it. At this point, I did the maths, big tech because the amount of money that these gpus cost the amount of money they put into data centers and the amount of this there's a thing called depreciation basically there's only so long you can use a gpu for before it either dies or becomes obsolete right they have until 2030 to make two trillion dollars of new revenue not total revenue i mean new ai revenue otherwise it was all right and this is the thing they're doing this because everyone needed something to do and they came up with one the problem is no one thought does this make more money than it costs and i get it for the first year or two i totally get it like you know what in and it's everyone trying to recreate the past everyone trying to say well in the past uber cost a lot of money amazon web services cost a lot of money.
44:00So if we spend lots of money, we'll make that, right? That was sort of my devil's advocate question of like, you know, what do you say to the people who are just saying, look, we're in the early innings of this, right? This is, you know, we're in the top of the first and this is going to go into extra innings. The web was like this too. There was a shakeout in the dot-com bubble, you know, but we still use the internet. And like, you know, the cost of not competing for guys like Zuckerberg and Altman is higher than whatever the cost of all of this capex is the thing is that's myth it is myth the dot-com boom and the dot-com bust if we look at it as the telecommunications bubbles so the cisco's loosens wind stars and all that and you look at it as the websites you have two things there the websites the dot-com websites that went bust they didn't go bust because they were not useful they can't they went bust because they were too useful like the idea of buying something online made sense immediately the problem was was like shipping something that costs five dollars that cost you twenty dollars like the math wasn't great those unit economics you could eventually balance them out right amazon as amazon did as amazon did the blues and all that that never went into the green that fiber wasn't even useful a lot of it like they really hasn't been useful at all like that alone is a myth but on top of that nothing we're doing resembles either of these things.
45:23Jim Cavallo of Goldman Sachs, middle of last year, did a great piece, I think it was called Gen AI, Too Much Spend for Too Little Return. And he made a point where it's, yeah, you had to buy a bunch of these Sun Microsystems servers to make the internet run. But the amount of money you spent on them, minuscule in comparison. But the internet already proved itself. And in fact, there was a Newsweek article that people love to quote, by the way. This one saying the internet was a fad. And what's great about that was he was actually correct about several things. This thing about education and like the online education was a good idea.
45:55It's correct. Putting all that aside, it doesn't resemble it. It doesn't resemble it at all because with the internet back then, you could still say you can buy stuff online. Perhaps you could say the businesses don't work, but you could point at it and go, this is the future. Right. Pets.com, maybe you don't need, but the concept is real. Chewy is an incredibly successful website. Like there are ways of making these businesses work. if there's a way of making an LM business work I'd love to see one because we are enough and people say oh it's only three years it's all the king's horses and all the king's men and all the money and all the media attention all the analysts and human on twitter saying AI is the biggest thing ever take him on back and saying Nvidia is going to go up forever you've got all the hype men all the money all the attention you still can't do it you still can't do this and people can try and say all they want oh yeah it's the early earnings not saying you here but it was early when how long how long do they get right how long do they get because we are running out of money i mean just flat out the data centers that are getting built are being built using private equity money that is running out we are the lot the data center deal that mark zuckerberg is doing have you heard about this hyperion this is complete this is the that when you hear it you're like oh we're so they are creating a joint facebook yes of meta a hyperion data center a joint vehicle where meta owns 20 of it and blue owl and various others are funding it with billions of dollars it won't be built until 2029 so like four more generations of gpus but Nevertheless, it won't be built until then.
47:37And then at that point, Meta is leasing it so they don't have to have it on their balance sheet. And then, this is the great part, they've guaranteed it. They've guaranteed that they will lease it. They don't own it. So they've just signed up to a really expensive apartment for four years in the future. 2029, we're not going to be talking about this wank. This is going to be blood and guts by then. but a really simple thing is when there's enough money and a good justification of something you don't have to like get the runes and the spells out you don't have to get a shaman you don't need a wizard to conjure the money up and create some obtuse debt vehicle if it if this was the future why isn't it on your balance sheet mate and the answer is meta makes no money from ai zip they all it does is they make money from ads just like they have been for whatever and they're claiming oh we've seen 5 % increase from generative AI ads.
48:38Bullshit. I'm sorry. Show me actual numeric. Like money. Money. Show me the money. And they never will. They never will. They changed how they, Meta changed I think late last year for how they measure active users. It's like daily active people now. They move the shit around. It's like how Netflix does it with like the watching hours or something. They all move, it's all fake. It is. But on top of this, it's all fake, but used to buy because all these companies microsoft meta google they used to be awesome because awesome financially i guess because their cash flow was just there and they didn't sink hundreds of billions of dollars into gpu they weren't asset heavy now they're asset heavy businesses software is meant to proliferate endlessly right not be this that's a good point right and now they're just attached to a bunch of gpus they microsoft owns hundreds of thousands of old generation GPUs.
49:33There's a thing called an impairment. When something loses its value, you have to mark it down. One of those is coming for all of them because they all own A100 GPUs, H100 GPUs, the ones that came out in 2020 and 2021. And those are just basically paperweights now. They have some value. They still work, but not for long. And also, if they still work, why are you buying the new ones? It's not really clear why anyone's doing anything. Let me ask you this. Is it possible? Could OpenAI conceivably go out of business? I mean, it's hard to say that about a company that only employs like 3 ,000 people and it makes$8 billion in revenue.
50:12Well, yeah. I mean, well, again, it tells us it makes$8 billion in revenue. Well, they leak. Yeah, they could absolutely go out of business. OpenAI claims they're going to – they've signed up to$1.4 trillion worth of compute deals now. Unbelievable. ball they have played they've agreed by the way nvidia everyone's saying nvidia is going to invest in open ai and nvidia's earnings they revealed it was a letter of intent with the option to invest right okay my mate said that my mate said i might invest one day money's never going but nevertheless open ai agreed to build six gigawatts of data centers for amd and 10 gigawatts worth for broadcom first of all two and a half years and 50 billion dollars a gigawatt just off just so tired of that But hundreds of billions, they've promised hundreds of billions of dollars, but their compute costs alone are meant to be in the hundreds of billions within the next few years.
51:06They don't have the money. And also, are they meant to just suck venture capital's dollars forever? And people say, oh, Uber did that. No one has done this. No one. SoftBank's meant to send them$22.5 billion. SoftBank, big Japanese company corporate monstrosity. Fun fact, they had to fund open... They were the guys behind WeWork. Yeah, they're one of the big funds. Yeah, and also Greensill Capital. Okay. Just a straight up fraud operation. It's so cool. It's so cool. But even better than that, so to fund this$22.5 billion, SoftBank has done really well because they made a really good bet on two companies.
51:45Alibaba, like very early, big Chinese e-commerce company, and Arm, big chips company. The chip company, yeah. They have had to take a margin loan out on their Armstock. and also sell all of their Nvidia stock just to give OpenAI their next funding round. And apparently a story came out, this Masayoshi-san, I haven't confirmed this, Masayoshi-san, CEO of SoftBank, cried when he sold his Nvidia stock. That's how you know it's good. That's how you know everyone's doing the right thing. You know you're in a good relationship where you're crying every night and you're bankrupting yourself. SoftBank's at the end of their money too.
52:19They have an agreement with their shareholders that they won't raise the amount of money they borrow above a certain threshold. And we're like, I think it's like 25%. I forget the exact thing, but we're near that. So OpenAI has really drained the venture capital. I was about to say swamp, but that means something else now. But the other thing is, is people are saying, well, what about the Souths? The Souths, right? Yeah. The Southie Public Investment Fund is running out of money. They're having to stop investing for a while because they've made so many bad calls. Right. Yeah, when the Saudis have run out of money.
52:56That's the thing. The Sauds are out. Everything that NVIDIA is doing is funded by private capital. It is private capital flowing through big tech or bond sales going to NVIDIA. And at some point, it's not being funded by revenue. So why the f*** are we doing it? I mentioned the possibility of open AI going under because I noticed that Altman recently said something about how the industry, or his company at least was too big to fail. And once you hear somebody start using that term, that to me is just like a major warning sign. They're not too big to fail. They're too small to pull into enough pieces to wait.
53:33There's too many people who are going to be angry at them. They're not. The systemic risk they have is to Oracle. Oracle has bet their future on OpenAI being their biggest customer. So if OpenAI goes under, maybe it pulls under an Oracle or – I think – I mean – But it wouldn't have an existential – it's not like J.P. Morgan going on. No. I think what we will see is OpenAI goes under, and it is a symbolic and philosophical shift. It's a – people will want – they're already trying to – That's the bubble popping. It is. And it's also – it could be something – the bubble might pop from something smaller.
54:09It could just be a scandal. It could be who knows at this point. But they'll try and shift to Anthropic. But Anthropic is just going to get absorbed into Amazon or Google. Like that's what people at Amazon Web Services have talked to about Anthropic. There's nothing special about what Amazon offers Anthropic. There's nothing. They could move to another company and they just, they're signing a big deal with Google. But nevertheless, everything about this is predicated on the money people being willing to shovel hundreds of millions or billions of dollars. And for no reason, it's not based on revenue growth or anything like that.
54:42It's based on. Number go up. Number go up. But fairy dust and magic. In the future, this will control everything. Can't even automate stuff. It doesn't do. Every time they sign a big customer, it's a f***. It's a project. It's a experiment. There's a term, experimental recurring revenue, where it's like, we're going to see if this works for a bit. And then it usually doesn't. And the thing is, all of these layoff stories you've heard, oh, it's for AI. It's wank. It can't replace people. It's not replacing people. Well, it seems to me that a lot of what that is is companies using AI as a fig leaf to be like, we overhired during COVID.
55:18Exactly. And we have got it. We want to cut costs. And this is a great opportunity and a great excuse to do it. And there was an R &D tax credit as well that changed. I forget exactly. 172, maybe. There was a code where it was how you were able to amortize expenses. I think you have to realize them in the quarter now. You can't deduct them in quite the same way. And that dramatic... I've spoken to numerous hyperscaler accountants about this. That affected a lot of things. So, yeah, they've been... Everyone was parking people during 2021. They were hiring everyone they could and just putting them in a room without a door handle on the inside and being like, see you in eight hours, your stock vest in a year.
55:54And it was just to make sure they couldn't hire. That shouldn't be legal either. But it's, yeah, it's the massive, I think what happened, I think 2021 broke the world in many ways. But 2021, there was this massive growth that happened because you had everyone inside and everyone went spending. Everyone's numbers went up. And I think they've all been trying to recapture that magic ever since. And they've now, it's like a cursed amulet situation. These GPUs are driving them mad, burning all their money. Really, it's a classic story. It does feel like that. Ed, before I let you go, I want to go back to Google for a minute.
56:28Sure. Gemini. Mickey, if you're in there, can you bring up this picture? This is a photo. I guess we wouldn't call it a photo. This is an image that was generated by, I don't know if you would consider this Gemini, But it's Google AI using a program called Nano Banana 2. Yeah, this one was good. Now, I can't get up close to the pixels, but. Yeah. So the woman on the left, both of these women are fake, right? The one on the left is sort of like it has all what these AI images have been looking like. You can kind of spot them, right? They look a little bit shiny and they've got like a little too perfect.
57:04Now, the one on the right is also an AI generated image. Again, that woman does not exist. that to me i can't tell i think i'm pretty good at this stuff i cannot that to me could be a real close enough to be a dick about it here's the thing i always say about this how many prompts how many prompts did that take because with sora you can get some reality adjacent stuff with nano banana you can too how many prompts did it take to get there because you can cherry pick from these things but every time i have gone to diddle around with these things can't really get It's something that high quality the first time.
57:38It takes a good amount of times. And that's the thing. Sure, they're going to end up doing realistic stuff out of how many tries. Also, what does this end up actually doing? Well, I think what it ends up doing is that it craters any shared sense of reality that we still have, does it not? I mean, if you can't believe your eyes, if we get to a place where anything you see on the internet, somebody could conceivably be like, well, that's not real. Hey, I generated that. doesn't that strike you as an existential risk? We already think we're already there. We've been there for years. Yeah. We already had consensus reality breakdown numerous times.
58:14That's true. COVID did that. It was already kind of fringe already. Something like that. Okay. How reliably can you do it? What can you actually guarantee that we'll do? Because it gets down to the core problem of large language models, which is can you guarantee that it would be able to do that photo doing exactly the thing you want? Sure. There's going to be, there's absolutely going to be deep fake stuff. Deep fake stuff already. happened sure you're going to still see you're going to see stuff like that happen you're going to see these images but to what end what is the the idea is scary and this is actually the real marketing pitch of ai it's like well imagine if this happened okay has it happened yet well what if propaganda well they're already doing a propaganda trump administration's already doing propaganda with ai so i'm not sure oh they'll do a fake people have already been doing that you've like this is already happening so yeah it's not great but at the same time the ship has sailed on yeah the ship has sailed but also to what end how reliable what are people going to do with it when it starts happening okay i think i actually think that by the way the first one of those that actually causes problems they actually the platforms themselves are going to be in more trouble than they want to be because this is how you start up section 230 right this is how you start getting into a situation where platforms...
59:33Section 230 being the thing that keeps all of these companies from legal liability for things on their platform. Like a platform cannot be, like something created on a platform, something on a platform, you can't sue the company. I don't know, this feels like a really good way to get that repealed. Not that I think that should happen, I'm just saying, that feels like it too. Also, we have no idea how much it costs Google to plonk out that. And even though aforementioned MIT Tech Review thing says it's cheaper to do an image we have no idea about nano banana pro whenever you're pro by the way think expensive right and because all of this is so heavily subsidized like vo3 again vo3 you get 10 seconds of wow that's pretty interesting looking try and get 15 or try and get the same bloke to do the same thing in the next shot which is how all that statement is built sure you're gonna get like deepfake things but you've had people able to do similar to that okay for a year there's an article from Sarah Young and The Verge showing how you could edit in bomb-related stuff to photos.
1:00:31It's scary, but what's actually happening as a result? What are people actually scared of? Like, is it consensus reality breakdown? You already have millions of people who believe that ivermectin could cure them. Right. You've already got anti-vaxxers. You've already got reality breaking down in many ways. You've already got people being fooled by Photoshop. Right. You've already got Mimars watching AI slop. I mean, what next? Like, it's not reliable enough to make a consistent, I mean, anyone trying to do a consistent deepfake with that is probably already doing them. I guess it's sped up their workflow.
1:01:06I mean, I guess, yeah, just to leave us on something of an optimistic note, because this has not been an optimistic conversation, but very useful. Yeah. I guess, but maybe, you know, maybe there's a place where all of this AI slop, it just, it gets to a point where everyone's just like, you know what? I don't want to do this anymore. Like, I don't want to, I don't want to be on my phone. I don't want to be scrolling all day. Like if I can't, it was one thing when I could, you know, lust after pictures of, you know, the girl I wanted to date or like see my friend's baby shower photos. But if it's just crap, it's just slop.
1:01:39It's just fake stuff. Maybe people are just like, I don't want this. I actually think that that's where we're headed because it's two problems. It's the slop problem, but it's the thing I talked about with the raw economy. It's this growth at all costs thing where every product has got worse. This is the end shitification. Yeah, we had him on. And he's a f*** dog. He's going to join me in CES 2026. He's an amazing man. So every platform is getting worse. And then on top of this, you have Mark Zuckerberg saying, actually, I want more slop on the platform. Yeah. At some point, people just got to go, I mean, I don't use Facebook anymore.
1:02:10No, do I? Why would you? I don't get AI slop on Facebook, by the way. Yeah. I'm being gang stalked. They know. But people go on their feed. They don't see the things they want to see. They see ads or they see slop. Right. They go on Google search. they don't get what they want to see right they go every platform they use is actively antagonistic i think there is a good chance that this actually exacerbates an existing problem where people are hired at the tech industry 10 years ago we loved them we love the apps we loved them we love the apps folks we have folks we love google um ben gomes uh but we're in the situation now where the evil management consultants are just crushing the life of everything and they're like hey guess what you know how all your sucks and it stinks you can't find what you want we made a magic machine that creates more right in the way we've made we made an obstacle machine what do you think oh you hate it what if i told you it was really expensive and lost me a lot of money but that's the thing and i think that this exacerbates that problem but also no one has an answer for what happens when the bubble bursts when the bubble bursts it's not just going to be wow, companies lost a lot of money.
1:03:20OpenAI is dead. Something has shifted because we have now seen a few things. This is, I know it doesn't sound, it is optimistic. We've now seen, first of all, how many f***ing wits we know. How many people just like, I know quantum physics now, chat GPT told me. We've seen how many CEOs are just straight up horny for laying people out. Finally, I can automate everything. I've wanted to do this my whole life. We've seen how many people we know who don't actually know very much. What's the other thing you can tell? I at least can spot AI writing from a mile away. Two seconds. Right. It's not just big.
1:03:53It's huge. Right. You got this. And it's just shining a light on how many CEOs don't know what the f*** they're talking about. How many rich people are just wrong and stupid. How many people are really disconnected from any good source of information. I don't mean regular people. I mean all the people going, AI is here to stay. Is it? Yeah. And I don't mean this against you, but please. How many, I think around 25 people have said to me, it's like Uber. These are people who are like established analysts and tech journalists. It's like Uber lost a lot of money and now it makes money. But that's the thing.
1:04:27We've learned how many people, why am I the person who wrote the, hey, Amazon Web Services was this much? Why me? I'm a, I'm a PR person, blogger, podcaster, manticore situation. but i like it's shown how many people just don't do their homework and don't know stuff and it's also shown how many powerful people are just disconnected from the things they make i think it's a radicalizing event i think it could push people away from tech and it could also punish the tech companies in a way they which they deserve because they don't have they have a way of keeping people on the platforms they don't have a way of bringing them back that's a great point that's a great point let's leave it there actually i want to think more on that that's um Ed Zitron, thank you so much, man.
1:05:12Much appreciated. Tell everybody where they can find your work. I know I misstated your website. It's totally fine. Betteroffline.com. Betteroffline.com. And Better Offline is the podcast. Where's your ad at? It's in the newsletter. Gotcha. Okay. Thank you, everyone, for watching. Please feel free to follow Ed. In fact, do follow Ed. Follow us. Let's get these numbers up down here, folks, so I can stay employed and leave us a comment and tell us your thoughts on what you think of the whole AI revolution. Is this thing, you know, where is this going? I don't know. I feel better about it now, though, Ed.
1:05:47Thank you. My pleasure.
From the publisher
How profitable is OpenAI? Is the bubble about to burst? Ed Zitron of "Where's Your Ed At?" pulls back the curtain on the AI industry.
To watch this episode, visit: https://youtu.be/Z5Pl9FxHZBQ
Check out Where's Your Ed At?: https://www.wheresyoured.at/
Subscribe to The 1600 newsletter: https://link.newsweek.com/join/79a/the-1600
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.