Season 1 Episode 5: Did Bernie Sanders fail the Turing test?

[Nick]
Welcome to this episode of The Last Enclosure.

[Mike]
I’m Mike.

[Nick]
And I’m Nick.

And welcome to our little revolution as we attempt to, I don’t know, how would you put it, Mike? We are basically fighting against the modern enclosure movement, right?

[Mike]
So historically, landowners and capitalists were enclosing the commons that people used to live. And now big tech and other large providers, big pharma, whoever your big boogeyman of the day is, are trying to enclose the very resources used to live.

So our goal is to educate you about the myths of the day so you can understand the ways in which the world is being enclosed. You can fight back against it in your own way, in your own life. We’re giving you the knowledge to dismantle the fences and the gates and enclosures surrounding society.

[Nick]
Well put, Mike. And of course, that very last of enclosures is the place we should be the most free, the human mind. In that vein, let’s get into today’s topic.

So if you wouldn’t mind, join me for an imaginary moment. I want you to picture Bernie Sanders, the great socialist grandpa of modern American scene, in a closed-door meeting with one of my nemeses, personal boogeyman, however you want to put him, Mr. Sam Altman. And as we’re recording this, this is July 2026.

And only slightly more than a month ago did this closed-door meeting happen. Afterward, Bernie went on to explain that in the meeting, he was pitching, he was describing a policy that he would like to enact to mitigate the damage that AI could cause to our nation, our economy, our democracy. And it was a very Bernie-style proposal.

Basically, that a sort of one-time tax haircut would be taken from the big tech companies, specifically the ones developing and implementing AI. And that this money, in the nearly billions of dollars, depending on how it slices out, would go to a kind of sovereign wealth fund. So again, very Bernie response, very classical sort of power to the people kind of policy to a very new problem.

And the question is, how did we get to this point? How did we get to a closed-door meeting between Altman and Sanders? And what does this mean?

How should we feel about all this? Let me put it that way.

[Mike]
Yeah, I mean, I don’t know if you remember, but in our last episode, this will be episode five. So episode four, we were talking about the AI bubble and there being four types of bubbles. Towards the end of the episode, I talked about Paper Tigers and Kayfabe, right?

And proposals like Sanders, effectively, you know, while they come from a good place, they come from a place of wanting to help people, right? Deal with what is believed to be the fallout of lost jobs and, you know, basically getting in front of the harms of the industrial revolution, as it were. But in doing so, if you believe the harms are going to be real and that you believe the value is going to be real, you’re basically pricing into existence the valuations that OpenAI and Anthropic are proposing that they are worth.

And then you are then putting the taxpayer on the hook for maintaining those valuations, right? And so it actually is, you know, while Bernie’s proposal is coming from a humane place, it actually is a double-edged sword. And it’s one sword that is coming just in time to cleave away the negative press and the down valuations that these companies are facing, right?

Their investors are getting, you know, more anxiously nervous about getting returns. And here the U.S. government says, oh, we think you guys are worth about a trillion dollars. Why don’t you give us a, you know, 5% tax or 50% tax in Bernie’s case?

And, you know, by doing so, by taxing them at that rate based off of their equity, their valuation, you’re kind of actualizing the dollar value these companies say that they’re worth, right? Which is insidious. It’s insidious because that’s what they want.

And it’s no coincidence that this month, right? So Bernie’s proposal was a month ago. He’s been talking about it for a while.

We’ll get into that in a bit. But this month or last week, I should say, Sam Altman himself was saying, hey, you know, Bernie’s proposal is a bit tight for us, but we’ll give you 5% equity and we’ll put the money towards a sovereign AI fund for the country. How does that sound, right?

So it should be a moment of pause for us all if Bernie Sanders and Sam Altman are, you know, it should be a moment of pause if they are kind of having overlapping policy worlds, right? Sam Altman believes that giving the government money based off of a stake of his company is a good idea for him, right? Now, you know, him and Bernie are arguing whether it’s 50% or 5%, right?

But Altman clearly wants this idea of the company, OpenAI, his company being valued by some third party at a particular valuation. Valuations are very important in the finance world. Financialization obviously is the increasing role finance plays in our markets, our world, right?

And there’s a lot of financial voodoo that goes on that us regular folk don’t really understand. But the very beginning place where that all starts is capitalization. I’ve talked about this on my blog, Misaligned Markets, and I won’t bore you with the details.

One of my favorite writers, Blair Fix, says capitalization is the magic ritual of finance, right? So you draw out a number, you say, based off of, they have very complicated financial equations, but they kind of boil down to vibes. Based off of historical data and some other stuff, my company or my asset is worth this.

And the minute you draw a dollar value that someone else believes, you can go to a bank, you can go to some other investor and say, this guy down the street, he says he’s gonna give me a million for this. What about you? And remember, if you don’t give me at least a million, I’m gonna walk away to that guy down the street, right?

And so this idea of putting up a number that the company can be then taxed by, even if the tax is sort of a loss to the company, the valuation obviously is a prize, right? And so it is not a coincidence that Altman himself is now just kind of approaching this story and saying, hey, this tax thing, it’s fine, right? We normally talk about companies getting ahead of regulation by trying to propose regulation themselves.

That is what’s happening here. But the icing on top, the cherry on top of the cake here is that getting a valuation from the US government by the equity tax is going to allow OpenAI to go to investors and say, we are worth a trillion, whatever the stake is, we’re worth this, right? And that is undeniably very valuable detail for OpenAI.

[Nick]
It’s even invaluable because they’d be getting the sacred imprimatur of the federal government and therefore the American taxpayer on the hook to essentially, and let me introduce this term into our conversation, elicit a bailout should they ever get into trouble.

[Mike]
Yeah. So the other thing, obviously, the valuation is the real insidious part. But on top of that, what you just said is true too.

If the company, if the government, sorry, has equity, I’m talking about our country as a company. If one company gives another company that is a country, there’s equity. So if our country, definitely not a company, if our country has equity in OpenAI, we’re not just interested in their ability to pay us, but we’re interested in their material well-being because obviously we have a stake in the company.

And so like any stakeholder, we’re going to begin favoring that company. Because it is an investment, it’s an asset for us. And so that does things like distort the market.

So we’re going to have favoritism towards OpenAI. We’re going to want to protect OpenAI from losses. Their problems become our problems, basically, if we go down this route.

Now, Sanders is wisely saying, hey, a one-time 50% tax-based, but it’s equity-based and he wants voting rights. We’ll still be stakeholders, even if it is a one-time tax. So there is this issue of not being sufficiently distanced from these companies, not just because, you know, hey, look, they’re bad companies.

But fundamentally, the problem here to me is that the things that they’re promising, these are kind of regulatory issues, right? So the job losses, et cetera, if they are real, which we spent five episodes now saying that they’re not real, right, on the scale that they are proposing because the technology is not suited for purpose. But even if they were real, we have tools to address this.

We have regulation. We have other types of legal and social options on the table that I think the- The Fed has levers of action.

[Nick]
Congress has things it can do. It has public, it has fiscal policy.

[Mike]
Right, right, right. Like, this is sort of, I mean, even if Bernie’s heart is in the right place, this is so draconian. It’s draconian in ways that, again, you know, benefit these companies.

Timing is kind of- Not a coincidence?

[Nick]
Without getting conspiratorial?

[Mike]
Not a coincidence. Yeah, and I’m going to go through. I have my own experiences with, I think, where Sanders may have, not maybe where he got it, but kind of what has nudged him along this path.

But yeah, I mean, as I was saying, this timing’s weird. Timing’s bad. This is draconian.

These may not even be the right tools to address this problem. And now you’ve got industry on board with these exact same solutions, right? So Sam Altman, not to his credit, but Sam Altman is saying, hey, I’m proposing this 5% equity investment, not just for me, but maybe all my other AI friends get on the grift too, right?

So Altman’s trying to propose this as an industry-wide kind of policy to, I guess, legitimize it. I don’t exactly know what he’s thinking. Maybe he thinks having more companies in on the bandwagon will make Bernie lower his 50% number.

I don’t exactly know what the logic is, but it does kind of create an air of legitimacy if more companies are getting on board for the bailout, which it definitely- I said last episode that there will be no bailout. This is the one form it would take. As I said in the last episode, hey, this is kayfabe.

Do not fall for it, right? And unfortunately, Sanders has fallen for it. Ironically, David Sachs, Trump’s AI czar, he was saying, hey, look, these companies, they’ve been going around scaring the shit out of people.

Call this a stupidity tax and feed it to them. I’m almost of the mind that we should make it happen, even though I hate what this will do for markets. It’s just like, you can’t go around screaming at the top of your lungs, my technology’s dangerous.

My technology’s dangerous, and not expect people to take you seriously. And Dario Amadei and Anthropic themselves, independently, had kind of thrown their own hacks in the ring and said something similar. They have their own proposals that do not involve equity, interestingly enough.

But they’re saying, hey, regulate us, tax us, of course, on our terms. But adding to the narrative, again, the role that these actions play in legitimizing these companies and their insane valuations.

[Nick]
And also, as a point of difference here, as a point of comparison, Dario’s proposal is conditional. Like, if unemployment reaches some threshold, I think 5%, whatever was just announced, then these triggers will fall into place. If there’s already catastrophic damage to the economy, then we will do something.

Yeah, yeah, yeah. I think it was Anthropics.

[Mike]
Anthropics, right. Yeah, I mean, Dario’s proposals are of a similar vein. They’re sort of just kind of gesturing vaguely towards something should happen slightly before the bad things happen or right as the bad things are happening, right?

So I don’t really think there’s a sense of narrative control here. But interestingly enough, in my own experience, kind of alluding to what I was talking about at the top of the episode, I have been seeing Sanders increasingly talking to people in the industry since the beginning of the year. And I’m going to kind of spin you a story.

Imagine me as Charlie Day in sunny Philadelphia with the conspiracy board. I don’t want you to think of this as a conspiracy per se, but I’m seeing linkages between Sanders’ proposal and relationships in the broader, I would say, AI ecosystem that have me kind of scratching my head, right? So let’s start at the top here.

So on February 16th, an article released by Transformers AI, there are a little sub-stack that is ran by people who are kind of in the, I would say, an effective altruist adjacent space, mostly focusing on AI news, a little bit of AI safety stuff too. Safety, of course, is sort of the hobby horse of the people that went into companies like OpenAI and Anthropic to protect us from an AI that will kill us all or cause immeasurable harm, et cetera, right?

[Nick]
And look for our deeper dive into the AI community coming up. But this is one example.

[Mike]
Yeah. I mean, we alluded to some of this in episode three, and we were talking about the buckets that people go into, right? So these guys are in bucket one, right?

They think AI is super capable, and people should be talking about it with more respect. And they’re fearful of it. Well, they’re fearful and or positive about it, depending on which side of the coin they land on, right?

[Nick]
So someone like Eleizer Yudkowsky – Which side of the bucket? Yeah, which side of the bucket they land on.

[Mike]
That’s right. So Eleizer Yudkowsky would be like, we’re all going to die. Someone like, let’s say, Holtz is his last name.

I forget his first name. But someone like Holtz is like, oh yeah, this is so good. It’s going to be so capable.

He’s a tech bro. And he’s just like, yeah, AI is going to be so capable. I’m going to be able to finally escape humanity, right?

So anyway, Transformers AI, little outfit, does AI safety news and AI technical news. They write an article called, The Left is Missing Out on AI. It blew up all over my feed.

In fact, Ed Zitron was sharing it. That’s how I learned about it. A lot of our colleagues were talking about it.

Yeah, Taylor Lorenz, another tech journalist, she was sharing it. So my feed was blowing up with this story. And this story freaked me the fuck out.

So one thing I noticed that no one mentioned is, actually, let me start with the framing of the article. So The Left is Missing Out on AI kind of sounds very intuitive to a certain type of person who is very online. So if you go online, there are a lot of people who are left-leaning.

And they are very openly so, who are vocally anti-AI. And some of these people go as far as to say, if you ever use AI, you’re immoral, et cetera, right? They’re literally a part of the great anti-human machine.

Right, right, right. I mean, they’re not coming from the Doomer, Yudakowskian view. They’re just saying, hey, a machine that plagiarizes and lies to people and causes psychosis and has all these environmental harms- And sucks up water and does all of the things.

[Nick]
Right.

[Mike]
That’s bad. And you should not be supporting that, right? OK, it’s more reasonable than this thing will devour all the matter in my cone, right?

So the article basically says, leftists, they are talking about AI incorrectly because they’re underestimating the rate at which these capabilities are increasing. They, ironically, of all people, take a quote from Cory Doctorow, who later that exact week was shat on by leftists for actually saying, I use AI sometimes and it’s not useless. So these guys very clearly at Transformer AI didn’t know their audience, but that wasn’t the point, right?

They got a quote of Doctorow kind of saying, oh, AI is spicy autocomplete, which we have talked about that point on and off again here at this podcast. I won’t litigate it again. But directionally, I think it’s correct, right?

Not true, but directionally correct. But they go, there are two types of leftists, basically. Doctorow, who in that one quote was kind of not deferential enough to their AI god.

And then Bernie Sanders is the other leftist they mentioned, who they’re saying is taking AI job loss seriously. He’s taking AI capabilities seriously. And so the left, if we’re going to have a dialogue about AI, we need a left that is actually aware of the benefits and dangers of AI, as Senator Sanders is, right?

And so he was kind of the voice they were pointing to as the reasonable voice on the left about AI. There are only two leftists in the world, Cory Doctorow, who again, actually, his AI position was not covered in that article because he has the reverse centaur thesis, as we talked about in previous episodes, which is way more complex than it’s spicy autocomplete. He also, again, that week got shot on for using AI.

And then Sanders. So there’s only two leftists. One of them, they didn’t even bother to look at.

The rest of the article actually was dedicated to slamming Emily Bender, who is, she was actually mentioned, Nick and I saw a movie last week, a documentary about the history of AI. So Bender was actually in the documentary. She was one of the voices in the documentary talking about AI.

She’s a linguist. She is one of the people who coined the term stochastic parrot, right? So these language models, they don’t actually understand anything because they don’t have minds, et cetera.

They are trying to just repeating what they’ve heard in training, right? And so that makes them kind of, if you’re relying on them like you would a human, kind of a dangerous principle because they’re just stochastic parrots. And so most of the article actually is attacking Bender vis-a-vis her being not deferential like Senator Sanders is.

And actually capability, regardless of the stochastic parrot thesis is scaling, right? The reason I bring up Bender is because there actually was an article in Vox. It’s the voice of the hipster left.

Yeah, of course. And so Vox actually has an effective altruist. They have an effective altruist.

I guess you’d call it like a section of their outlet called Future Perfect ran by a local Bay Area named Kelsey Piper. Okay. So I know a lot of names, a lot of dates.

Sorry. Kelsey Piper wrote an article like almost a year to this day called AI’s critics are not taking AI as seriously as we need them. I don’t remember the exact title, but it’s something like that.

And in the article, guess who Kelsey Piper is attacking? Emily Bender. Again, vis-a-vis criticism of some vague left.

I don’t even know what Bender’s political affiliation is. I mean, I assume she is because she’s concerned with these sort of like AI ethic question about is AI fair, right? I would assume she’s left leaning, but I don’t know if that’s true.

And it’s not exactly relevant to her stochastic parrot point, which is about the actual function of the technology, right? But again, we have EA people attacking people who are not kind of talking about their precious, precious AI light cone devouring God in the right way, right? And I don’t think it is a coincidence that these articles named the same person in the exact same way.

And they basically are like, hey, look, this person who is like not deferential enough to our view of AI, we got to address that so that more people can be deferential to AI. And oh, hey, look, Bernie Sanders is talking about AI in the exact same ways that we want people to be talking about them in, right? Well, talking about them in the ways if you are not on board for the technology, right?

So they want critics of technology, but critics who fear the technology in the way that they approve.

[Nick]
In the same way that they do, right. Right.

[Mike]
Well, some of these people are not, they’re not actually, like, I don’t know if the guys that transform AI are, so some EAs are AI doomers. They are Yudakowsky and doomers. They’re like, we got to stop this.

Some EAs are actually AI is so good, it’s going to solve all the problems in the world, right? Again, bucket one, which going back to episode three, please listen. Bucket one is people who are true believers.

There are true believers either in the direction that they are like AI is capable and AI will do all the things it says it will do, right? And that is exactly what they want, right? They want to control the narrative.

They want to control the discussion in ways in which, you know, they approve. And so what they’re kind of doing is they’re putting their thumbs on a scale. They’re saying, hey, look, we cannot literally make everyone love or hate AI the exact same ways we do, but we can put our thumb on a scale in a way where we can kind of nudge the voices that are directionally aligned with us to kind of speak in ways that we approve.

It’s not exactly what I would call a conspiracy. It’s more of like an influencing campaign, like an astral turfing campaign, right? You kind of start from where the world is.

And instead of trying to puppet people, you were kind of nudging them to behave in ways that you kind of approve or predict. So the scary thing about this for me is that when I saw this article and I was like laughing at it because my doctor doesn’t say this. And okay, I don’t really know about Sanders proposals.

I guess they sound like what Sanders would say. But they’re really just trying to attack Bender in her sarcastic parrot term like Kelsey Piper was a year ago. This article reeks of kind of like an influence campaign.

But I saw leftists engaging with this article as if it’s like, yeah, I’m on the left and I hate AI. It’s evil. It’s like they literally were getting people to behave in the ways that they said they would by writing an article that was not very kind of well sourced or well thought out, right?

[Nick]
Isn’t this sort of the danger of accepting the premise of a question in an attempt to answer it? Yes. So all of these people you’re talking about, whether they were in the article or then responding to it, essentially accepted the premise.

[Mike]
Well, there are only three people in the article. There’s Emily Bender, there’s Cory Doctorow, and Sanders. And to Doctorow and Bender’s credit, they didn’t engage the article, right?

But I’m saying the article is addressed to the left who is missing out on AI. But plenty of leftists, online leftists, saw the article and they’re like, yeah, I hate AI. Even Zitron is like, okay, yeah, you wrote this article, but what about the ROI of AI?

Which means it’s sort of, that’s a very relevant question. The article does not engage that. Article just vaguely says capabilities are increasing.

If you can kind of count the number of things that large language models are doing in 2023 and count what they’re doing today, like, okay, yeah, capability is increasing. What the hell does that mean? We spent the last five episodes saying economically it doesn’t mean anything, right?

[Nick]
And by use case, it doesn’t really mean much either.

[Mike]
I mean, it’s a novelty. And I guess people who are academics who are watching this thing grow five years ago, I guess, are happy or whatever.

[Nick]
And framing it as, oh, look, the capabilities are increasing is a way of trying to spin the fact that the number of use cases are not also increasing. So it’s kind of trying to put the best foot forward.

[Mike]
They are, yeah, I mean, the whole point was to put their thumb on a scale and then elevate voices for whom they are speaking about AI correctly. The funny thing is, again, that week Docker was attacked one for coming out and saying he uses AI. If they were actually smarter, they would have probably used more voices on the left who actually make their point for them.

Instead, they actually framed Doctorow, one of the voices on the left that is kind of not in disagreement with them. I mean, I guess he’s not in agreement with them either, right? But I’m saying if you’re looking for voice on the left that use AI or say, look, AI usage is okay or whatever.

And I don’t want to put words in Doctorow’s mouth. Doctorow’s more nuanced point that AI usage is okay. But I’m saying he’s one of the few who’s not, his inclination is not just shame the shit out of anyone who uses AI, right?

Something more nuanced than that, right? The irony is you would use a Doctorow to make a more reasonable point, but they are instead actually, they’re actually diving to the most extreme ends of the discussion and even maybe unreasonably framing somebody in the discussion, all to create the type of, this is a word I learned from Corey Doctorow actually, schismogenesis that they desire. Now we’re in serious academic discourse here.

Yeah, so schismogenesis is not Doctorow’s term. I think Doctorow himself got it from anthropologist David Gaber, but it’s not Gaber’s term either. But schismogenesis basically is the creation of communities that are defined on the opposition to what they don’t stand for, right?

And anthropologists actually observe this tendency in communities that are not even developed, like kind of like communal groups and tribal band groups. So like this, this tendency is not even unique to our kind of like hyperpolarized environment, but obviously in our hyperpolarized environment, you find schismogenesis everywhere. And arguably I, what I consider this, the left has been out on the article written in February to be is weaponized mimetic schismogenesis, right?

So they took the meme of, boy, leftists really hate AI, huh? They wrote an article kind of saying that, kind of choosing the slices of reality that they pointed to, right? To kind of, you know, validate what was effectively to many people, truism, right?

When you go online, you can very quickly find evidence of leftist artists, leftist writers, right? And other types of leftists who are just like, yeah, AI is bad. And if you use it, you’re bad, right?

I mean, like you don’t have to look very far to find this viewpoint. And so this article takes that kind of, that meme, right? Because I mean, look, this is, it is position, but to say this is the entire left is, is dumb.

And again, Dr. Who they mentioned does not say this.

[Nick]
And it’s painting a very large group of people with the same brush, which is never helpful. It’s never accurate.

[Mike]
But because this exists and it rings true to many people’s experiences, and because people reacted to the article in ways the article said they would react, right? The article didn’t even person verbatim say this is what they will do. But the article was kind of painting a type of critic who is doing exactly the types of things that the critics did in responding to the article, right?

The article sort of self-justifying itself, right? It is, it is itself creating the evidence that it is using to promote by, by, by truncating the discussion to slices of reality that the article approves of, right? You know, tell the story that it’s telling is kind of validated in the reaction to the story, right?

And so this freaked me out. I wrote about it repeatedly on Blue Sky, the greatest social media website of all time. That’s a joke for you on Blue Sky.

I’m, I’m there kind of out of obligation. A lot of my, my Twitter click went there. And for three years, I was kind of in the dark doing this intellectual work.

So I’m like, I said, I’ll join Blue Sky, but it’s not the same as Twitter. And Twitter has gone to me. That X shit is not Twitter.

I call it X Twitter. Twitter is dead. Yeah.

Um, yeah. So this article came out. Sanders has mentioned the article.

If it were just that I would kind of left it at that. I may not even have mentioned it, right? I did want to write about this article because I found the reaction to it.

And this idea of this mimetic schizmogenesis is very interesting, but I kind of buried the hatchet and kind of moved on. But lo and behold, I see Bernie Sanders talking to Eliezer Yudkowsky, the big, big doomer himself on March 4th, right? So not long after.

Not long after, February 16th. So we have Transformer AI, an EA outfit that basically regurgitated an argument that Kelsey Piper made last year in Vox on Future Perfect, which is an EA-funded segment of Vox, right? Now we have Bernie talking to Eliezer Yudkowsky, who while himself is not EA, is in the EA kind of rationalist diaspora, right?

There are very many people who are kind of affiliated with EA who either know or interact with Eliezer, right? And so Eliezer and a handful of other AI satheists are talking to Sanders sort of like, you know, conference style, right? He met Jeffrey Laddish, too, another AI sathe or guy.

[Nick]
Presumably they invited him to some kind of meeting. And this wasn’t originated by Bernie’s people. This was originated by Eliezer and his team.

I don’t know who initiated it. We don’t know, but we’re presuming.

[Mike]
Yeah, I mean, like, and by Eliezer, it might very well be that there is an arm of EA or an arm of sort of these movements that are kind of connected to one another, as I’ll kind of talk about later.

[Nick]
Right.

[Mike]
But I don’t know whose idea this was. I don’t know who funded it. I don’t know, you know, because I mean.

[Nick]
It would be weird for Bernie Sanders to know who Eliezer is, though.

[Mike]
It’s not because Eliezer has a book out. And Eliezer has been propped up by the media as an expert on AI for a long time. I’m not kidding about this, right?

Or early this year, he was talking to Azure client. And maybe that’s because of Vox. I mean, look, Vox is not a is not a EA organization.

It’s just future perfect. That is financed by a single segment. Yeah, yeah.

But he went on tour for this book. Right. So I knew about him.

I learned about him, I would say, and learned about him in 2018. But I didn’t see him. I kind of saw him escape the EA rationalist community in 2021, 2022.

So my first, like, in the wild encounter with Eliezer, not personally, but I’m saying in the media, was on a podcast called Bankless, which is a crypto podcast. I have a, an associate of mine was a big crypto person. And so they wanted me to watch this.

And so, but this is him kind of escaping his, his containment, right? So, I mean, crypto was obviously more simpatico with him and his kind of, like, worldview than, say, the average normie media environment. But I’m saying as early as early 2020s, he was already kind of leaving sort of the closed gates of the rationalist EA diaspora and moving on to other kind of pastures.

And then by 2023, 2024, you know, he has his book out. Maybe the book came out last year, but he started writing this book, you know, if anyone builds it, everyone dies. And so he’s, he’s been platformed by a lot of people who were not in crypto and not in EA anymore.

Right. So. And he does get cited and quoted as, like, an AI expert, Eliezer Yudkowsky.

I guess, I think. The big note, actually, I think he, he wrote an article in Time, I think, in, in 2022 or 2023. I’ll put this in the show notes.

And this is Time Magazine, which, you know, you can kind of pay to play in Time. But the article was titled, you know, basically, the article basically made a case for bombing data centers. I’m not making this up.

Now, the conditions under which they would be bombed, and Eliezer will be the first to tell you, is, like, very contingent, right? But I’m saying, we don’t have to have to pussyfoot around this. The article was about bombing data centers, because, you know, in his analogy, and he’s been consistent about this.

[Nick]
Because they’re going to give birth to Skynet.

[Mike]
It’s going to give birth to Skynet. Not, it’s worse than Skynet, because it’s going to devour all the matter in the light cone, right? If you can see it in space.

Right. It will devour it, right? And so bombing a data center or two to stop that reality where 10 to the quadrillion souls are eaten alive by a matter-devouring mechanical fungus.

Yeah. Is worth utilitarian calculus, right? Right.

So I can rant about these people all day. I’m so sorry about this. But Sanders is talking to Eliezer on March 4th.

Literally, Eliezer and Jeffrey Ladish and a bunch of other, there are a handful of these organizations. So it’s not just Eliezer’s organization. And this is why I’m saying it’s not just Eliezer’s people.

Because I’m saying, there is a coalition of AI safety research companies, or sorry, nonprofits. They’re nonprofits. They’re not for, they’re not for profits.

[Nick]
Right.

[Mike]
There’s a whole coalition of them that were at that table. So it wasn’t just Eliezer. So somebody who is in a broader coalition with these groups must have put this together.

Or Sanders, who saw the news, he’s an old man, whereas people were like, hey, these must be the experts. Let’s talk to them.

[Nick]
He does watch the news.

[Mike]
Or they’re the only voices talking about slowing down and safety. And part of Sanders’ proposal is also getting AI companies to slow down, right? So these are the people who are on the same kind of page about slowing down.

There could also be a less cynical read of this, where it’s just like, you are building a coalition of people who want to slow down. You go to the people who have some capital and want to slow down, right? Even if they are weirdos.

So maybe Sanders is knowingly walking into this kind of weird, bizarre arrangement. Maybe he was nudged there. My story is not quite over.

We can continue in a bit. Go ahead, what were you going to say?

[Nick]
No, I’m just remembering back, you know, not too long ago, when Mark Zuckerberg himself flop sweated his way through an appearance at the U.S. Senate. And I don’t think any of the senators knew who he was before that day. I’m sure their staff would eventually leave them a note like, that’s the Facebook guy.

But this was a group of aging gerontocrats being exposed to the heads of various tech companies for the first time. Perhaps Bernie Sanders has become a lot more literate in the time period after that. But that’s, I guess, my cynicism coming out about members of the U.S. Senate. That is very terrifying. And the clips are hilarious, so go check it out. But yes, finish this part of the story, please.

[Mike]
It’s invaluable. But to your point really quickly, and I’ll come back, I’ll go back to my main story. Yes.

These people have done a lot of work to legitimize themselves, right? Certainly. And if it weren’t for, you know, errors like the Sam Bakeman Frees of the World kind of like imploding, I mean, there is a world where EA is actually so normalized and regular that it’s all people talk about to sound hip and cool, right?

That was the world we were, that was the world we were kind of ascending to if, you know, FTX and Sam Bakeman Free didn’t implode, right? That’s true.

[Nick]
We could be living in that universe for sure, yeah.

[Mike]
Right. And so even with that fact, you know, like you’ve got a bunch of people who spent their entire careers kind of yelling about AI safety. And even a CEO in Dario Amadei who kind of says the exact same things, because he comes from the same cultural milieu, giving these people kind of validation, right?

So that Sanders thinks these people are kind of the adults in the room may not quite be an accident, even though I would implore his staffers to do their, do more homework, do better homework, right? But, you know, if you’re just kind of relying on the world to be correct, then there are people who have legitimized, not just Eleazer, but I’m saying, you know, like Vox is a mainstream outlet, they have an EA segment, right? Like, this is not a coincidence, right?

So let’s continue with Michael’s story time. The saga of Bernie Sanders as he becomes aware of AI, yeah. So he talks to Eleazer and all these other AI safety people, right, on March 4th.

Then on March 19th, there’s a video literally titled Bernie versus Claude, okay? Bernie talks to the Claude LLM. I’m not kidding you.

They give Claude a voice. Claude has a voice mode anyway. It wasn’t specific to this encounter, but I’m saying, you know, Bernie talks to a phone and an AI computer voice comes out and he’s like, you know, isn’t it scary what’s happening with the AI advancements?

The old man yells at phone basically should be the title of the video.

[Nick]
And what was Senator Sanders’ response to this groundbreaking moment?

[Mike]
He’s like, he’s like, it’s like a, it’s like the weirdest cross interrogation. And it’s like, do you know what, what, you know, what is happening? What, you know, what’s going on?

It’s like, yeah, I don’t think Claude’s responsible for any of that. Why don’t you talk to the humans doing this, Bernie?

[Nick]
So basically, correct me if I’m wrong here. Bernie is blown away by this, this interaction.

[Mike]
I don’t think he’s blown away. I mean, he must, I mean, I hope he knows that, you know, that it’s just a computer and that it doesn’t actually really know anything, but who knows what his staffers are.

[Nick]
Many people are baffled by this technology. He could be amongst them. We know Claude didn’t create itself.

Right. So like I. Did Claude pass the Turing test or did Bernie Sanders fail the Turing test?

The world may never know. That is, that should be the title of this episode. Welcome to episode four.

Did Bernie Sanders fail the Turing test? So next up on this.

[Mike]
Yeah, so Claude is ran by Anthropic and Anthropic itself has connections to EA, which I’ll kind of highlight later, but for our last stop on this grand tour, Hank Green and Bernie Sanders talked together on April 7th about AI. So Hank Green is probably one of the most well-known and beloved YouTubers. He created Crash Course, SciShow, right?

If you’re familiar with edutainment YouTube, Hank Green is a central part of that. I love Hank. I still watch him today.

I love Hank. You know, like Hank is great, but I’ve started to notice that Hank, I’m not saying Hank is EA, but I’m saying maybe he has EA sympathies. You know, a lot of Hank’s, so I want to be very clear.

Hank has his own kind of philanthropic ambitions and goals and they are independent of EA. And that has nothing to do with EA, but I’ve noticed recently, you know, Hank earlier this year was promoting a channel called AI in Context. AI in Context is, I believe it’s ran by an EA organization called 80,000 Hours.

80,000 Hours is actually one of the most seminal EA organizations, because it’s actually kind of what normalized this idea of earning to give. 80,000 Hours, the name refers to, that’s sort of the length of time you have in your career. And so, you know, if you choose a good career, you can make enough to donate enough to the most important causes in the world, right?

So it’s a very kind of like- This is a core EA tenant. Yeah. And so this organization is sort of designed to help people kind of find jobs that they love, that help the planet, you know, help the world, right?

Kind of on the surface, like kind of benign, right? But its connections to the broader EA ecosystem are kind of, make it kind of a little concerning. And obviously, the fact that they kind of help support a channel called AI in Context, which is commenting theories about AI doom indirectly, they take the stories of the day and they kind of, the analysis is kind of grounded, but obviously, there’s always this subtext of, what if the next one’s scarier, right?

So one of the first videos that AI in Context did was on Grock, and when it became MechaHitler, and saying, you know, they’re having trouble aligning the AI now when it’s so small and weak, imagine what will happen when it’s stronger and more dangerous, you know, like- And, you know, in Context, or even out of Context, these things are reasonable questions to ask, right? If you’re saying you’re going to deploy stronger models or bigger models, you know, and you’re not securing the ones you have today, it’s not concerning. But, you know, kind of going back to your point earlier in our conversation, asking the question, taking the premise at face value kind of legitimizes the premise, right?

And so we’re not even, if we’re kind of embracing that framing, we’re not even taking the consideration that, hey, this iteration of AI may not actually be what they’re saying it is, right? It might actually be closer to a tool or whatever, and when you believe things in, like, AI in Context kind of fear-mongering about what could be around the corner, you are kind of giving them, you’re kind of giving into their power, you’re giving them the power they say they have by giving into their premise, right? And so Hank, you know, was promoting AI in Context.

The fact that Sanders had kind of had his own EA journey and is talking to Hank is also kind of suspicious. Hank also, one of, I think, SciShow’s videos very recently was sponsored by Control AI. Control AI seems to be a consortium of, I don’t think they’re EA, but they are a consortium of AI labs that are kind of talking about EA-adjacent talking points about AI and AI safety.

[Nick]
Did any of those talking points make it into what Hank Green was saying directly that you saw?

[Mike]
It was SciShow, so it wasn’t Hank, but SciShow was his creation, right? So I’m saying one of the channels that he started was taking a Control AI sponsorship, but independently of that, Hank has his own channel where he talks about the topics of the day. Hank is a very prolific, you know, autodidact and polymath, so I mean, I value Hank’s opinions, but to give you a rogue’s gallery of guests he’s had on, Hank had on Nate Sores.

So Nate Sores is the co-founder of MIRI, the Machine Intelligence Research Institute, that is Eliezer Yudkowsky’s organization. So he is the co-founder of Eliezer Yudkowsky, or the AI Will Kill Us All organization, right? He is also the co-writer of If Anyone Builds It, Everyone Dies, right?

So I’m saying he talked to Sores kind of after the book came out, kind of legitimizing, you know, even if, you know, I didn’t watch the video because it frankly would’ve been a waste of my time, but I think any engagement with this premise is sort of legitimizing it in ways that it doesn’t really deserve to be legitimized. And if you’re going to talk to people about AI doom, well, the Foundation Labs themselves, OpenAI and Anthropic, they’re saying that AI could kill us too. Why don’t you talk to them, right?

At least there you have people working in technology talking about it, right? So Hank was uncritical at best. I don’t want to speak to him, I’m just saying I think having on Sores itself is like a yellow flag to me, right?

Because we established that Sores, you know, along with his co-founder Eleizer Yudkowsky, are literally selling us a sci-fi version of AI where it ends with the light cone of all matter being devoured, right? And if that’s our premise, especially for LLMs at this early stage, right? Like, I think that’s a non-starter, right?

So like, there’s just, if you wanted to, the only context to have Sores and Yudkowsky on is, I’m writing a sci-fi story and I want to know the best tropes in science fiction. Can you help me, right? And even then, look, this is clichéd and hackneyed.

AI devours everything in sight. Like, okay, like, real original, bro.

[Nick]
It’s been pretty clichéd, let’s say, since at least the early 90s. Right. In film, television, media, so yeah.

[Mike]
I mean, going back to, I remember the guy’s name now, Nikolai Fedorov, our little, at the turn of the century, our Orthodox Christian friend. Yes. He’s actually from the 19th century, but I think he died in 1903.

But these people, again, the story of technology is so good, it will either save us or destroy us. That impudence has been with us for time immemorial. So I don’t want to say that there’s something novel in what the EAs are doing and what the rationalists are doing, like Eliezer and such.

But yeah, so Bernie’s Pit Stop, going back to Bernie, is with Hank Green, who himself, I think, has a separate journey to EA in this way that I’ve not really done a very good job of lining up. Hank also talked to some person at Anthropic when Cloud Mythos was leaked and kind of freaking out about what, you know, I think the title of that YouTube video on Hank’s channel was called This Changes Everything, right, which is like Cloud Mythos is so dangerous and so powerful that cybersecurity is being upended. And he was kind of uncritically, I saw parts of this interview, not all the parts of it, is uncritically kind of just eating the hype of Anthropic, honestly.

And so the fact that Bernie’s journey kind of ends talking to Hank, who himself seems to have been having a separate, distinct journey, kind of nudged by the forces in this community, is like, this is weird, right? Like, I don’t want to kind of cry foul and say this is a conspiracy. But like I’m saying, this is weird.

This doesn’t seem organic completely. So this is the story, but I want to kind of lay out some details for the audience. I think it’s really important to consider.

So Anthropic itself has connections to EA. A lot of the more safety-minded people at OpenAI left OpenAI to found Anthropic, right? But on top of that, one of Anthropic’s most important executives, Daniela Amadei, is married to Holden Karnosky, who was the founder of one of the most important EA organizations called Open Philanthropy.

So Daniela is married to Holden Karnosky. Holden Karnosky’s co-founder of Open Philanthropy, Duskin Moskovitz, actually gave Anthropic seed money way back in the day, right? So there is a material tie from EA to Anthropic.

Now, I’m not claiming that EA kind of has this talent in Anthropic today. I think Anthropic is sort of its own thing. Like I was saying, the fact that there is this synergy, that one of the executives at Anthropic, Daniela, who is Dario’s sister, I think she’s the president of Anthropic.

But Daniela, Dario, Duskin, and Holden all know each other. They’re kind of all friends, right? And Duskin gave Anthropic money.

He also co-created Open Philanthropy, right? Which gave Anthropic money, right? There’s a sense in which, you know, there’s synergy here.

There is a relationship here. And Anthropic, unsurprisingly, kind of mirrors the tenor and posture that a lot of AI safetyist people have, even going so far as to worry about existential risk, right? And at the beginning of February, Anthropic closed their Series G rounds.

They got an infusion of cash. And then immediately after, there’s this media blitz where Sanders and Hank Green are talking about AI safety. And, you know, again, I’m not alleging anything necessarily as a conspiracy when I’m saying there does directionally seem to be, within the communities that Anthropic is part of, a push to kind of shape the narrative.

Now, to what extent that Anthropic itself had a hand in shaping that, you know, I don’t know. But Bernie did not sit down to talk to ChatGBT. He sat down with Claude, right?

Which is Anthropic’s AI, right? You would think that if Bernie was doing some sort of media blitz that was sort of like not at least being nudged a little bit by Anthropic or the EA organizations that he would sit down with opening eye to, he was like, I’ve talked to Claude yesterday. Today, I’m talking to ChatGBT, and it tells me the sky is blue.

Is that true, ChatGBT? Right? Like, you would think that he would talk to, hell, Grok.

I don’t fucking know.

[Nick]
I would pay money to see Bernie Sanders talk to Grok. But yeah, that would be a slightly more even-handed way of approaching it. And in the absence of that kind of comparison, we’re left with, okay, he only spoke to Claude.

Yeah.

[Mike]
And it was sort of the capstone of his journey, too. I mean, I don’t consider Hank Green technically part of the, it just so happens he did talk to Hank Green.

[Nick]
It’s just that there are connections. Right.

[Mike]
But, you know, his last kind of journey through this funnel is talking to Claude before talking to Hank Green. So, yeah, I don’t know. I don’t know what to make of it.

But I’ve not really seen anyone kind of talk about these connections. If there’s any journalists out there, they should look into this. I’m not a journalist or a listener.

I’m not trying to be. I’m just pointing out what I’ve observed. And, you know, given my situation, my location in all this, I happen to be kind of aware of all these communities to the extent that I am.

And it gives me a little bit of insight into there could be something here. Right? I can see the linkages.

I don’t know what connects them. But it’s possible there’s something here.

[Nick]
And it is, of course, our job as we’re also our self-appointed job as residents of the Bay Area, very close to Silicon Valley, in the shadow of Stanford and the VC in San Francisco and everything else. We freak out so you don’t have to is basically, I think, a short way of summarizing our mission mandate. And yeah, this is a lot of connections.

This is a perfect conspiratorial Charlie Day murder board. I can see the red strings connecting all these various individuals and organizations. To me, it’s quite clear.

And I think we can make that clear for the audience as well.

[Mike]
Maybe. I mean, I always like to hedge, not because I’m not confident, but it’s because, you know, if you don’t have the full picture in front of you, the best position is one of agnosticism, right? Right.

But I am saying I am presenting this to you, not because I know for sure what’s happening, but because at the very least, you know, I think the linkages themselves are kind of important to know about, right? If you’re kind of looking at evidence and trying to weigh different possibilities, I fucking sound like a Bayesian now, one of these rationalists, you know, one node of evidence to consider is like, I don’t know that Sanders himself is completely neutral from this cultural milieu that he interacted with, right? That had their own reasons, perhaps, for kind of controlling the narrative, right?

Now, whether or not the extent to which they did this, I don’t know. I’d have to prove that, but it’s hypothesis on the table now, right?

[Nick]
And the result, of course, is, well, the results. Bernie Sanders is currently the only member of the Senate and the only Democrat to have proposed anything as comprehensive or meaningful as his tax, his sovereign wealth fund, whatever they’re going to call it. And so we’re kind of then returning to this original premise, whether we accept it or not, is the left missing out on AI.

The main person who’s not missing out is Bernie Sanders. And he is now kind of the sole voice in the left-leaning corridors of Washington. How do we interpret this?

Has he been, therefore, captured by the movement itself, at least in his resistance, is he being captured at the same time?

[Mike]
Yeah, I mean, I would put him squarely in bucket one that we talked about in episode three, right? With true believers, right? So he believes in the kind of fearful vision that we’ve been given.

And so he’s like, let’s try and control the downside, right? Maybe it was too effective because Sam Altman’s counterproposal is, I mean, 50 is too high, old man, but I’ll do 5%, right? But they are kind of talking terms, right?

They’re speaking the same language. They’re kind of haggling over the equity, the valuation to give this company and to tax it at.

[Nick]
Yeah, it is distressing. And, you know, like you referenced before, we did basically say there won’t be a bailout because there can’t be a bailout. And this is looking more and more like Sam Altman and these various tech moguls seem to be angling for one in a very direct way.

[Mike]
This could be the bailout if, so these are the conditions in which there could be a bailout, right? So condition one is that Trump unilaterally does this. I know you and I have talked kind of, you know, off the pod that, you know, regulatorily, you know, the action space where Trump is kind of constrained here, while Trump can kind of- Even as the president, there are limitations on what he can do in terms of interacting with private companies and fostering or creating bailouts.

[Nick]
It’s up to Congress, has the power of the purse. However, to your very point, if perception is indeed reality, just the president himself putting his kind of vague blessing on the idea can affect the marketplace, can affect the way banks and various financial institutions value these companies. And again, that’s the perception.

And then the reality follows.

[Mike]
Right. Yeah. And that’s the point I’ve always kind of been trying to angle.

I mean, especially the number one piece of that valuation, that number, once it’s realized, right? Once someone says the number out loud, it’s like an invocation, as Blair Fix says. The invocation basically manifests that number, that value into reality.

And other people begin to orient themselves around that, right? But to your point, and this is probably something for another episode, given your background in the federal government and military, because a lot of people hear, oh, Trump can’t do that. And Congress has approved that.

And they’re kind of rolling their eyes. Because I’m saying public perception is that Trump has run roughshod over the Congress and the courts too, to some extent. Certainly many times in recent memory, yeah.

Right. Yeah. So we’ll do another episode where you kind of…

Nick has walked me off the ledge a few times about some of these stories about what Trump has done, and sort of interpreting the kind of actual story from the headline, as it were. So when Nick says, Trump likely can’t do what they’re saying he can, I do put some weight on that, given Nick’s personal experience. But I know for you, the listener, you have every reason to be skeptical, given the reporting around Trump’s behavior and what he seems to be able to get away with.

[Nick]
And the weakness of this particular Congress, which is the… I think this is the weakest Congress in 120 years.

[Mike]
I think Congress has checked out.

[Nick]
Congress has checked out. They’ve engaged a little bit in the lead up to the midterms, and they’ve pushed back a little bit on a few ideas. But would they push back on a sovereign wealth fund for the American people?

I don’t know. I think it’s got maybe a nice ring to it, and maybe they might get behind it. I have to acknowledge that.

[Mike]
But what really matters, actually, is just that valuation will get other market actors to orient around it. And maybe OpenAg is more investment based off that number. I don’t know.

Right. Right. So, I mean, just the fact that we’ve been talking about this is kind of dangerous.

And Sanders is actually kind of helping foment that reality, whether he’s a willing participant or a stooge or whatever. You know, I know I spent this entire episode kind of laying out the case that he might be not captured in that he’s actually shaking hands and greasing palms with Anthropic, but…

[Nick]
But downstream of this manipulation, at least.

[Mike]
They’re nudging him, just as he nudged or just as Transformer AI nudged, quote, unquote, the left in the response to the left is missing out on AI, right?

[Nick]
It’s something that we’re going to keep our eye on, for sure. And now we should probably explain why is this magic number? Why is this valuation so dangerous?

And we can begin that discussion with the fact that none of these companies are profitable. Right. They are, in fact, wildly unprofitable to the tune of…

Billions of dollars.

[Mike]
Billions of dollars. Yeah. OpenAI lost, I think, $21 billion.

[Nick]
According to Ed Zitron’s recent reporting.

[Mike]
Right. Yeah. Ed Zitron got the numbers directly from a source at OpenAI themselves.

So these are real numbers. These are real losses. We don’t know how unprofitable Anthropic is.

What was happening a little bit in May and June, mostly June, people were looking at OpenAI’s numbers and they’re like, oh, okay. Well, if OpenAI is not good for it, Anthropic surely is good for it. I mean, they were profitable last month.

Well, Anthropic was actually getting compute from XAI because XAI kind of tapped out of the race. They said, hey, we can’t use our current GPU cluster. We’ll give Anthropic two free months, the months that they reported profitability.

[Nick]
That was a literal gift. That was a donation to essentially Anthropic.

[Mike]
Yeah. I mean, obviously it’s not altruistic and any of these earnings make… Securing that the tenant is in XAI’s best interest too.

And it’s also, I don’t know. I mean, you’d have to go back looking at that diagram of all the circular dealing, right? All the round tripping, all the round tripping of the AI revenue.

I don’t know what XAI and Anthropic’s linkages are, but they might have one via Google or something like that, right? So I’m saying there might actually be material interest for XAI and Anthropic’s wellbeing, not just because of this business relationship, but because of the round tripping of the revenue. I don’t have the chart in front of me, so I can’t tell you which companies are linked together.

We all know obviously Nvidia and CoreWeaver are linked to everything. They’re the center of the circle, but I don’t know what XAI and Anthropic’s relationship is.

[Nick]
So I’ve heard now from a couple of different sources, if OpenAI has lost 21 billion, Anthropic probably lost in the neighborhood of 11 billion. That means the Sam Altman proposal is deeply dangerous. It is, they wanna peg their value at the current amount, ignoring all the losses.

And therefore that 21 billion gets kind of, not just wiped out, the American taxpayer is on the hook when the bill comes due. And that bill will be at minimum $21 billion. Given everything that’s going on in the country, I don’t think that there will be a lot of appetite for that.

But again, we have a President Trump in the White House. We have a weak Congress. We have these strangely charismatic tech moguls.

So we’re kind of saying, look, this is not impossible. It exists down the road. And right now, let’s all start advocating and avoiding that outcome, please.

[Mike]
Don’t fall for the kayfabe.

[Nick]
Don’t fall for the kayfabe.

[Mike]
You’re not gonna lose your job. They’re just threatening you with that. They’re threatening you with all of these scary things.

And whether you fear them or you love them, that is what they’re preying on, right?

[Nick]
They need both of those results. And Bernie Sanders is now, whatever his intentions were, has fallen into that trap. That is how we are assessing his recent behavior.

[Mike]
Just don’t give them the time of day. Just like you don’t enter strange vans with people offering candy. Don’t give them the time of day, right?

You could use their products. I’m actually kind of agnostic on that. I mean, although I think you should try and find more ethical AI.

There are AIs, for example, that are not trained on stolen data. There are open source AIs that are kind of less, you know, I don’t want to say less problematic, but they’re not contributing to the environmental harms that anthropic or open AI is contributing to, right? But whatever you do, right, use their product, don’t use their product, but don’t believe their kayfabe of this is the greatest technology that ever exists.

It’s so dangerous slash so good that we have to preserve it at all costs or protect ourselves from it at all costs because that means you’re playing their game, right? Even if you just talk, you know, in ways where you’re like, it’s possible for them to do what they’re saying. I know that they want to take our jobs away.

Yes, they do. They do. But if they could, they would have already done it, right?

The kayfabe is in the fact that they can’t do it, but they are convincing you it’s happening imminently, right? We were talking about this in episode one or two, I think, with the Uber example. Uber example, yeah.

Driverless cars are coming. They’re coming. They’re coming.

Any year now, Elon Musk keeps promising that we’re going to get these. Kicking your feet up. You can drive from New York to San Francisco, you know, not paying attention to the road.

People have died that way riding in Tesla’s autopilot, or we should call it that, but, you know, whatever. And it’s not, tech just promises you what you think the next big thing is. And that next big thing has been shaped by your culture, right?

Going back to the rationalists and the EAs, right? Like, they believe AI will sublimate all matter in a light comb, because they’ve read science fiction about optimizers that do that, right?

[Nick]
And we’re sitting here opposite each other talking, two fans of science fiction raised and brought up on it. But yeah, these folks who made it to the top of the industry, they are taking it so far and so seriously to a degree that it could affect all of our lives materially and permanently. Oh, it will.

We talked about it in episode four, don’t you remember? Yep, yeah. The four bubbles are going to burst.

The four bubbles, yeah. That’s the thing to be afraid of, is like real material- Reality. Economic reality, not the state of technology that’s going to consume the universe.

And as a point of context, there are great driverless cars in downtown San Francisco right now, they’re called Waymo, and they are locked on that grid and can’t leave it. So learning the specifics of how these things actually function, just being more aware, being more educated, is the first step towards getting out of this trap that’s been set up for us. And certainly, breaking out and using different open source AI models is a very good idea.

In fact, it’s such a good idea that major American corporations are doing it as well. Let’s list them off. I remember three from earlier, Airbnb, Siemens, Microsoft.

So Siemens and Microsoft are both using- DeepSeek. DeepSeek, yeah.

[Mike]
Microsoft is considering it. We don’t actually know if they’re evaluating, or they might be evaluating it.

[Nick]
And the fact that they’re not using any of the big two- Microsoft also has, Microsoft has open source models of their own design, their own inventing. They might be moving away from to embrace DeepSeek.

[Mike]
Yeah. So it’s like hosting the compute yourself is expensive too. Right.

Obviously, depending on what you need it for, right? And it’s just that, well, if we completely offload the cost, this is Microsoft saying it, it saves them probably a lot more money than running their own models on their own Azure infrastructure. Microsoft Azure, for those who don’t know, is their cloud hosting service.

So you can deploy applications in the cloud on their servers using Azure. But they obviously have their own access to Azure because it’s their product. And they’re saying, rather than deploying FI, FI is one of their AI families.

We’re going to go to DeepSeek in China to get compute and do AI stuff.

[Nick]
For our listeners, what’s the origin of DeepSeek? Obviously, the association with China is usually pretty made clear, but- DeepSeek is just a Chinese AI lab, right?

[Mike]
So they’re probably one of the bigger AI labs in China. Back in January 2025, actually, the stock market had a little oopsie-doopsie because when the market learned that DeepSeek was making open-source models for cheaper than what Chachabuti and Andropic were running for, they’re like, is the AI a bubble? And then they promptly forgot about that and kept piling in.

What a temporary affliction of wisdom that the markets had. Yeah, it turns out that you guys were onto something nearly two years ago. So DeepSeek, it’s a company.

DeepSeek is a major player in AI, but probably one of the bigger ones in China is Alibaba, which you may have heard of as a company in China.

[Nick]
It’s their version of Amazon. It’s even bigger.

[Mike]
It does everything. So Alibaba has a research arm that does AI. Their models are under the Quinn family.

So a bunch of models called Quinn. Quinn is also a pretty popular model that’s being adopted by a lot of tech companies in the West now. So I think, yeah, you and I were reacting to some news.

I think open-source adoption amongst tech companies is up nearly 40%, 50%. Basically, they’re starting to gain most of the market share kind of not so slowly for AI usage. And this goes back to the story.

Again, a lot of the threads we’re talking about here, we’ve talked about in all our episodes, all the names I keep saying, like Eliezer. All these through lines are the same through lines. But one through line we’ve had is that, look, AI as it is, is expensive and economically doesn’t justify itself.

So of course, lo and behold, when Chinese companies promise cheaper, lower costs, and they offer them, then of course, people are going to jump on the bandwagon. They can still have their cake and eat it too. And the market will reward that.

Right. Now, to note, and this is sort of where the coverage of the Chinese models kind of falters, is that these companies aren’t profitable either. Right?

So they have a different business model. So the American companies, many of them, not all of them, but many of them, if you exclude Meta and Microsoft and Google, right? So OpenAI and Probit are the ones I’m mainly talking about.

[Nick]
The big two, right?

[Mike]
Yeah. They’re like, hey, we got to scale at all costs. So we’re going to have closed source models that are very large, that can do the largest number of tasks possible and capture all the market.

That was sort of their thesis, right? Scale is all you need.

[Nick]
This is the hyperscaling phenomenon. Yeah.

[Mike]
So scale is all you need. You can build a big model. Big model gets you closer to AGI.

And then at one point, if they’re really delusional, they believe they’re going to devour all the matter in the light cone of the universe. And, you know, big model is better, right? For everything.

That’s turned out it’s not true. There are actually trade-offs when you kind of build an optimization system, right? And so there are American companies like Meta and Google to some small extent, and Microsoft to a small extent.

They have open source models. Microsoft, Meta, and Google are doing what I call a hybrid strategy. Or I guess I should say Microsoft and Google.

Meta has actually dropped out of the AI race. Meta is saying it’s going to become a hyperscaler. So hyperscaler, actually what that term refers to is a provider of compute at a large scale, right?

But hyperscaling is the act of, you know, building out more compute for these massive, massive AI workloads, right? So Meta is like, we can’t compete with AI labs. By offering their own AI model to the public.

And they were doing predominantly what I would call open source model like these Chinese companies. So Meta was releasing open source models called Lama. I think Zuckerberg saw the popularity of these Chinese models much earlier on.

And he saw the enthusiasm around open source. And he’s like, oh, we can, you know, because everyone hates open AI because they went closed source. Open AI was kind of like, back in the GPT-1 two days, it was kind of like, going back to the story I’ve been telling, AI was an academic field.

It was very open. People shared knowledge freely, right? And so open AI is very much a child of that culture.

And then when open AI kind of caught the commoditization bug, where Sam Altman’s like, hey, we can make money from this. They began to enclose. They began to enclose their models and enclose their knowledge so that, you know, they could keep their lead, right?

I choke a little bit when I call it open AI, because like, it’s closed AI. People started making that joke in 2023. So it’s not, it’s very, very ancient joke.

But, you know, the shift, you know, meant there was an opportunity for any company that could offer open source models to kind of regain some of that luster from, you know, the old days of AI. And so a lot of companies kind of dipped both feet in both waters. To Meta’s credit, though, their primary strategy, at least in 2023, 2024, was open source models.

Then when they caught up, when they started making models that were approaching, you know, quote, unquote, frontier levels, they were like, time to pivot to closed source, right? But, you know, I guess that didn’t go well. And that’s why they are now committed to hyperscaling instead, you know, offering their servers for people who want to deploy AI workloads.

It turns out, though, I think only Anthropic and OpenAI are the only companies that want to buy this much compute. So they’re not gonna have a very big market. Meta’s not gonna have a very big market.

But going back to my main point, Chinese companies are not profitable.

[Nick]
There are American companies- There are no profitable AI companies as of the recording of any of this.

[Mike]
Yeah. So like, you know, it is true that a Google or a Microsoft can give, you know, the public, the pleblians as, you know, you want to call them little crumbs of open source models. Like Gemma 4 is a very good model. I have a home lab, I’ve tried using it.

From Google. From Google, yeah. So Gemma 4 is basically an open source model that is not as good as Gemini, but is good enough for most tasks, right?

And it’s a pretty small model, so you can actually run it on consumer hardware. I’m actually running it on a 3090, which is a consumer card from 2020. So this is as good as you can get.

Now, 3090 is kind of a high-end consumer card, but, you know, given that it’s like six years old, it’s not like it’s top of the line tech. I mean, it was back in its day. But, you know, Google and Microsoft kind of dipped their toes in both waters, and they can do this because they have other business revenues, business, you know, lines of income that let them finance a losing project, like training an open source model, which, you know, if you’re going to train a model that costs compute, right, you have to buy compute time to train the model.

It’s running on computers. Which is deeply- Deeply expensive. Expensive, yeah.

[Nick]
Right.

[Mike]
And then you’re going to offer it for free, right? You spent all this money training the model. You’re going to offer it for free, right?

So now in theory, you know, not everyone wants to host their own model. I built my server. I do regret it a little bit because I was, the reason why I wrote all these articles, you know, AI is a dark pattern, whatever, is because when I began working with my server and deploying these models, it’s like, holy shit, all the value comes from people actually building the pipelines and building all the scaffolding for the models.

They’re not, they’re actually not this good on their own. In fact, models by themselves are kind of dumb. There’s a lot of human labor that goes into making a model seem like it actually knows things and is magical.

And having learned that bit of lesson myself, I’m like, I should warn everybody. It’s actually, it was, it’s a good learning experience. I was trying to get an IT too.

[Nick]
So, you know, I don’t consider it a complete waste, but- This feels a bit like if you sit down at the poker table and you can’t spot the sucker, you’re the sucker. I mean, that kind of is the theme of a lot of this.

[Mike]
I think that’s the theme. I think there are people who are actually kind of having their cake and eating it too. And that, you know, look, if they’re using AI for a very bespoke task and all these companies are going to release open source models or open AI and Anthropic are going to subsidize their product, then it’s your buffet.

Like they’re not losing, right? They know what they want out of the AI. They’re getting it out of the AI.

Now for the rest of us who are footing the bill in terms of the environmental costs and at least for the closed source models and, you know, the labor costs, whatever, right? Like it’s not a great deal, right? But my lesson to everybody is like, look, like the technology itself, you know, is being used by terrible people.

They’re promoting a terrible vision for it. The way we get out of this, because we can’t ban it, is, you know, we can’t ban these open source models from China. We can’t ban these open source models that are running on my hardware.

[Nick]
The genie is firmly out of the bottle.

[Mike]
Not because I’m saying you can’t take it from me, but I’m saying like literally a ban will not reverse the entropy of having released these models into the world, right? The way out of this is basically telling people the way you get value out of these things is to scope them aggressively, right? That’s how we get out of this.

Once in everyone’s mind, LLMs go from some magical human-like intelligence to, this is a workflow process machine. We will- It is a tool only. Right.

We are going to go into a world where the scope of the harm begin to naturally diminish. Now, will they go away entirely? No, like factories are still harmful, right?

But factories in the heyday of their existence, where we talked about this in episode one, they’re terrifying, right? They’re terrifying. I think it was episode two, actually, the way I cut it.

They were terrifying facilities, right? Children getting mangled and women burning in buildings alive, right? I mean, factories are horrible.

And you probably wouldn’t blame anyone for saying, factories shouldn’t exist back in 18-whatever. And in Bangladesh, they still are horrible. But now here in America, they’re much better.

You’ve made that point. I’m not trying to speak against our brothers and sisters in the global south.

[Nick]
Overseas, yeah, of course.

[Mike]
No, of course, right? But this is the way we fight the enclosure with the double movement, right? That’s right.

That’s right.

[Nick]
No, so it’s a question of I guess the lesson of this episode, though, is don’t be like Bernie Sanders, in a way.

[Mike]
Well, I guess it’s Bernie’s cool, and he’s being true to himself. But yeah, I mean, I think a little bit of homework would have helped Bernie kind of- His staff should have explained a lot of this to him. The reason why I started this podcast with you and why I’m writing about AI and my misaligned markets, which is not an AI blog, by the way, if you noticed, it’s a political economy blog.

But there’s such a dearth of good discussion of AI. I think now that the bubble’s starting to pop, people are kind of actually beginning to surface good explanations of what’s happening. It also gives a permission structure to be more critical.

[Nick]
So this is opening up more.

[Mike]
But I was writing about AI intelligence as a dark pattern back in August of last year, because I’m like, I spent this much money on a server, AI is not doing the things they tell me to do. I also bought this because of my work. I work in a tech company, and it was helping me in my work too.

But it’s just like, none of the things that I was sort of promised are true. And if I’m learning this, right, people, I hope they’re learning this, and I was learning very quickly. They weren’t learning it.

I just got firsthand experience of why this is true. And I’ve kind of made it my mission to tell the story, even though I think at this point in the bubble, more people are kind of on my side. I don’t have to really shout as much as I was back in August of last year.

[Nick]
Yeah, it’s less of an uphill trudge to have this conversation than it was even a year, a year and a half ago, or the first time that you brought up with me, which is now almost two years ago. Or even three months ago.

[Mike]
I’m saying like literally the material reality of this bubble is becoming day by day.

[Nick]
Very evident.

[Mike]
Yeah, even when I was writing about the AI bubble in March in my, you know, what to expect when you’re expecting a bubble post, there was resistance to the idea of what I was saying. I mean, granted, I felt confident in publishing that because I’m saying, you know, a lot of the financial reporting I saw from Ed Zitron and other sources were like indicating it’s definitely a bubble. But I had the nugget of that post, the idea of the post back in December of 2025.

I was actually working on it on Christmas Eve. So, you know, like these ideas, you know, they were out there for a while. They’re not, they were obscure, you know, six, seven months ago.

But I think the reality is bearing fruit that this is true. These things have been overhyped and we’re in a position where we have to consider, you know, what are the material realities that are going to affect the AI bubble? And how can we kind of soften the blow, soften the landing?

I guess Bernie to his own, in his own way is trying to do that. But obviously he’s doing it in terms that the AI industry favors.

[Nick]
He’s doing it on their terms in a sense, yeah. So that’s the lesson of the show.

[Mike]
Yeah, do you, yeah, be yourself, be true to yourself like Bernie Sanders is, but do your research.

[Nick]
But also do your research, which he didn’t do.

[Mike]
Yes, he trusted Eliezer Yudkowsky. I would not trust that man to write science fiction, actually. Which I said, I said, go to him for science fiction advice, but that’s actually- But don’t even trust him for science fiction.

He’s actually written science fiction. It’s actually terrible. So I take it back.

[Nick]
You’re rescinding your- I guess ideation is good.

[Mike]
If you’ve never heard of science fiction before and you want some, you know, basic exposure to tropes in science fiction, go to him.

[Nick]
All right, we’ll make that, we’ll make that the first lesson. The follow-up lesson is, we always say time and again, this is a billion-dollar industry masquerading as a trillion-dollar industry. It looks like, again, with Ed Zitron’s recent reporting, it’s barely even a billion-dollar industry.

[Mike]
No, no, it’s probably a multi-billion-dollar industry. It’s probably at least, he used to say 65 billion-dollar total adjustable market.

[Nick]
Yeah.

[Mike]
He rounded down to like 13. I don’t know why. I asked him on Blue Sky, hey, you used to say 65 billion-dollar TAM, now you’re saying 13.

A TAM just means the total size of your market, how much money you think you can get from your demand base. So he went down from 65 to 13.

[Nick]
To 13.

[Mike]
I don’t know. I don’t know where he got the numbers from. I mean, I’m assuming he’s sort of like directionally kind of sussing out like use cases and kind of pricing them, but he’s not, I don’t think to my knowledge, he’s written extensively about how to price actual AI use cases.

Because his reporting mostly is on what are the actual losses, right? And before we can even have the discussion of, how to price actual use cases, we have to discuss the losses, which he’s been shouting about for the last three years. And to his credit, the world’s finally hearing him.

I hope the man has receipts. He probably is going to do a victory lap very shortly. Kudos to him.

[Nick]
I hope we’re invited to that party somehow. So we usually also end an episode with a recommendation. I’d love to posit one.

Everyone, if you can, go see or rent or purchase somewhere, Ghost in the Machine. Amazing documentary that I found truly eyeopening about sort of the intellectual history of AI in the modern era and really every other topic on the list. So yeah, go see Ghost in the Machine and we’ll see you on the next pod.

[Mike]
I’ve been Mike.

[Nick]
I’ve been Nick. And thank you for joining us as ever.

[Mike]
Until next time. Take care.