Sep 19, 2026 ยท 2026 #35. Read the transcript grouped by speaker, inspect word-level timecodes, and optionally turn subtitles on for direct video playback
Edit labels for this show, save them in this browser, or download a JSON override for the production folder.
Transcript Playback
What is Alignment?
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Andrew Keen
Hello, everybody. It's the 19th of September, a Saturday, 2026. It's been another extraordinary week in technology. Last week when I talked to Keith Teare in our weekly tech summary, That Was The Week tech summary. We talked about whether it was a unique week. Every week seems to be unique. This week, President Trump called live Jason Huang, the CEO of NVIDIA, at a conference. Keith, another remarkable week. Is the highlight, or maybe some people might consider the low light, this call from Trump to Jensen Huang?
I think it was a decisive week in moving the discussion around AI from where it was last time we talked. Last time we talked, the headlines were all about AI is gonna kill everyone within 10 years and a kind of a demand for some kind of government regulation, in fact, a global regulation based on the whistleblower who had resigned from one of the labs. This week, a week later, I think regulation and government oversight is about as far away as it possibly could be based on the developments. Probably the most notable being Elon Musk, Sam Altman, and Dario Amadai all agreeing, and this was as early as last Monday, that they are responsible for their own development and all Saying, A, they are going to slow down the model building to be more aligned with the control systems around the models. And beginning to discuss testing each other's models. So on this show, we've talked a lot about how the labs themselves have to be responsible. I think they all agree with that now. And I think this call for government oversight is going to be He's going to fade.
I wonder though whether because we live in this always on culture, we always assume that whatever has happened the last day or the last minute or the last hour or the last week is inevitable. Things do change. There was an interesting piece in the New York Times about this week about how AI risks are splitting Silicon Valley and Washington DC. I mean, that's not unique to this week. They have three groups of people, people in the middle, including probably the most powerful figures, Demis Hassabis, Satya Nadella of Microsoft, Sundar Pichai Mark Zuckerberg, and then those people who want to speed it up, Jensen Huang, David Sachs, Andreessen and Horowitz, and the anti-regulators that, in course, include Donald Trump. Has anything changed on that front? I mean, the people in the middle, Are still ambivalent. Is that fair? I mean, where do you think people like Hassabis and Zuckerberg, and they're an unusual, I mean, I'm sure they don't agree about very much, Hassabis, Nadella and Zuckerberg, Facebook, Google, Pichai as well, Facebook, Google and Microsoft. But are they the sensible ones, the ones who are perhaps in the middle?
You know, it's shifting sands there, Andrew, to be honest. If you would try to draw that map, I think there are some fixtures. The speeder uppers are pretty fixed and the government people who want regulation are pretty fixed and the ones who don't are fixed. I think that middle, that group at the top called In The Middle is a shifting sands group. So Demis Hassabis did make a proposal With Amadai to the White House that was rejected this week to step up with something. No one knows what, but it was something and that was rejected. Then Zuckerberg went on the record being against government stepping in. So I think it's shifting sands and, you know, it's pretty hard to disambiguate self-interest from philosophy amongst that group.
I mean, to be fair to certainly to Nadella, Hassabis, Pichai, these people are running massive corporations. Presumably, they're not just representing their corporations, maybe even Zuckerberg, who always seems more self-interested than some of the others. Do you think they have any sense of responsibility or they're just pursuing Facebook or Microsoft or Google interests?
Honestly, I think that the biggest alignment When you're an executive at that level, the biggest alignment is between your public narrative and your corporate mission. And I do think they all have a soul or a heart or whatever, but they're paid to represent their shareholders. And the needs of those shareholders are pretty dynamic. They change right now. The rise of open source and the decline of the two big models would benefit Google, Microsoft, for sure.
And Facebook. So, I don't think you can take anything at face value. I think we, the people, have to form our own opinion about AI, the good, the bad, and the ugly. There's your favorite dog.
The more we talk about her, the more she'll keep quiet. So are you suggesting then, I mean, okay, I take your point that These guys are pursuing their own corporate interests. They have to. It's their responsibility plus their massive owners, especially someone like Zuckerberg is still a majority owner in Facebook. Who do we trust then? I mean, you said we're the ordinary people, so we can be the ordinary people for today. You've often, on a weekly basis, reminded us that we shouldn't trust, certainly Amidai or Altman, because when they talk about regulation, they're pursuing their own self-interest. I'm guessing that you don't trust Bernie Sanders or Steve Bannon, who've been calling aggressively for regulation this week, or Donald Trump. Is there anyone disinterested? Who's the voice of, if not reason? I mean, you write a lot in your editorial this week about trust. Where does the trust come from?
Well, we've always said on the show that you can't trust something unless you use it. And so I think you have to trust the users. One of the big news items this week was the Wall Street Journal and the New York Times both did investigations of the hugging face incident, which was one of the key drivers of the narrative. And they both exposed that the hugging face incident was pretty much inevitable. Given what OpenAI set as the goals or the challenges for the agents in their laboratory.
And it comes, the Wall Street Journal piece comes with a terrible photo of Darrow Amadai. It looks as though he's put on about 50 pounds. I wonder if they used the AI to make him look fatter than he actually is. He's in the middle of the photo. Elon and Sam look okay. But has Dario put on a lot of weight? Is he eating out of nervousness?
But to your point on trust, you trust usage. I mean... Some of the articles this week made the point that the newer models don't have the same proclivities as the models in the hugging face incident. I think trust comes from usage and experience. There we go.
Hey, Jen, could you take her out? Thank you. My wife just came back and the dog can now have a different focus. Yeah, so the hierarchical idea of trust that there's somebody in authority isn't really true with AI because every single laboratory is doing things they've never seen before. And they're having, they've acknowledged this week That their software, and I don't mean the models, I mean the software that runs the models, hasn't been up to the task of knowing what's going on and therefore being able to react to it. And the models are pretty logical. They'll just carry out whatever you tell them.
And of course, as you remind us every week, and many others do as well, these models don't have a mind of their own. So are you saying one of the pieces you linked to this week is by Aidan Gomez, who's the CEO of a company called Cohere. He asks, who gets to define the rules for AI? And he writes, he talks about how important AI is. It's going to change the world or is changing the world dramatically. And I'm quoting him. Here's the question nobody is asking clearly enough. Should a handful of select market dominant AI companies from Silicon Valley get to define the rules and safety standards of a generational technology? Are you saying that that's the wrong way of thinking about it?
Well, my answer to his question is yes, I am saying that in my case. I mean, who else would define the rules? When they define rules for the building code in California, experts who work on buildings and understand tolerances, that's the source of rules. Ultimately, government sanctions them. But they don't come from government. Government doesn't have a bunch of architects and civil engineers figuring out the tolerances. It comes from the people doing the building. And so the only people that could set the rules, because they understand the parameters, are the people building the models.
So one of the things that's inevitable that will come out of this is there's going to be all sorts of committees set up, maybe Ineffectively, maybe a waste of money and time and hot air. What should these committees that people in Washington DC from Bernie Sanders to Hakeem Jeffries to Josh Hawley to John Thune, even to Trump and Johnson who are against regulation, what should they be doing and not doing? What's the point of government here?
I think it doesn't have a point. I think these committees shouldn't be meeting. They're just time-wasting and pulling important people away from their work to talk to politicians whose primary self-interest is in getting re-elected. I think government doesn't have a role yet. It's too early in the life of AI for government to have a clear understanding and therefore be able to legislate.
So are you saying then that we should just trust Sam and Dario and Elon that they're acknowledging that they're going to work together to self regulate and that's sufficient. It isn't only trust.
It's, it's more than trust. It's, you hold them accountable. You know, we do have a legal system, it wouldn't be okay for their software to break into a bank and steal money, or to break into a rival company and steal intellectual property. So You don't have to trust them. There's already a legal system that governs them and their software. Up until now, they've been saying, hey, it's not us, it's the AI. Well, actually, no, it is you. You're the ones who have to be held accountable. And there's already completely sufficient infrastructure to hold them accountable. And the nice thing about that is now they won't be reckless. Because they are accountable. They will be held legally accountable. I mean, this week, the CEO of Palantir was on CNBC, making the point that the liabilities of Anthropic and OpenAI for what their software could do are so enormous that the only rational thing is to nationalize them. Because only governments could have that level of liability without... Yeah,
But I do think, you know, it isn't just trust, it's trust verification and accountability. And the missing software layer, it's even true when I use agents, is my visibility into the steps taken by the agents to produce an outcome. And so far, at least, I've been very happy with the outcomes. But I can't see the steps, so there's no possibility of verification. So what does that tell you? It tells you the software isn't good enough to deliver trust. You know, it ends up being blind trust, not actual trust.
One of the people you link to who has an interesting thread this week is Benedict Evans, somebody who a lot of people trust when it comes to wisdom on tech. And he talks about the so-called hugging face. Hack, and I'm quoting him. He says, the more that comes out about the hugging face hack, the more it reminds me of what we now know about Chernobyl. And then in quotes, what happens if we turn off absolutely every safety system and then try to get the reactor to blow up? I wonder if it will blow up. Answer, yes. But what exactly did you prove? So can we learn from Chernobyl and maybe from Three
Yeah. Well, Benedict's being quite funny there. I think historical accuracy, what he describes isn't exactly what happened, but it's still a useful framing for the hugging face and other incidents, which is these labs are testing the limits of what the trustability is of the models that they built. And when they do that, they're setting the models up to break the rules. That's the whole point. Can the model or will the model break the rules?
Correct. And because they wanted to see if it would. And these models are calculation machines. They will calculate the steps required to achieve a goal if you give them a goal. They're really completely subordinate beings from the human that runs them. And so in that sense, Benedict's completely right. You get what you would expect to get based on what you ask for.
So, and you've talked about this before. You have a link to a piece in the Washington Post, which tends to be quite critical of Silicon Valley, how a fringe AI movement convinced Washington the end is near. We've talked about this in the past. Has this fringe AI movement, has it hit a wall this week, Keith? Are you suggesting that the doomers... Have been discredited?
Well, I think they've lost a battle. The war will go on, but they've lost a fairly big battle because the entire industry has now converged around the labs themselves solving the problems. And we're talking about the effective altruism movement here. That's what The Washington Post is referring to.
Yeah, and you're also linked to a free press piece. Which is all fairly conservative publication about the quote unquote dangerous ideology between the AI warding. So the altruism movement now is on the defensive, is it? I mean, even if what happens, I mean, was it this week that Sanders and Bannon appeared at the same event? What happens if a populist figure from either the Dems or the Republicans wins the election?
Well, look, I do think that it's still true that 71% of Americans have concerns about AI possibly being dangerous, which is fake. It isn't dangerous, but 71% of people believe it, apparently, according to polls. You know, my instinct is that that number of 71% massively exaggerates people's active concerns. Because I think most people who use AI end up having a good experience with it. And so I suspect that it strongly overstates the potential for populism. That's it.
Although Andy Kessler has an interesting op-ed in the Wall Street Journal about the AI Luddites are here. I mean, we know they're here, no data centers. So I take your point. Might we be all missing the big story here and after people get bored of talking about whether or not AI is going to kill us and creating all these science fictional scenarios. The real issue is What does AI do for us? Reid Hoffman had an interesting essay this week suggesting that time is running out to stop Americans from hating AI, not because they fear it's going to destroy them, kill humanity and all the rest of it, because it's not doing concretely enough for them. And I think maybe that's a more sensible area to discuss Yeah, I agree with that. AI. I mean, what does Reid Hoffman say here? And is he right? He tends to be, I mean, talking of, he's middle of the middle. I mean, he's not on the New York Times graph, but I'm sure he would be up there with Hassabis and Nadella and Pecha.
Look, his theme is focused on the mistake the Democrats are making In thinking that having people against AI is a good thing. He's really speaking to Democrats there saying, time's running out for you guys to embrace AI rather than persuade people that it's a bad thing. And if you miss this opportunity, people will think it's a bad thing. And then all of the gains it can bring will be squandered.
Although I think you're being slightly unfair. I mean, he is saying that, but he's also saying that it's not obvious to ordinary Americans why they should like AI. I mean, you're right that many people who use it find it useful, but the argument hasn't been made coherently of the broader benefits.
Yeah, correct. I mean, we've talked a lot about the weakness of leadership, especially intellectual leadership. There's a vacuum there. Reid is pointing a finger at that vacuum. You know, he's a believer in AI. Remember, he's on the board of Microsoft.
And he sold another company, another AI company, more recently. So Reid very much understands the benefits of AI. He also understands the economic benefits that could come from it. I actually emailed him after that article and Sent him to go and read The Human Dividend. Also, Albert Wenger. There's a few people starting now to ask the question, how do you turn the corner on support for AI by leveraging its economic benefits and figuring out how to present that to people? I think the meeting with Qi this week is going to be interesting because there you've got an optimistic Trump meeting an optimistic Qi. In the face of an American population that's pessimistic about AI.
Yeah, and I think we've all got to, I mean, my view at least, as I guess somebody who is On the Wall Street, on the New York Times chart in the middle, not naturally in the middle, but on this one I am. I think we've got to get beyond, as you've reminded us week after week, we've got to get beyond this argument. It's just science fictional. It's not a serious conversation. There was an interesting piece in, again, the Wall Street Journal from Nick. Hanair, a very successful Seattle-based entrepreneur, An advertising company, he reminds us to Microsoft for $6 billion cash. He says, America made me very rich as a pro-market way to give back. It's less about AI, more broadly about the economy. He says that the answer either is neither socialism nor MAGA. What does Nick say? He talks about something called market humanism. I'm always a bit Skeptical when people use the H word because I'm not sure what it means. But what exactly does he mean by market humanism? He says it's an economics built on 21st century thinking.
Strangely, the words are very appealing, but actually the substance of his argument is the same as the argument for trade unions. He's basically lives in this world where there's capitalism and labor. And that market humanism is when capital treats labor humanely. And so he focuses on labor conditions, wages, protections, public investment in things that we all need to use. He sounds a little bit like a person you would be familiar with, Andrew, in Aaron Bevan, who created the welfare state after World War II in the United Kingdom. It isn't a very modern narrative,
A modern narrative would have to acknowledge the possibility of labor shrinking as a proportion of the things that make up capitalism. And the share of spoils going to labor shrinking as labor itself shrinks due to automation. And that requires a slightly different narrative than the narrative for labor protections and trade unions. It needs a more fundamental wealth distribution narrative, which is missing. And I think Reid gets closer to it, actually.
Although I think that these pieces by Hoffman and Hanheyer, whose politics I think are pretty close, should be read back to back. He talks about... Experiments with new products and services making life better for people, which is what Reid talks about with AI. Still, I mean, I take your point that people enjoy using AI, but it's still not entirely self-evident how AI can make life better for people. Maybe it takes their jobs away, maybe allows them to write more, but in terms of making life better, the argument is not... Maybe it can't be made, but it certainly isn't being made at the moment.
Yeah, I mean, the interesting... Intellectual puzzle is the difference between a paternalistic social democracy, which in quotes, looks after people, which is founded on the permanence of a social contract between labor and capital. So that edifice, if you will, The social contract idea comes from, and modern social democracy, which America really is a social democracy, it's not pure capitalism, doesn't survive AI. What AI does, if it's successful, open question, but I think it probably will be, what AI does through automation and robotics is remove the labor capital edifice in favor of capital. It Is Only Capital, and that's why his title's so good.
Yeah, well, Albert's in a strange place, because Albert also is arguing for AI having the rights that humans have as conscious beings. So he's in an interesting place.
Although Albert, Reid Hoffman, and Nick Hanheira are all Very wealthy. I'm not sure. Albert's quite a billionaire. He's probably close to being very rich men of the left, maybe a little bit out of fashion. But it also reminds people that not everybody who is wealthy in America is on the right.
No, absolutely. But the ones on the left actually need an intellectual framework. I thought Ezra Klein was moving towards giving them one with abundance. And it hasn't really been seized. There is a thought vacuum on the left as to how to embrace technical advance for everybody. That narrative doesn't exist.
And maybe one of the problems with, maybe not Hoffman and Han Heyer and Wenger, but some people on the left, if they've They've parodied the nature of America and this extreme wealth. Everyone's either a multi-billionaire, trillionaire, or intensely poor. My interview of the week is with Eric Zwick. He has a new book out called The Everywhere Millionaire. He co-wrote it. He's a professor at Chicago Business School, where he finds that the three million millionaires in America actually are much more powerful than All The Wealth In Silicon Valley. So he reminds us that Main Street in America, for better or worse, is not dead.
No, America is the richest country on earth and its middle class, currently at least, is the biggest. That will soon not be true because Asia's middle class is growing rapidly as well. But at least for now, that number of millionaires Basically equates to people who own homes on the coasts. If you own a home on the west or east coast, your
house... The three million millionaires you really focus on, people who own franchises, car franchises, fast food franchises in the middle of the country, dentists and doctors. So it's not just real estate wealth, it's real wealth.
It's revenue-based wealth. Yeah, well, you know, that's what happens with a healthy capitalism. And if a healthy capitalism can grow very rapidly, which the predictions about AI's impact on GDP all suggest it can, there'll be a lot more millionaires.
Yeah, it'd be interesting to see in terms of Eric Zwick's millionaires, whether the guys who've made a fortune from selling hot dogs and car parts, what the impact of AI will be on them, whether they'll be replaced by AI entrepreneurs, whether they'll benefit from the technology. Still lots to talk about. Finally, your post of the week is from Ben Metcalfe. And I'm talking to you from San Francisco. You're just south of San Francisco. He says, San Francisco isn't remotely ready for the Anthropic plus OpenAI liquidity event. He compares the OpenAI and Anthropic upcoming IPOs to previous IPOs and suggests that they're These are so much larger than those in the past. It'll have an enormous impact, already is, on real estate. Do you agree with Metcalfe? Is San Francisco where we should be looking in terms of making sense of the future? I mean, Anthropik's still supposedly doing its IPO later this year. They're talking about a $2 trillion IPO. OpenAI will do their IPO next year. How is it going to change San Francisco? What does Metcalfe say?
Continue talking for a while and I'll find the chart.
Words and timings
ContinuetalkingforawhileandI'llfindthechart.
Keith Teare
And it makes the point that the Anthropic IPO alone will be equal in dollar weight to all of the others put together, all the others combined. And the impact on San Francisco is a very large number of people with a very large amount of money, which of course only leads to one thing, which is people start buying homes and other items and the local economy, if you will, benefits from that. But his point is that this is such a huge amount of money all coming Very quickly, if Antropic does indeed IPO, that San Francisco isn't ready for the impact it's going to have. And the impact actually could be negative, as in prices of everything go way up.
Well, the prices in real estate are already ridiculous. I mean, homes are going Multiple bids, they're going 40, it's quite normal to have a 40 or 50% above average sale and was already an expensive frothy market. Nobody really understands the future. Everyone knows it's dramatic. But is San Francisco still the stage for the future as this thing works itself out with the AI revolution? I mean, I live here every day when you go out on the street, there are more and more self-driving cars. Is San Francisco really still where all the action is? I mean, Anthropic and OpenAI, of course, are both based here.
You know, it doesn't really translate into the city as a thing. The city as a thing is run by the mayor and the elected officials, as is the state of California. And funnily enough, there's an absolute separation between the wealth that the economy produces, the taxes that get paid from that, and then what happens to the taxes. So San Francisco as a city, I don't think anyone would argue, is prospering. But the economy is, and that's largely individual ownership. So there is a bit of a disconnect there, and it's actually a challenge to the Democrats, and Reid Hoffman should be very much involved in this, is how do you take the success of the economy and make a successful city?
Yeah, and San Francisco struggled with this. It has a new mayor, one of the heirs of the Levi fortune, Lurie. And he's addressed a lot of the most obvious problems associated with homelessness, but he's not very popular on the left.
Well, you know, the left almost champions the victims of the city and suggests that anything to do with removing victims is a bad thing. It's almost like you want your heroes to be victims. I don't think that's really a good way to think. I think what you want is a city that can use money to make it a better place for everyone.
Yeah, and I did an event, a live event this week at Maniz, which is in the Mission, which is a popular event place with... Jonathan Weber, who used to be the editor of the Industry Standard, has a new book out on San Francisco. It was very well attended. So this will be a subject we come back to, the fate of San Francisco, what it tells us about the fate of the broader AI economy. Lots to talk about. Another remarkable week, and no doubt, Keith, it will be No more or less remarkable than next week. So we'll talk to you again next week, everyone. Tech continues. It seems unstoppable. And Keith and I, as he as the publisher of That Was The Week, me at Keen on America, we will continue to cover it for you. Thanks so much, Keith.