Sep 12, 2025 ยท 2025 #34. 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
Optimizing for Change: Why Exuberance is Required
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker 1
Hello, everybody. It's late in the afternoon in California, Friday the 26th of September. I'll probably put this out tomorrow on Saturday. It's time for technology. That was the week. And this is the week that, according to Keith, OpenAI just shifted the interface and the power. And I thought... We usually end on the post of the week, but Keith, this week we may begin with the post of the week from OpenAI, which is about chat GPT pulse. So tell us what this new product is and why it's such a big deal, why you chose to lead. Usually you don't lead that was the week with a product.
Mainly it's because the product reinforces a narrative that we've been talking about on the show a lot, which is that AI is consumer-led, that consumers are the first customer and a massive scale. It's already a billion users a week are getting up there. And Pulse is a consumer app. Basically, if you have an iPhone or an Android phone and you are paying them, right now you have to be paying them $200 a month to get it because it's only in preview for high payers, but it's gonna be for everyone and free, I imagine, pretty soon. And what it is, is an intelligent agent that knows your email, your calendar, the stuff you've been interested in. And it gives you a daily update about all those things, including suggestions for things it could do for you to improve your productivity on those things. So it really is like having an assistant wake you up in the morning with requesting your permission to do various tasks for you, very correlated to what you're interested in.
So the interview of the week this week is Jacob Ward, who's the author of The Loop. He came on the show. The Loop came out a few years ago. It predicted some of the AI revolution. And Jacob came on the show to talk about why smart machines are making humans Dumber and Dumber, he talked about the idiocracy trap. Some people might interpret this, Keith, as being part of this idiocracy crap. Everything now is presented. You say we're consumer-led, but the consumers aren't deciding what goes into Pulse. It's chat GPT. So where's the consumer here?
Well, actually, it's 100% decided by the consumer because it's only acting on your behavior. It's a little bit like back in the old days, AltaVista used to, when he did a search, he put 10 other searches underneath, which he had learned from users who'd done the same search. So basically, ChatGPT is learning from you and then enhancing you. So the consumer is in the loop, but I think that doesn't necessarily fully discuss his point, which I think is a broader point, which is the idea that as humans...
Jacob's point, yeah, which is the idea that as humans need to do less and less, and machines do more and more, humans will get dumber. I think there's a whole conversation to be had around that. I disagree with that as well, but it's at least an understandable point of view.
And what about, Keith, the issue of privacy here? You and I were talking beforehand. You're an open AI guy. I use Claude, and this hasn't come to Claude yet from Anthropic. But do we... implicitly or explicitly give OpenAI or Claude permission to look at all our previous interactions so that it can create this thing called pulse? Isn't it going to open up a hornet's nest of legal challenges on privacy?
Well, they've covered the backside there because there's many ways for you to have private conversations that are not shared with OpenAI inside OpenAI. You can create what is effectively projects that are self-contained, for example. So you have the ability to let it see, if you will, or not see what you're discussing. So I think legally speaking, at least, they've covered themselves. Nonetheless, you're right. There's bound to be people who suddenly realize that it's able to see things they didn't understand.
Right. I mean, you're a high roller, as you acknowledge. The only people who have access now to Pulse are people who spend $200 a month. which presumably includes you, but most people aren't spending $200 a month. Most people don't use these. I mean, it goes back to all the issues that we have dealt with in Web 2, in Google, about whether people actually read the terms of conditions. This is going to become a huge issue, isn't it?
Who knows? It could become a huge hit that people just say.
Words and timings
Whoknows?Itcouldbecomeahugehitthatpeoplejustsay.
Speaker 1
No, I don't mean a hit, but an issue, not just in terms of pulse, but what these AI systems are actually doing with their knowledge of our desires and interests and intelligence.
I think that comes down to who else knows. If it's between you and what is increasingly going to become your personal AI, and even OpenAI itself doesn't know, it's just between you and your personal AI, then there probably are no explicit privacy issues. If OpenAI started sharing that with adjacent people, yeah, and there were leaks, there'd be a ton of issues. I don't think it does that, at least not on purpose. So I think for the most part, you don't need to fear violations of your privacy. I think it more comes down to, is it useful for you in helping you think through your day or the issues that you're working on? And that is a private domain. It's not a public or even a semi-public domain.
Sam Altman has an interesting essay this week on abundant intelligence. Sam Altman, of course, runs OpenAI, CheckGPT, and presumably was pretty influential in Pulse. Does he write about Pulse in Abundant Intelligence? What's he writing about?
He doesn't particularly write about Pulse because Pulse came out after he wrote that. What he writes about is much, much smarter models that... think in quotes, and he points to a super important thing, which is how much power will be required to drive the NVIDIA GPUs at human scale, global human scale, billions of users. And he talks about having signed a deal this week for 10 gigawatts of computing power. Now, 10 gigawatts is enormous.
Even one gigawatt is enormous And if you think about the infrastructure of power If we build data centers that consume a gigawatt or 10 gigawatts What is it that is consuming that power? It's NVIDIA GPUs And what is it that's forcing the GPUs to consume the power? It's token production from user AI requests So what Altman is saying is there's going to be an unlimited number of tokens sitting on an unlimited number of GPUs, sitting on an unlimited number of data centers consuming a massive amount of power to give humans effectively knowledge of everything. That's what the SA is saying. And, you know, if you listen to the NVIDIA CEO is on the B2 squared, B squared podcast this week, which is Bruce Gurley's podcast and Brad Gerstner's. The CEO is on there and he explains that you can predict NVIDIA revenues if you can predict power growth. because the growth of power equates to, you know, NVIDIA's GPU sales directly.
Yeah, and we've talked about this before, and I don't think you and I agree on it. There was an interesting piece, which I forwarded to you, didn't include in this week's newsletter, in the Wall Street Journal. Spending on AI is at epic levels. Will it ever pay off? There's some suspicion. in your sort of teleological way believes that it would these things always pay off i mean sometimes i assume they won't but they are all this spending is at epic levels and these deals the one you talked about for example between open ai and nvidia um is sort of it benefits both parties but It might be the froth, the ultimate froth on the boom. There are skeptics who believe that, that these companies are doing all these deals with each other. Ultimately, perhaps the revenue isn't there. So it's all very well, Sam Altman, talking about the needs for power, but there's also a need for revenue on his side.
Yeah, you know, I've mostly been a skeptic when large amounts of capital expenditure happened to build infrastructure. I was a massive skeptic back in the 90s about interactive TV. and made the point that I don't believe most living rooms want their TV to be interacted with because that would interrupt any shared experience in the living room between more than one human being. So I don't have a knee-jerk love for capital expenditure. I think what it comes down to is, is there real utility and value in the thing that is being manufactured or produced? And with AI, I think it's unambiguous that the the outcome is human positive to a very large degree. So the capital expenditure is probably going to turn into human attention, and human attention invariably turns into dollars. So I do think in this case, and I didn't include that article because it came out after I'd finished the newsletter, but there's one in the newsletter which talks about an 800...
So that article is kind of making the same point. And he's literally right, as is the Wall Street Journal article, if you don't imagine revenue being produced from the investment. But if you do imagine that, and I think You know, you and I talk about $10 trillion value of OpenAI in the future. You disagree. I've been saying it for more than a year now. This week, several other people are using the same number to say that it's very likely that that will happen. One of them is the NVIDIA CEO.
Well, the thing is he can do the math on his order book based on current ratios between GPUs bought and revenue generated. So he probably has some insight that I can only guess at. But honestly, I think an open AI being worth $10 trillion is a very strong bet now.
Well, I hope you're always honest, Keith. You said, honestly, we shall see. This is one of our unanswered questions that will be revealed in the next few years. In terms of your analysis this week, you mentioned monetization. You said that because of development of products like Pulse, you believe that the next trillion dollar platforms will control three assets simultaneously, attention, compute, monetization, and settlement. What exactly does that mean? And what kinds of products are they going to be?
Well, that is a little bit deeper dive on what we already discussed, which is attention costs money to serve. So if human beings are using AIs, any of the AIs, that isn't for free. There's a data center somewhere where compute, that is to say thinking, is going on. and hardware is being used and power is being used. So at the top of the tree is attention. I think we can see that that attention is growing both in the number of people using it and in the amount of time they use it for and in the complexity of the tasks they use it for. So then the next thing is power generation. which is required for the compute, that's where the 10 gig... Yeah, that's the 10 gig analysis.
Of course, there are also, leaving aside the economics of that, there are environmental consequences, more and more articles about that in the last few weeks, but that's another issue gone.
And then the last thing is, if you've got those two things, how is AI going to make money? And the current answer is a combination of subscriptions and developers paying for uses of APIs, which are two ways of driving that attention and that compute and that power use. The third way is to have a monetization platform within AI, which this week OpenAI announced that they're hiring ahead of monetization. They haven't yet hired.
They've got a business person, but what I mean is content monetization, not subscription-based or API-based. And so Fiji, the person that came from Instacart, who's running product, is hiring that position. And that implies whatever the AI version of Google AdWords is. That is to say that there are paid links that OpenAI gets paid for if somebody clicks on the link. That is probably the single biggest monetization opportunity because as you get more and more attention, more and more people are clicking on things, more and more money will flow in both directions between publishers and AI.
One of the issues is whether OpenAI can essentially control the entire market. We've had a number of conversations about this, not maybe so recently, but certainly in the last few months and couple of years, about whether OpenAI is a different kind of company. It's more than a platform. It controls the whole stack. Is Pulse an attempt to control the stack?
And the two things, one of the things that you note in your editor is the front door is migrating from search feeds to the proactive assistant. But OpenAI wants to control that front door.
Yeah, so that word bundling is key. Bundling is when you become the default. and everything happens through you. Unbundling is when somebody else builds, let's say, a real estate AI for buying and selling homes, and people buying and selling homes go through that instead of through you. That's unbundling. This week implies that bundling is winning. And that does speak to one or a small number of successful front doors. Anthropic clearly is one as well.
Well, Anthropic, I mean, they're like Hertz and Avis or the Soviet Union and the United States in the height of the Cold War. When one comes out with one thing, the next week, the other will come out with it. So no doubt, Anthropic has something like Pulse in its arsenal, so to speak.
Yes. And so I do think this is the week when bundling seems to be the hot signal. But that could easily change. It depends on what people build. It's a great time for building things because the truth is none of us really know yet.
Although it is and it isn't because if it's really true and actually... open AI control their own platform and technology, then for a startup person who wants to bundle and create smart assistants, the emergence of Pulse is rather bad news, isn't it?
Yeah, and Hux is a front door by design. It's meant to start your day with a summary of all the things you might be interested in. It heavily leans on audio because the founders come from Notebook LM, the audio podcast tool at Google. And it launched first. And within a few days, the Hux founder had to write a blog post explaining that he's not afraid of Pulse. It validates the direction. But, you know, deep down in his gut, he's thinking, oh, crap. they're bigger than us and they're more likely to win the race to become the front door than we are.
I mean, the real question is whether Sam can pull this miracle off. I mean, it's a war on many, many fronts. It's hard to imagine how many fronts at the moment. Are there any historical... I mean, when Google was building itself, it was much less obvious what they were doing. And they somehow suddenly emerged out of their garage or out of Stanford, and they controlled the entire internet. Yeah,
But it's a grain of sand in a desert. So your first law in terms of the emergence of pulse is the migration of the front door. And your second one is consumer pull is driving industrial scale push. You say compute energy rights become the moat. So what do you mean by that?
Well, so... this isn't a build it and they will come. It kind of was that two or three years ago, but now it's they're here, we need to scale up. And so by consumer pull, I mean the amount of usage and requests dwarfs the data centers and the power. And so they're racing to keep up with consumer demand, which is the opposite of of the Gary Marcus point of view. I don't mean him personally, but that negative point of view, which implies that this is all BS,
like interactive TV. Yeah, smoke and mirrors. But doesn't consumer pull, Keith, depend on price? I mean, I might be interested in Pulse, but not at $200 a month. I paid $20, and if it was included, I'd probably use it. So it's one way for... a company like OpenAI or Anthropic to control what you call consumer pull through price. Are they somehow perhaps overpricing these things to make sure that there isn't too much demand? and then their energy centers get overrun?
They are definitely doing that for sure as a short-term way to both launch new products that wow us whilst not collapsing under the weight of usage. So they're definitely doing that. I think the trend is towards free. This Computer Gradient article that's in this week's newsletter talks about The Spread of Power Needs to the Edge. It's the first essay at the top there. And it basically is talking how the new Apple iPhone 17 has a bunch of AI in it that only uses the battery on your phone. And it's clever. Things like language translation through the earphones.
Oh, I see. Just for distributed power, essentially.
Words and timings
Oh,Isee.Justfordistributedpower,essentially.
Speaker 3
Yeah. So, I mean, but the trend will be for everything to be free and for monetization to happen within the content streams. And what does that mean? Advertising again? I hope not, but there's going to be some people... What's the alternative to advertising? I think paid links are different to advertising. I don't know if you can see that nuance, but when I click on a link that's relevant to a conversation, OpenAI may be getting paid, but I don't see an ad, I just see a link.
Well, that's a microeconomy that's always been the dream of technologists. It's never seemed to me to be realized or realizable, unless it's all managed by OpenAI or Google.
I think it's gonna be managed in the center of the network. Cloudflare this week had a whole week of announcements, one of which was a stable coin called NET for transactional micropayments. Another was to fulfill their promise of blocking crawlers unless they pay, if the publisher wants that. So the trend is towards monetization with users getting it for free.
So our interview of the week is skeptical of tech. Jacob Ward, who is a journalist, a tech journalist, is very skeptical. His book, The Loop, came out a few years ago. The zeitgeist might be shifting a little bit. Noah Smith, who's always very influential, his Substack essays are excellent and they have a huge audience. has a piece this week on techs can fix most of our problems if we let it. I sent it to you, Keith. I'm sure you were thrilled by this one.
Well, I agree with the sentiment. Look, I think the more interesting one is your interview because... On the face of it, it's logical, right? If humans are doing less, you imagine humans will get done. Right, it's the idiocy trap. Right, which seems logical and it's hard to disagree with. Now, in my mind, I think of humans doing less as giving them the capability to do more of what they want to do. some of which will be leisure and non-intellectual.
What do you mean some of which? I mean, this is the old Albert Wenger argument, and you and I have talked about this endlessly. If most people, I mean, it's Marx's vision in the German ideology. If suddenly everyone in the world never had to do another minute of work, Are they all going to start reading Shakespeare, doing podcasts? Unlikely.
You know, Marx's famous quote segments the day, and the last part of the day is a philosopher after dinner. So it does include reading and thinking and discussing.
And farming and doing all sorts of other creative things. I mean, he sees this as being so creative. I mean, maybe it's another conversation, but, I mean, even you have to acknowledge that If suddenly we were liberated from work, it wouldn't necessarily mean that we'd all start our own podcast or music label or... No,
Think of ourselves. I definitely, and you can speak for yourself, but I definitely feel cleverer two years after I started using AI. I'm doing less of what I used to do I'm learning a lot more things I didn't know I could learn, including practical tasks. And I definitely don't feel dumber, even though AI is doing some of the stuff I used to do.
It certainly makes me more humble, which is quite an achievement, makes me realize the limits of my own intelligence when it can digest pieces that would take me days, sometimes weeks to understand. It can do it in about 10 seconds. I'm not sure whether always the conclusions are coherent or right, but it is incredible in that sense. In that sense, it's like magic.
I gave it the code base for signal rank scoring mechanism last week and asked it to write a mathematically based paper that could describe the algorithms to a math proficient reader. And it produced this 25 page paper that was completely true to the algorithms in the code, but was written like a math professor would write it. And I tried to read it. And I could read everything except for the equations. So I copy and pasted the equations back into another area and said, tell me what this means. And it changed it back to normal language. And it was perfect. And I couldn't have done that. But in having it do it, me read it, I learned a lot from it.
So back to the Smith essay, he is a bellwether in the sense that he's influential and smart. Do you think that the zeitgeist is shifting a little bit back towards a belief in technology? Are we going to see that? There's so much skepticism these days.
The zeitgeist is completely divided between the Charlie Kirk generation that is pro-tech, pro-capitalist, believes that you can intervene in society to good outcome, and the people who are more like I was when I was young, which is the critics, usually thought of as critical thinkers who focus on societal problems like race and class and gender and believe big tech to be a problem. So I think we're at that point where there's no consensus here. I'm definitely on the side of Noah Smith, but I don't think that's yet a majority opinion.
Yeah, and of course, the Smith piece, for people not watching, has the image of a self-driving car, one of those cars owned by Google on its front page. I had an out-of-town guest this week staying with us in San Francisco, and they were utterly amazed by the abundance of self-driving cars in San Francisco. I think that the self-driving car is probably the best example of how technology can make the world a better place at the moment, isn't it?
Yeah, you know, I had a meeting this morning at the Marriott at Fisherman's Wharf in San Francisco, which is for me about a 40-mile drive. We could have done this show live, Keith. Oh, we could have. I hadn't finished the newsletter, though. I came back to do it. But my Tesla drove me self-driving in both directions.
No, it forces you to pay attention. It has a camera on you to make sure your hands are available if needed. But I could definitely think and not have to drive.
Yeah, I think Waymo is a really interesting company. It may be one of the reasons why Google stock is doing so well. Speaking of Charlie Kirk, last week we... focused, we didn't do AI for a change. We talked about the implications of the Kirk assassination, particularly on the Jimmy Kimmel front. You have a section in this week's newsletter on media. Were you surprised with how, I don't know if ABC folded or pushed back on the Trump administration, but everything seems to have changed. We wondered why, or I wondered at least, why Kirk didn't simply shift his show to YouTube. It's obvious now that, sorry, Kimmel, that he's negotiating with ABC. So is ABC, what's your interpretation of what's happened this week on the Kimmel, ABC, Trump, Kirk front?
I think what happened is the backlash against the decision to suspend the show was more scary than the administration's feelings. 400 very, very well-known people signed an anti-Disney petition. They're all potential employees of Disney, people that Disney needs to have in their camp. And so I think they made a quick calculation of which way the wind was blowing. and decided that they needed to put Kimmel back. And I just saw today Sinclair, the owner of a lot of the stations that still hadn't put him back on air, is now doing so this evening. So it looks as if Kimmel is going to be fully back. His monologue on the first night was very funny. But he did skirt over the fact that he totally misjudged.
Well, I think he apologized. He recognized he'd made an error and he made it clear that he wasn't celebrating the assassination of Kirk. I thought, I didn't watch it all, but I saw that highlight. You also have a Wall Street Journal piece, classic Keith Teer piece on when the left censored speech, a piece about Google admitting that the Biden administration's pressed it to silence COVID critics. Does that justify, in a way, in your mind at least, it doesn't in mine, but does it justify what Trump's trying to do now with media? Does it suggest, well, under Biden, the left did it, so we can do it now under Trump?
I think what it tells us is that instinct we all have, that all politicians are narcissistic, self-serving ideologues, has a lot of truth in it, that you can't trust a politician with principles I think you have principles, I have principles. Speak for yourself, I don't have any. But I think we both wouldn't agree with what Biden did, and we don't agree with what Trump tried to do.
Although a defense of the Biden people is that they thought that they were saving people's lives, or at least that's what they would argue. I'm not sure whether it's always true.
I believe that, but That's the whole point of principles. You have to stick to them even when it's hard. So, you know, I don't believe they're lying about that, but I do believe they were wrong about it in retrospect, but there was no way of knowing that at the time. But, you know, freedom of speech is usually, you know, when you don't like something, the best remedy is more speech.
Yeah, and there was an interesting piece in The Times, which was too late for this week's newsletter. fearing that Trump might try to start interfering with the Google search engine in terms of its results. I don't know if you saw that. Is that conceivable?
Yeah, although just as there was a pushback against Disney on... on the Kimmel decision, you would see exactly the same thing. It's hard to imagine. Finally, last week, the interview of the week was with Marshall Poe, who talked about Gutted and Glutted, the dire economics of podcasting in the AI age. You watched it. It was an interesting interview, although you didn't agree with Poe. Will the product of the week, your startup of the week, will Hux amplify the gutted and glutted nature of podcasting, what actually will it mean?
Well, it's creating synthetic podcasts that are topic-specific on a daily basis. Sounds like my show. It isn't very good yet. I mean, you know... Yeah, but it will get good. That's one thing we know with AI, that it radically improves. It will almost certainly get better. And, you know, they'll probably introduce... They have about 50 voices already compared to the two that... Do they have yours and mine? They do not, but they need to get those, Andrew.
Yeah, well, we need to call out. I'll have to have a word with my wife. We'll call up Google. Should we be fearing, Keith? That was the week, our weekly summary and conversation. Will we be able to use Hux, maybe borrowing from your newsletter, to create a show without having to spend half an hour on a Friday afternoon doing it?
Here's what I suspect. I suspect that news... or backward-looking, historically speaking, backward-looking content will become more and more automated. I think anyone trying to shape the zeitgeist will remain primarily human.
I couldn't agree with you more. As always, that was the week for tech. A big week, at least according to Keith. It was the week. that OpenAI shifted its interface and the power changing everything once again. We shall see if that's true, Keith. We're not doing a show next week because we're both on the road, but we will see everyone in a couple of weeks. So keep well, everybody. Use Pulse, use Hux, and maybe make Keith and I redundant. Thanks.