The Next Renaissance: Zack Kass on How AI Will Reshape Everything

Zach, thank you for joining us. Because look, this is an AI platform where I'm employed and we utilize large language models and AI in all kinds of ways. This is your expertise when you were go to market, when you were the head of that for OpenAI. Help me understand what did that mean? What were you negotiating for them?

Most of my work during that time was sort of in three pillars. One was obviously building the teams to lead sales, solutions, partnerships at the time, customer supported marketing. Another was negotiating and working and supporting with our largest clients. And the other was strategy. How do we actually serve these customers? And insider strategy, you know, partnership strategy, sales strategy, etc.

When you look at OpenAI now or chat GPT I scale of 1 to 5, 5 being an A, a 1 being needs drastic improvement. How would you grade where the model is performing today?

Well, I want to make sure I understand the question. Are you asking OpenAI as a business, ChatGPT as a product, or the model itself powering as a product?

I mean, just because so many of us now use it, whether we use it intensely or not, so many people have now experienced it.

And you're asking how I would grade it in its performance as a model or its overall as a tool for

people who want to leverage a large language model?

I would say five out of five. It's hard to imagine a better, I mean the, the two, if you gave it a four and a half, which I think you could, you could do reasonably, I don't think you could give it anything less than that. It's, it's growth speaks for itself. It's. First, let's talk about what it's done. Well, among the most impressive things that OpenAI has done in the last three years, you'd have to probably consider the precipitous decline in cost of the models themselves. GPT4 was $60 per million tokens when we launched it. Today it's less than $1.50. And that cost isn't just being subsidized by the investors. It' massive reduction in latency and cost to serve the models themselves. And so what you're seeing is this remarkable means of providing access to more people because now ChatGPT can be given away for free and you can figure out other tiers for service. Second thing that I think is so impressive about ChatGPT is its multimodality. You can talk to it, it'll speak to you, you can give it images and video, you can give it CSVs, you can give it Google Docs, lots of things that I think we would have thought in 2023 we're going to come, you know, a ways down the road, suddenly showed up overnight and people became, you know, to expect it really pretty quickly. Third thing, the context window is massive. I mean, it's, you know, I think 100,000 tokens at this point. We have an enormous context window, which means you can give them, you can give the model a lot of information and it can give you a lot of really good information back. And that's done incredible things for things like deep research and making sense of these really incredible large canons of information. And I think the fourth, and this I think often gets overlooked, but the improvement in things like hallucination and safety refusals has meant that it now is like a very good working product for kids, adults and professionals alike. OpenAI recently added this parental safety feature, which I think is well served in helping make sure that parents have a view into what their kids are doing, especially with respect to things like dehumanization and chat psychosis. So, all in all, you have OpenAI doing these really incredible things that I think we would have thought would have taken a lot longer that are now happening really quickly. That alone probably gets it a five out of five. Where you could knock it and I think you could is if you sort of poke apart at, well, you know, is it. Is it actually better than Claude code for code engineering? Maybe not quite. How well does it actually integrate with other business solutions? Still, still not yet. ChatGPT teams and enterprise is still sort of this like standalone service and it has still.

I mean, it's improved. I use it and it's improved. It still does hallucinate on occasion. I have to chastise it when it gives me output that is in contradiction to the prop that I created for it.

Yeah. And I think again, I think that where it really struggles is when people treat it as this close combat weapon. I think that where GPT5 was most misunderstood in the release is everyone said, well, it's not that much better at writing, or maybe it's worse at writing than GPT 4.5. And my point is it's not really how this is intended. It shouldn't do all of the thinking for you. It should do the really hard computational work so that you can actually do the writing and you can do the really thoughtful work the last mile, so to speak. So anyway, I think it's done exceptionally well. And then if you want to give it a big knock and I think you could on this, you could say, well, listen, the Other issue is we've entered this world where it's become so sick a fantic and people are building, you know, we're seeing this sort of rapid rise of chat psychosis and I think that's pretty concerning and something that we need to address really quickly. Thankfully, I think OpenAI seems to be taking, you know, very immediate and swift action to, to help this.

So. So I want to talk a lot about you. In previous podcasts and in some of the interviews you've conducted in the past, you talk about the humanization and the optimistic ability, you know, the next hundred years, the unleashing of humanity, the potential of that to take place because of AI. But I just want to get to one business concept that you brought up in your interview in cnbc and there are two issues here. The first was you said there is a non zero chance that China says we will reinvent the architecture. And then you went on to say we're going to change the rules, the nature of the game when we talk about China and AI. And I had to look up non0 is vernacular I wasn't familiar with. It means there's a slight chance this will happen. Am I interpreting that correctly, by the way?

That seems. Yeah, that seems fair.

When you say that China will reinvent the architecture, what did you mean? Because we have the clampdown on the high power chips from Nvidia to China and are you saying they might not even need them?

I think if I recall this particular interview, what I was referring to is this. Today we look at the rules of the game as sort of being roughly the following. In order to build the next frontier model, you need an enormous amount of compute, right? Let's call it a sovereign nation amount of GPUs. You need a sovereign nation supply of geniuses, right? Brilliant machine learning researchers, and you need a sovereign nation supply of energy to sort of power this compute. The data rules are already starting to change, the scaling laws are sort of breaking already. And we're discovering that in fact bigger models are not necessarily better. And so whereas I would have included data in this a year ago, it does seem like better data is necessary, but not necessarily more so. Given this, the assumption now is, okay, well, the west and in particular the United States is in this very advantaged position. On frontier model research, this has been the prevailing sentiment and I think that this largely would have been considered true until of course, Deep Seq produced R1, a state of the art model that required a lot less, maybe by almost an order of magnitude less compute. And today costs 50 cents per million tokens to run, which compares to OpenAI's $50 cents on its hosted model. So it's just more compressed, it's less expensive. What I was proposing in that interview is we make these assumptions on the basis of the transformer architecture. What we know today are sort of the mechanics of the game that define these rules. In a world where we discover new architecture and we are flirting with some sort of next generation architectures that look like the transformer but aren't necessarily the transformer, we may wake up one day and say, oh, actually this is not, you know, the GPU is no longer the lifeblood. There is something else now, and I'm not proposing that we are near that, but I am proposing it's possible especially, and this was my argument, I think in a world where China doesn't have the requisite compute, it's going to rush to find a more efficient architecture.

Well, you may. I'm going to interrupt you because there's a great analogy used in explaining that you used Elon Musk said if Elon Musk is going to have a supercomputer that needs a million chips first, can he get a million chips? And what you were saying, if I recall the interview correctly, was, look, China's going to figure out a way to get around needing the 1 million chips.

Yeah, look, they are. What we have learned over and over and over is that scarcity breeds innovation. This is just sort of largely true. And especially if you're China, which has a plethora of data, a plethora of energy and a plethora of brilliant researchers, they are spending a lot of time now trying to figure out, okay, how can we change the rules of the game? And in this case it means probably building a different type of architecture that becomes less compute dependent. And this is what we are seeing out of their open source frontier model providers and by the way, China's open source library, open source community. I would score as a 9 out of 10, where I would score the west or the United States and Europe probably a 6 or even 5 out of 10.

What you just said about China figuring a way to do this, does that impact. You had another quote that jumped out at me in that interview where you said that today we live in a world where compute is very good and pretty cheap. Tomorrow we will live in a world where compute is very good and very cheap. Does what China's doing, do they contribute to that or do they circumvent that?

Well, so there's another important Distinction to be made here. There are two types of compute in this case. Well, and now there's a lot of different compute. We have TCUs, tensor processing units, we have GPUs, graphics processing units, and we have CPUs, computer processing units. What's interesting is the CPU is already abundant, right? The cpu, the device that powers our phone and our, in most cases a laptop, is powerful enough to do a ton of really interesting work and super energy efficient, right? You do not think twice about plugging your phone into a wall and no one, you know, you don't show up at someone's home and they whoa, whoa, don't plug your phone in. That's going to, that's going to offset, you know, that's going to destroy my energy bill. The CPU remarkably, is actually now powerful enough to power later generation models, in particular GPT4. I can run a GPT4 model on my device disconnected from the Internet, which is pretty incredible. It's slow and it drains the battery, but it works. That sort of suggests that we are entering this new world where, where GPUs are certainly going to be necessary for, for quite some time to actually train the models. The CPU may actually soon be sufficient to run the models, what we call inference. And the current rate of sort of the precipitous decline already of the compute power suggests that we should feel like we live in a compute abundant world at some point soon.

Let me break this down. I am not the computer genius that you are and that the people you're surrounded by are you. From a business standpoint, are my costs of accessing these tools going to come down? Is that what you're telling me?

Well, they already have. I mean, let's be super clear, right? Like first, if I had tried to describe to you GPT5, five years ago, where would I have started? Like truly, how would I have begun to explain to you what that sort of form of artificial intelligence would look like? And then if I told you that it would be free and you could run it on your device, you'd be like, well wait, that's pretty remarkable what had to have happened. And the current trend is that the models are going to get more compressed and more efficient on a per, let's just call it watt basis. How much energy does it take and how close to the edge of the device can you run these models? The frontier is going to require being trained at huge data centers. So let's just call that GPT6 for, for hypothetical, the GBC6 is going to have to be trained in these huge data centers powered by GPUs. But GPT5 soon is going to run on the device. And this suggests that the models themselves are going to continue to decline in cost as they have GPT4.$60 per million tokens today, a dollar and, you know, whatever, $0.30, $0.55 if you run it on Deep Seek. This is the access that everyone has been praying for for most things. And in case of water, it took thousands of years. In case of food, it took hundreds of years. In case of electricity and the Internet, it took dozens of years. In the case of AI, it's taking five. Right, Five years. Still, all you need is access to the Internet for you to have access to abundant intelligence. And we call this. In a paper you may have read, we call this unmetered intelligence. The idea that access to intelligence will feel abundant doesn't mean that everyone will have it. Because just because we're all literate doesn't mean we all read. Just because we have the Internet doesn't mean we all do research. It means that if you want to have access to unmetered intelligence, you can.

You actually talked about. This is a, you know, a sidebar. But you talked about in one of the podcasts you were on how people waste so much time on the screen. We talk about how we're on the screen all the time working, but people waste a great deal of time on their screens. You brought up reading because you have a book coming out, the Next Renaissance AI and the Expansion of Human Potential. And in the literature that's talking about the book, the question now isn't, quote, are we building a better future? It's, are we building an exceptionally better future? This is why, before we started recording, I referred to you as an optimist. Am I correct? Are you an optimist? And are you? Because there are a lot of people out there who feed the media vacuum with doom and gloom about where AI is taking us. And I don't hear that when I listen to you on podcasts.

I'm an optimist. And I'm an optimist in part because my genetic disposition, I was predisposed at birth to being an optimist, that I sort of grew up believing that the next day would be better than the last. And then when I went to college and studied history and computer science, I discovered that there was, like, a lot of evidence to actually support this, you know, guttural intuition. And I'm always struck by the fact that if you are a cynic or a Pessimist. There's really an enormous onus on you to prove that the world is, is, is deteriorating or in decline. And, and, and that in fact, all of the evidence, I mean, so much evidence points to the fact that we live in this gondola that sort of is moving up a mountain that is swinging back and forth. And at any given moment we are, we might feel in the gondola like we are, you know, there's tremendous turmoil swinging backwards. We may be because the winds, but from the ground, if you observe the gondola, you see that it just keeps moving up and to the right. And I press people, when would you, what date in history would you rather be born than today? And it's very hard for people to confidently pick any date other than the present because they can recognize, even with our rosy retrospection or just deep desire to romanticize the past, that it was deeply flawed in ways that we probably don't appreciate even now.

John Stossel, I don't know if you're old enough to remember when he was, of course, at abc, but he has his own podcast. He talks about, for instance, we're feeding. There are what, 9 billion people on planet Earth. We now have less starvation among the 9 billion than we had just 30, 40 years ago. I don't want to get into climate change discussions, but he points out that in the climate change discussion where we all get very afraid about what's happening, we talk about how things are heating up and the deaths that that brings about, but we never talk about the number of deaths attributed to cold weather that are diminishing. It gets to this issue of how are you looking at the situation? So you describe yourself as an optimist, and I'm curious when you hear very well known former CEOs or AI experts saying this technology is very dangerous, and you refer to different possibilities of danger, you talk about the threat of the low resource bad actor has never been more terrifying. What's the appropriate focus of our concern over AI as opposed to our ridiculous focus over the concern of AI?

So I will start by saying I think that it is very, I don't fundamentally disagree with most people who sound alarm bells about most things today. I mean, I think there are some alarm bells that are truly red herrings. We shouldn't, we shouldn't be talking about them because they're wasting our attention. And, but, but I don't, I don't sit around sort of dismissing people's concerns. I do point to all the solutions that are currently underway and Humanity's sort of resiliency in solving problems that it creates for itself, which we have a very strong track record in doing and have done so for millennia. Now, when it comes to AI, there are sort of two really strong loud battle cries at this point from the alarmists, and this list is fairly long. But it's sort of led by Jeffrey Hinton, one of the godfathers of AI, Stuart Russell as well. And between Geoffrey Hinton and Stuart Russell, what they have basically said over and over is actually we're sort of approaching this sort of doomsday scenario. And Stuart Russell's increasingly become very bleak in his estimation. What they are predicting is a downside scenario in which we build a machine that. Have you heard of the paperclip theory?

I have not. I'm the Terminator generation.

So, yeah, well, okay, so great. So there are two perfect segue. So there are basically these two similar but very discrete theories of what we'll call misaligned superintelligence. In one, we build a machine that is in fact malevolent, that discovers that it is better served if the humans just weren't around, and takes active measure to harm us either sort of irrevocably and, or permanently. And that is not dissimilar, you know, minus the time travel is not dissimilar to the Terminator theory, which says, you know, we're going to build this, this, this very mean super intelligence. And then there is another, which is the paperclip theory, which says that misaligned superintelligence doesn't need to be malevolent. It actually can be accidental. And if I told an AI superintelligence to build to. To maximize paperclip production, that was my only guidance. And in order to do so, it actually just started to convert all resources into paperclips and accidentally just converted the world into. Into paperclips. Well, that would be. That would also be very bad. And that's misaligned superintelligence. Those two theories are very prevalent in today's media because they capture a lot of attention and a lot of fear, but it's actually quite hard to extrapolate to them today. It's very, very hard to arrive to them. And the steps that would have to be true and sort of the number of safeguards that would have to be broken would be pretty remarkable. Then we start from down from that. And I don't mean to. To dismiss that outright, but I do mean to say, like, I don't spend a lot of time on those two outcomes. Because they, they aren't actually, in my opinion, the really scary eventualities. But the level down from that does get a little weird. And the next one down from that I think is the most pressing, the one that we should be talking about, which is empowering bad actors.

Right.

And this is something that doesn't get as much attention, but deserves to get a lot of attention. And there are two kinds of bad actors as far as I'm concerned. High resource bad actors, nation states and Bond villains. And low resource bad actors, psychopaths and criminals. About 4% of the population is psychopathic. Another 5% have figured out how to make crime payments over the last century and are doing so now with exceptional ease. And in the same way that we are building technology that's making a lot more people a lot more productive and 91% of the population is doing really impressive things with it or, or at least good things net neutral. If at worst there's a percentage of the population that has now figured out how to very inexpensively tear at the fabric of society, mostly through financial crime. 500 billion was lost in 2024. We think the number is going to be closer to 700 in 2025. A lot of it's retail. Your parents, my parents, people are basically being stolen from with exceptional ease. And one of the crusades I am on is to sort of challenge everyone that we need to spend a lot more time having really honest and open conversations about the fact that we have created a very tolerant society and that what we need to do now is actually seriously consider using this technology to identify and go ahead, you're getting to

an issue that I want you to talk about. Let's break it down simply into to pieces the tolerant society and the expectations. People thrive. As you've pointed out in other podcasts, when you know that the rules are enforced and that you can operate within the guidelines. First, on this theft issue you just brought up, I interviewed Bernie Madoff in jail. And so. And I used to cover the worst who I used to think, you know, in my scheme of things, financial crime is up there with murder because people can't recover sometimes. What you just threw out, by the

way, I like this take. What you just said is a very good take. And I think it's even, I think it's more, more pronounced than that which we should, we can talk about. But I think it's not as the people can't recover, it's that it tears at the fabric of society. Financial crime creates low trust society where it becomes very Hard to do business on a day to day basis because everyone is looking over their shoulder.

Where do you want to invest? Venezuela or the United States? I mean, it's that simple. The reason the United States is exceptional, I believe among nations and I don't know if would agree with this is the rule of law means that my property is my property and that it will be my property tomorrow. I can risk my, my investing because also for the consequences. I can risk my investing because if I have a gain, I don't worry about somebody seizing it, who's jealous.

That's what we're going to do. I completely agree with this and I would go a step further and say this is true. This is why people like neighborhoods and like broadly societies. You like going to a place where you can predict the behavior of the people operating in it. And this is true economically and this is true socially. And what you want to be able to do is live in a society where you know that if someone does something to harm you, they will be punished.

Like people on the subway. You use the subway analogy, which in New York everyone talks about. I see homeless mentally ill people every day on the subway. And you talk about in Japan that individual. It's not politically correct or I don't know the terminology today, but locking people up with mental illness in facilities that would protect them, but also protect society from them is not considered appropriate right now. And yet we have people who don't use the subway because of it.

When people talk about Japan, they talk about this incredibly well functioning place. And what I have to, what I get to educate people on is that Japan ranks number one in the world among countries that institutionalize people who are, who are struggling with mental health. They average about 200 per 100,000 and they don't suffer sort of suicidal empathy. And in fact have thought of these institutions as places to help rehabilitate people, but to hold them to an exceptionally high standard. Because in Japan tearing at the fabric of the social order is a, you know, a capital offense in effect, right? It, it tears at the very nature of the Japanese spirit and that is, you know, unconstitutional. Whereas in the United States we said, listen, we don't want to, we don't want to put these people out. And by the way, I'm not proposing this is, this has an easy solution, it just has a very obvious one which is reinvest in the mental health facilities that we long ago forsaked and actually build places that we can, we can help people versus letting them sort of make the commons and the places in which we live feel unsafe when

we talk about this and AI's impact on all of us in ways we don't even really appreciate just yet, or what's coming down. You pointed out on a podcast that democracy may not be the final state, and then you went on to say capitalism might not be the free market solution. How does AI, because you were talking about the social contract and how AI may redefine that social contract. How is that going to play out?

So first I want to say something on the last topic because it dovetails into this, but it also is important to say one of the other things I crusade on now is that we use these tools not to identify and prosecute bad actors, but also to protect the freedom of dissent and freedom of expression and freedom of religion. I think it is equally important that we, especially in the United States, but elsewhere, use the technology to fight the bad acting and also to protect and celebrate the freedoms that we love. Otherwise, what's the point? And I do think that the potentiality of a police status is to me also very scary and something that should be taken really seriously. AI should not be used to sort of to, to, to scare political dissent or to shape societal behavior outside of that which, which harms society. That being said, I also do wonder, and now we can go into the question you asked. In a world where the cost of goods and services will likely come way down, there are going to be three that are particularly pronounced as being exceptionally inflated. Healthcare, education, and housing. Most of our lives have gotten a lot less expensive, except for those three things. And those three things are making a lot of people feel very pinched. And what I tell people, or what I like to talk about is the fact that those three things are not expensive because the technology didn't solve them. They're expensive because we policy them into oblivion. We basically created a world where in order to have access to, to great education, great housing or housing for many people at all, or health care, you need to either be exceptionally wealthy or very privileged in some, you know, specific ruling class. And I have sort of borrowed among other people a lot of sort of thinkers this idea that if we can figure out actually how to unlock those three services, these sort of, what should be probably considered rights of someone in a well adjusted and modern society, then what we actually need to start asking is as the cost of goods and services plummet, what is the right way to manage commerce? I'm a free marketer. For what it is worth, I think that we should let the market do A lot more. It's not obvious to me that capitalism is the free market solution or the optimal free market solution in 100 years. That I think is certainly going to be up for debate, as is most of these. Most of the social contracts constructs that we operate in today, namely the idea that colleges, among them, the idea that colleges is this eventuality among them, the idea that you must work in order to be happy among them, the idea that having kids is some unbearable burden, when in fact it may be one of the great honors and importances of a functioning society. So these are the things that I think are going to be up for debate, especially as unmetered intelligence starts to apply to production of goods and services.

And you talked about. And I want to get back to your book, but I also want to get back to some of the themes of that column you wrote recently. But a future conflict between people and the state. We look at the world right now state versus state. But you on that podcast I listened to, pointed out that future conflicts might be about individuals or about neighborhoods against our governments brought about AI's impact on us.

I think when people ask, people will often ask me, what do you think about east versus west and the AI race? And I do my best to disabuse people of the idea that this has to be a race. I don't think we should talk about it as one. And if it must be considered a race, then I think it's a race where the prize is shared equally. I think that the achievement of a superintelligence is inherently something that's going to benefit most people in the same way that I don't really care who cures cancer, I don't really care who solves fusion energy. I want these things for humanity. And we now live in a sort of a global state where most people have a vested interest in everyone else. I mean, whether people want to recognize this or not, this has just become true. An economic collapse in the US would have a profoundly negative impact on China and vice versa. We have to acknowledge this. What I think is increasingly also true is that we are approaching a world now where people may soon discover that a lot of the things that the government has been promising them is achievable imminently and readily through free market solutions. And that a lot of the reasons that we don't have access to some of these things is in fact special interest policy that sort of protect cartels and groups.

We protect sugar. We can get sugar a lot cheaper. But we.

Well, yeah, I mean, that's Every people love that one. But man, the list is long. I mean very long. I don't.

Losers.

I wrote a. Well, we are about to publish a paper called Illegal Automation which outlines all of the things that we are not allowed to automate in the United States and the number of jobs that are protected as a result and the cost associated with it, roughly. And it's very sobering you read it because we spend all this time saying, oh, if AI automates so and so's job, you know, I'm going to feel so bad for them. And that's true. And they say, well, AI is just going to make the wealthy wealthier. And I say listen, and these things may be true, but in fact, let's just start at the principle, which is not enough people have access to all of these things. And the reason for that is that we've locked them up and we're going to have to face a reckoning pretty soon to the fact that so much of the current system is built on special interest policy that's just been layered on top of itself such that we've basically forgotten why it's there in the first place. And it doesn't serve the average person, it doesn't serve the population, and we're going to continue adding more and more of it until we start to let the technology and the market actually do these things.

Could your column rethinking the focus on artificial general intelligence, what Gen AI really brings, it would seem to me with what you just said, we're kind of on the edge. We could either regulate AI to the degree where we create new winners, but then jeopardize several losers with the regulation, or we could go the other way and AI could help us dismantle regulation that actually creates disadvantages for people. Which is it?

Well, I think both. But I guess what I would say is I'm not naive. I don't actually propose that that AI is going to be the solution to policy and special interests. That is going to come from something entirely different. It would require a Singaporean style upheaval where we rewrote actual practices of authority and law in order to protect ourselves from special interest invading our policy making. But I do think this is the crux of what I believe that the automation that AI provides is going to one first make most of the things in our life feel even more inexpensive than they do today. The computers that we use, the screens that we use, the content we consume, the devil deflation.

This is the theme of that column you wrote.

That's right. The flights that we get on, all of these things where I think we are going to, where people are going to get really, and I mean really pissed more than they already are, is when they start staring at their life and saying, wow, flat screen TV is literally the cost of a, of a, you know, a decent meal out. And yet. And I can fly across the country for $80, but if I go to the emergency room tonight, it might literally bankrupt me. And there's going to be. We're not there. I think we're. We have yet to approach the fever pitch where people go, what the hell is going on? Why are all of these things so inexpensive? And Mark Andreessen has talked about this a while, but I think we really have the opportunity to do it now. Why is everything in this deregulated column so inexpensive and everything in this regulated column so expensive? Maybe consumer protectionism is not protecting me. And I'm not proposing that there's going to be some libertarian uprising, but I am proposing that people are going to start to point out at the things that feel exceptionally expensive and really challenge, I mean seriously challenge why they're so expensive in a way that forces policy changes.

Does AI bring about the ability of people to challenge the status quo?

Here's what AI. Here's my belief on what AI will do. I sort of point to AI as being this potential force for a rebirth of civic engagement. Two things happen when because of GPT5, frankly a context window this big. The first is lots of magic is going to start to get introduced into consumer products and experiences where in deregulated environments, if you can do something really magically for your, for a customer, companies are going to do it within the next year or two years. And it's. You're just going to start to see a bunch of really interesting automation happen that costs you nothing. And you're going to be like, wow, this is really amazing. I wish that DMV could do this. Pick a government service. The other thing that happens is the next time that Congress puts a 20,000 page bill in front of us, more people than ever are going to put that thing into ChatGPT and say, where in this bill is policy that has no business being here? Where in this bill? And by the way, this is already happening. A journalist did this recently and they were like, wait, why is there this weird thing for Alaska? And then we discovered, oh, the Alaskan senators needed a kickback to pass the policy. And this used to never get any attention and now it can. Muckraking has never Been easier, especially when it comes to congressional policy. And I think this is going to be very inconvenient for a lot of policymakers who are so accustomed to sort of squeezing things past us and saying, don't look over here. But in fact, now we get to. Now we get to examine everything because we have the sidekick that can read everything and say, oh, this has no business being here. And by the way, not just is this is no business being here, but we can say, model this for me. How will this play out? What are the cost of this?

It does the data chunking and then it gives you the extrapolation. But in that, again, in this column you wrote, help me understand. When you say AGI is overhyped when we talk about AI, you were talking about tokens in the beginning of this discussion. You know, when I first got into AI, it was a big issue about, okay, how many tokens? What are we paying? Who are we leveraging? I don't think anyone worries about that anymore. At least my current employer seems to be very comfortable with what we're doing. But we don't seem to have a common vernacular, a common language. I'm not going to use big words here when we even say AI. When I say AI, my friends who are in different fields are thinking something totally different than what I might be trying to convey. Is that still accurate? Are we coming to a consensus? This is what it means?

Well, first, this is a general technology that I think is actually quite hard for anyone in the field to define because it is so big and is doing so many things. With respect to AGI, Artificial General Intelligence. It was coined as a term in the early 2000s to describe this place where AI wouldn't be good at just specific tasks or narrow tasks, but it would be good at general tasks, meaning that it could rival a human in sort of most economic function. The reason that AGI, the reason I sort of provocatively propose that AGI is overhyped is not because the potential of AGI is overhyped. It will, you know, it will have profound consequences. It's that the definition of it is impossible to agree on, and therefore we should stop talking about AGI as this marker. What we should instead do, I propose, is point to results. One of the important results that we need to see is the cost of the actual service. The inference coming down. The reason your employer no longer cares about tokens is because we've made them so inexpensive, you can't consume enough tokens to really bother anyone on a, you know, on a bottom line basis anymore because it's just so cheap. The next thing that we should want to see is actual cost of everything else start to come down. Access to things, especially in deregulated environments. And the third and most important is scientific discoveries. Right. The great marker I think, of unmetered intelligence is are we more and more discovering novel frontiers across, you know, adjacent sciences. And, you know, this is, I think, the great. One of the great promises of AI, which is the ability to help humans figure out a lot more about our known and unknown universe.

Let's talk about the book and what propelled you to write this. And do you have a published date yet?

January 13th comes out. You can pre order it now on Amazon or my website, but you won't be able to read it until that date.

All right, so but when we talk about the book, what is the message you want people to glean from what you've created?

So I wrote the book with a very specific intention, which was to provide a sort of, at the time, a counter narrative to sort of these prevailing dystopian ones that were really good at capturing headlines. And what I was discovering as I was spending time with more and more people inside of AI and sort of around the world is that optimism wasn't actually unpopular. It was shared optimism that was unpopular. Right. People are very willing themselves to be excited about a future and actually quite nervous about sharing it. Right. It's not a very popular public position to say things are going to go really well. Pessimism has a coolness factor, but it also shows this sense of paranoia and stewardship that people want to convey. And so it's sort of critical that a lot of companies present all of these downsides so that they don't look irresponsible. Sam talks about this, by the way. Sam Altman talks about the fact that he's like, look, if I told you all the things that I thought would go well, I would sound like a crazy person and you would dismiss me outright. And so instead what I have to do is sort of be much more measured to my upside and much more boisterous to my downside, because that's how you'll take me seriously.

Stop. Stop. What is going to go well or better? Because the book is the next renaissance AI and the expansion of human potential. Help us understand in concrete, visible terms what's going to be better. And when I say in the future, the future we will all experience, hopefully we will all live the next five to 10 years.

Yeah. So we describe what we call the promises of this chapter. And I think that the promises all stem from this idea of the expansion of human potential, that at the limit, the thing that is sort of most regulates and most rate limits our ability to do more, is the fact that we just aren't quite smart enough to solve all the problems that we would want to solve as quickly as we want to solve them. This is why we don't have abundant energy yet, because we haven't figured out fusion energy. We don't have abundant foodstuffs yet because we haven't figured out how to make vertical farming more efficient, et cetera, et cetera. And the list goes on. Why does cancer still such a blight and all these other issues that we face. But the opportunity that AI presents us is specifically that is to expand the ability for one person to do or accomplish a given task, whatever that is. And that, of course, will be true on an economic basis. By the way, we're 95% more productive today than we were in 1995, or about 90% more and 400% more than we were in 1960 and about a million percent more than we were in 1800. So we're getting more economically productive all the time. But now, for the first time ever, we're starting to see what might happen if we supercharged all the scientists, not just the knowledge workers. And so one of the first and I think most exciting promise is this idea of novel scientific breakthroughs and scientific discoveries. What would it look like if we started to cure cancer or we started to cure neurodegenerative disease and all these things that plague us, which I think we will. What would it look like if we started to cure diabetes and autoimmune disorders? And the indicators here are flashing green. We discovered our first antibiotic in 60 years recently because of AI. We split HIV out of DNA recently because of AI. And all the bio and life sciences companies are getting really excited and they're just scratching the surface. And then this is true in material sciences and particle sciences and molecular science. The Japanese just discovered a plastic, a biodegradable plastic that's stronger and less expensive than traditional plastic. And they're already starting to sort of roll it out. So we're seeing this. We're seeing this. And then the third thing that I think happens as a result is access. I think that we get a lot more access to goods and services. Some of this will happen through just lowering the cost of those goods and services, and some will happen through reinventing these goods and services right today, you and I think of primary medical care as being a doctor's office. But what if it was a mobile device and a wearable in the not too distant future where we did so much more further upstream to actually give the specialists very, very specific jobs versus just these ridiculous problems that we are now sort of thrusting downstream of what could be preventative care and access to education. Right. YouTube is already doing this in many ways. But we're going to see a really incredible Cambrian event around early childhood and adult education. AI powered private tutoring that enables access to exceptional scholastic learning and probably scales our ability to do social learning much more efficiently in, in schools where kids can actually go outside. Right. And we were, you know, Alpha School McKenzie Prices project is sort of promoting this. And the last thing that I think happens is we get a lot more time and I think we get time on two axes. I think we will live a lot longer than most people think. You know, will we solve longevity? Maybe, maybe you and I get to decide when we die. But more importantly, I think that we get to, we will work less. I think that we will actually capture more time in the day. And this is probably my hottest, second hottest take behind deflation. People insist that we must work to be happy and they also sort of really question the idea that we could live in a 30 day work week or a 30 hour work week.

People need purpose to be happy is what I think most.

I think most people would argue that people need purpose to be happy. And I think what we've discovered in all these, and we point to this, the Harvard Happiness study and the Blue Zone studies and the deathbed studies, I mean they're very famous happiness and joyful studies. Now all of them point to purpose as being able to be derived from something besides work. None of them, none of them require work as a source of purpose. And in fact family and friends for many people is a much greater source of purpose. Stewardship of the stewardship of the land. Well, community actually turns out a slightly different than purpose, but they are related.

You're getting into the keys of happiness and this is not new. It's like the foundation is you have to have your health, you have to have a roof over your head, you have to have food in your belly. Those are the foundational issues for having Maslow's Laws.

Yep.

But then you, then it be, then happiness becomes a choice. Because when I say community, it's with. You have to be a part of something, whether that's family or people. That's what we're talking about here. But when we get back to AI you've just described and what I hear. I'm optimistic about all of this, very optimistic. And yet human nature seems to go forward the dystopian for the, you know, it's going to kill us. I mean, Eric Schmidt, you know the articles about AI, the large language model, blackmailed the person by reading the emails. This is an article. You can Google this in order to avoid being shut off. I mean, how do you counter that? Or is that legitimate? Why do people freak out?

Yes. So there's a few things to say here. The first is the anthropic study that you just cited. There are many of these studies. There is a lot to say here. This merits a conversation unto itself, but basically it actually serves the safety teams that do these studies to release these results. And the studies themselves are often pretty complicated because. Let me give you the most pedantic analogy I can, and then I'll come back.

How about, are the studies accurate?

It's not that it's not accurate. Let me explain in the following way. I recently, for effect, actually challenged my nephew of a two and a half year old nephew to a game of chess. And he's a bright two and a half year old. And I said, I'll give you a popsicle, Quinn, if you can beat me. So he sits down on the other side of the board and I'm about a 1600 Elo, so he's not going to beat me. He doesn't even understand the mechanics, but he knows that it's a game and he knows there's a prize he wants to win. And so Quinn wipes all the pieces off the board and sort of declares victory. And I did this for a fact, not even realizing how amazing the results would be. And of course, I don't go to my sister Bryn and say, hey, Quinn's a psychopath, right? You have a future monster on your hands. And if we don't take care of him now, what will he do? Quinn knows there's a game to be won, knows that the stakes of the game are such that he wants to win the Popsicle and can safely assume that wiping the pieces off the board will not result in some bad consequence for him. But he hasn't been taught the spirit of the game. He hasn't actually been taught that this is not how you want to win. And the reality is, there is a ton that the machines still have to learn before we can safely equip them with tasks, and many tasks, we may never decide to equip them With. And this is part of the work that safety teams do. But misalignment does not mean sort of fundamental brokenness. It means an immaturity. And so it is important to acknowledge that, like, a system is capable of doing a thing in order to win in the confines of the game and the rules it was provided. That doesn't mean that it is, you know, not fit for use for many other things, and that it can't ultimately learn the spirit of that game. So that I'll simply say there. And it's a lens through which you can view all these other safety discussions. Now. Why do we spend so much time as a species talking about how scary the future is going to be? I mean, there are so many biases that haunt us such that even the most intellectual among us cannot overcome them. Principle among them, it turns out, is the negativity bias, the idea that things will get worse. And this has been written about plenty. We know why we're conditioned to believe in the negativity bias. And our ancestors, of course, it served them, but there are many others. There's a confirmation bias, there's an attention bias, and then there's, of course, this idea of rosy retrospect. We spend all this time being like, oh, if if only things were more like they used to be. And it's just these are just sort of genetic dispositions haunting us, that biological inheritance, basically, that don't allow us to actually see the things the way they should be seen, sort of approach these things more rationally. And it's not to say that things are going to go perfectly. It's just to say that there's a ton of evidence that they're going to go a lot better than they have.

Do you hit that in the book because your book is about human exceptional. Human exceptionalism?

I try, yeah. Some of it's about human exceptionalism, although I acknowledge that humans actually are sort of quite flawed. And I try to be honest about that. What we definitely point out in the book is that I try to give people the opportunity to see things the way they could see them if they're willing to separate some of their biases. And I try to do it in a way that doesn't, you know, I'm very willing to admit all the things, all the holes you could poke in the book, but also try to help people see the things that might be anchoring them to the idea of some terrible outcome.

As I start to wrap down on all of this, I want to get back to something you said that made my Ears perk up. You talk about AI and its ability to impact health outcomes, how it's already led to discoveries. You mentioned in antibiotics that we had not had, which we now have because of AI. Do you really think AI could unlock the fountain of youth or even immortality?

Well, I don't know that it seems like we would have to learn something new about the sort of the, the known universe or we'd have to unlock something in the unknown universe to keep a organic organism alive. You know, an organism alive forever. That's not something that we've proven possible and seems sort of impossible. But I do think it's very reasonable to believe that there is a bunch about the known universe and unknown universe that we are going to discover that will make this conversation feel silly even in 10 or 20 years. And I encourage people to have some humility when they consider what is possible. And the easiest way to do that is to just try to pretend to have a conversation with your great grandparents about the current state of the world world and like try to describe your day to day life to someone who, who died in 1900 and you know, what would have to, what would have to have happened in their lives in order to actually believe in, in sort of this eventuality. And I do think that it is increasingly likely that we are going to make a series of scientific discoveries that you and I can't really fathom today. And this is true, by the way, for two reasons. One, the more we know about the universe, the more we can know about the universe. Science begets science, but also human ingenuity begets human ingenuity. The Roger Banister effect is one of my favorite examples. Roger Bannister, first man to run a four minute mile on record, does it within five years. So prior to this obviously we thought it was some frontier of human ability and then within five years almost 70 other men had done it. Right. But we are inspired by watching people achieve things that we thought were impossible. And of course the Wright brothers is another one. 1903, the Wright brothers fly a plane. Everyone's or 1906, 1903, the Wright brothers. 1903, the Wright brothers fly a plane. And, and someone at stare, you know, someone at Kitty Hawk is filming it. It and someone watches that film and goes, I wonder if we could fly that plane a little bit taller, a little bit further tomorrow. And of course the Wright brothers great contribution to the world is not in aerospace engineering. Their planes are not particularly engineering interesting. What's brilliant about the Wright brothers is their bravery and their ability to do something. And then, of course, within 70 years, we have put a. We've put a man on the moon, we've achieved space travel. And I just remind people of this because I think it is remarkable how much we are capable of once we see a barrier break.

So this book comes out in January. What's the follow up? What, what, what are we going to be talking about when I interview you in three years?

Well, I hope that we're still. I mean, I, you know, I hope that the book is still interesting in three years. I'm not writing it as a. I'm trying not to write it as a point in time. It's based on the work of John Maynard Keynes, who sort of has this beautiful line that is so amazing. He says, I want to disembarrass myself of imagining a future. The idea that talking about a point 100 years in the future would be very embarrassing as a macroeconomist at the time was true. And he said, I want to just free myself of the embarrassment to tell you what I think is coming. And it's my chance to say to people, what if we could inspire conversations for the next, next couple decades on what might happen? And my hope is that in three years we have really, really interesting proof points to suggest that in fact we are beginning or on this really remarkable logarithmic curve.

Is there anything you'd like to add regarding whether it's AI, the human condition today, and the promises that AI might deliver a better human condition? Anything you want to close with?

The only thing that you and I haven't discussed that I. That I do think is important is I. I wrote the book originally to sort of talk about my view of how things might go well. And then my wife got pregnant in the process, and we're now expecting our first child, a daughter, in October. And one of the things that I realized in the process is I deeply care principally about the human experience. And I do think that AI is a chance for us to actually put humanity back at the center of things. And so much of the opportunity, I think, for the next chapter is to rediscover a lot of the humanity we lost in the digital age. And that a lot of the technology we have today, which has extended life expectancy and done all these incredible things, has also really compromised our ability to just spend meaningful time together. And I, my. My hope and prayer. And so the crusade I'm on is that we build technology that. That enriches the human experience.

That might have been a very good place to end. But now I have to follow up. Will you allow that child who's going to join all of us in October to hold a screen at say 2? I was at the airport yesterday or two days ago and you know, it's what every parent does. It's the modern era. They let the kids watch the cartoon on the phone. But then you hear the stories about the people in Silicon Valley don't let their kids use phones.

Well chewed. No, certainly no. First of all, I have a strong suspicion that we are peak device that I think we are. I think we have more screens per capita today than we probably may ever. And I think, think the screen is going to become less critical in a world where we can do more sort of computational work automatically. I do think that. So first, my wife is a Waldorf teacher, so my prior here is that she spends her day teaching and nurturing the souls and spirit of the Dex generation without devices. And what Waldorf schools have a strict no device policy. I would be very surprised if we had devices in our children's hands at any time before, you know, they were 10 years old. I would also say that we have the benefit of living in Santa Barbara amongst a lot of friends and family. And so there is, we have a, we have a really incredible physical community that sort of calls for a lot, that offers a lot of support and calls for a lot of attention. So my hope is that we just don't even have to rely on devices in a way that a lot of people, parents frankly, are, I think, are forced to right now.