AI Is Smarter Than Ever. So Where’s the Revolution?

So let's say you're in high school. Your parents come home one day and they say, "We've decided to get you a private tutor." And boy, do we have the perfect guy for the job. He knows everything. He has literally memorized every fact and finding in every discipline. He can explain every concept to you with whatever metaphor or analogy you want. You're confused about bio. You need some Harry Potter metaphor to remember the difference between the sympathetic and parasympathetic nervous system. He is so good at that. And best of all, he's cheap. He's going to live in the basement. He never has to eat or sleep. He's available 24/7. How good does that sound? Sounds pretty good to you. And so one day this tutor comes by the house. And you ask him, "Is it true? Can you really memorize every fact? Teach me every single subject with Harry Potter analogies?" Of course, the tutor says, "And there's something else I can do for you as well." In fact, I've done it for most of my past clients. If you want me to just take the biology test for you, I can do it. If you want to just skip the geography lesson and have me just write the geography essay for you, I can do that too. Boring book. You don't want to read grapes of wrath. I get it. It's long. I'll read the book for you and write the book report. He's the best teacher ever. And he's the worst tutor in the world. A tutor so terrible, in fact, that the more you rely on him, the dumber you get. This isn't a magical tutor. It's not a make-believe thing. In fact, millions of students now have access to it. And we're learning the results. A team of researchers from California recently studied the effect of artificial intelligence, chatbots, on math scores on a popular learning website since 2015. These students could have used AI to unpack math concepts, come up with mnemonics for the quadratic equation. They could have used AI to become better at math. But what actually happened on problems that students could copy into their chatbots performance improved since 2022. But when students were given tests away from their computers, their scores plummeted. In fact, their scores after the invention of AI were worse than they'd been before the invention of AI. The tutor who could teach them anything with any metaphor, taught them to be dumber. There's several lessons here, I think. One lesson is the difference between the capability of a technology and its use. Just because we can use Wikipedia or Claude to become experts on quantum mechanics, doesn't mean the primary use case of Wikipedia or Claude is to increase the number of particle physicists. The second lesson comes from the media theorist, Marshall McCluen, who smartly said that every technology is both an extension and an amputation. A machine that extends human intelligence is also a machine that cuts off the need for humans to be intelligent. Like many technologies, AI forces us to ask a hard question. When you're in possession of a tool that can reduce struggle and friction, you have to ask yourself, what is the value of friction? When is the struggle of the point? And when is the solution the point? Ethan Mollick, I think, is probably the single best explainer of what AI can do for people and the difference between what it can do in theory and how it's actually used in practice. He's the author of a new book, Co-existence. And today we talk about the paradoxes of AI. AI makes experts more powerful while automating the junior work that creates experts. It makes our work smarter and faster by automating the process that makes us smart and fast. And according to Ethan, every challenge currently presented by AI will get harder and harder as its capabilities improve exponentially. I'm Derek Thompson. This is Plain English. This episode of Plain English is presented by Comcast Business. Did you know Comcast Business detected 79.3 billion cyber threats last year? That's why they're introducing security edge preferred. It's big business, cyber security designed for small business. Get started with Comcast Business Internet for $50 a month for one year on a qualifying Internet package and find out how to get one month of security edge preferred on them. Plus, ask about a $500 prepaid card. Limited time offer, restriction supply requires 300 megabits per second Internet security edge security edge preferred and wireless connect. Call today. Ethan Mollick, welcome to the show. Thanks for having me. So your book opens with four casters trying to predict the future of artificial intelligence and math. And they gave AI a 2.3 percent chance of winning a math Olympiad goal by 2025. Last year, super four casters gave a Millennium Prize problem, a 1.7 percent chance of being solved. Now open AI has said it has already solved one. I mean, we're in October. Maybe it'll solve another one in the next 24 hours. Starting here at this forecasting failure to anticipate how powerful AI would be even in just a few months, what does that failure teach us about the capabilities of AI? I think it teaches us a sort of lesson that I think feels like people have heard this since COVID, right, which is humans don't think it exponentials very well, no matter how hard we try. And I think that it's tough. Even the world's best forecasters and they are really good, right? Like the, it's a straight line on a log of the graph. Like the ability level has not actually varied very much over since in the last four years, right? It's actually ticked up a little bit. And I think it's really hard for us to grasp what that means, right? We it's better on a bunch of dimensions we never had to measure before. It's not horsepower, it's not speed, it's intelligence for whatever that is. And I think it's really hard to adjust what that what that means. I think that's a lot of failure. I see constantly as an assumption that the future looks like the past and that's not looking true anymore. It seems to be that forecasters have actually been wrong in two directions. I think one group has underrated AI's growing capabilities in, for example, solving complicated math problems. But another group, I think, has overrated the real world consequences of AI's capabilities. If you look at the unemployment rate, it looks basically normal. If you look at productivity growth, it's up a little bit, but there's a variety of reasons for that. And it's hard to see AI in the productivity numbers. If you look at corporate returns, they look pretty great. But if you kind of just like put your hand over the chip companies and the memory companies, a lot of the economy is saying we're not necessarily witnessing some enormous ROI from all this AI that's flowing through our bureaucratic structures. And so it raises this question that you and I have gone back and forth a bit online. So I'd love to get your thesis on the record here. If AI's capabilities are growing so fast, growing exponentially, why are the real world consequences of AI growth moving along so slowly? So this is kind of where I'll argue, I'm surprised how fast it is. I am a professor at a business school, an economic sociologist. And the lesson you learn, and I say entrepreneurship, the lesson you learn is like, entrepreneurship never win, right? Like then, you know, just to say you have a fact, they should never win because all of the resources, all the money, all the customers are all with large companies. Why do they win? Because big companies can't change. So the fact that we're actually seeing in surveys, high levels of ROI already, even if that hasn't been translated into, you know, 30 million is not enough to move the needle for a lot of these large companies. But they are seeing positive gains over a very short period of time that some of them are starting to reconfigure how they work. Like, to me, that's fast, not slow. So I'm not surprised that we don't see this happening, especially because the current phase of AI we're in is single player. It is about individuals getting performance gains. And you could see that, right? I can vibe code an entire game overnight. I can write academic papers faster than ever before. But when I start writing to systems, multiplayer, many people working together, there are still huge gaps. How does that even work? How do we work together in this? So you'll see, like in coding, a code will be a hundred times more productive, but they're still in a two week long sprint that requires input from people every two weeks in different kinds of ways. You're not going to have a change. In fact, what you're going to have is huge amounts of token spend. And you're not going to see an impact. I don't think that that is a long term persistent problem. But I think that it is a very obvious short term problem. And something to talk about a lot in the book as well. There's two concepts that you've used that I think help to make your argument especially sticky. You've argued or quoted other people arguing that jobs are like O rings and that companies are like garbage cans. Why is the economy in your view made up of garbage cans and O rings? Right. So we have two tragedies here in different ways, right? So the O ring is named after the spatial challenger where 16,000 parts or whatever worked great that day in 1986, but the O ring failed and caused a cash hash for free. And part of that is, you know, I talked with a jagged frontier a lot in this book. And last book was a term we coined in our papers. And it's the idea that AI is good at some stuff. You wouldn't expect bad at some stuff. You wouldn't expect it. It only takes one or two places for AI to be bad at a job for you to still need a human to sub in for that because you can't have everything work great and then one thing fail at most jobs is not additive. We're little bits out and subtract from each other. So the O-ring means that jobs are not easily kind of replaced one for another with AI, because if AI can't do human tasks exactly, you still have value accruing to human work. And you see that happening, and even in coding, where like AI now writes code, but you need to still need human engineers to manage it. Now maybe that will change in the next couple of months, and then we'll see another part of the O-ring gets stuck. So jobs are evolving very rapidly, but there's still these sticking points, especially as you go to organizations. So then the garbage can model is one of the more depressing models of organizational theory, which is that basically companies are just bundles of problems and solutions all looking for each other, paralleled on top of each other. So there is no plan, or at least the plan is only a small part of how organizations work. And organizations are instead full of like contingencies and complicated aspects and evolutionary data ends. And there's loops of work that no one ever knows even exists anymore, and organizations just do things. And as a result, that makes it hard both for to improve them immediately. And also means that the job of redesigning organizations to work with AI is more complicated than just, well, we will supplement this task, because that task is on paper somewhere, but doesn't represent the actual garbage can of the company itself. So you can see why we call this a garbage can metaphor, right? Everything is tossed together and like wasted a garbage can and mixed together randomly, and sometimes it works out, and sometimes it doesn't. Like you, I tried to write about and cover artificial intelligence for a wide range of readers and listeners, some of whom are working at or even executives at the frontier labs, and some of whom absolutely despise the frontier labs, hate the acronym AI, hate everything about this subject, and wish we could just get into a time machine and go back to October 2022. And I think there are a lot of reasons for this. I mean, you see it even in the survey data, where if you ask people who use AI, do you think it's helpful? By a four-to-one margin, they say yes. And you ask in the same survey, do you think AI is bad for society? And by a four-to-one margin, they say yes. So there's all sorts of interesting discrepancies here. But one that I would love you to dilate on a bit is that we're all using the word AI to describe what are often completely different products. When the frontier labs say our AI solved Navier Stokes or some other math problem, they're talking about an internal model that won't be made public maybe four months. Then you have the most advanced public models that people spend a lot of money for, the corporations or individuals subscribe to. But only about three percent of individuals in the country actually personally subscribe to an AI. Many of them, if they're using it on a monthly basis, are using some model that's been public for years, which means that it was internally available even more years. So how do you think about this? I think very difficult public communication challenge of talking about a product that is at least in my last recounting three different things. The most powerful internal AI models, which are much better than the most powerful public AI models, which are much better than the models that most people are using on a week by week basis. How do you address this problem? You're probably short of few by the way. There's also the AI model that has people's heads from science fiction, which determines a lot what they think the future is, is Terminator, or whatever they watch, and there's the AI, and then there's still inside companies, or Big Issue, is AI still means machine learning on your own data, where data was the new oil, and all the other techniques around artificial intelligence. So there's a lot of mess here. I would say that part of this is that it is confusing, right? Like we have a general purpose technology. So there's many things happening at once, and the uses are different everywhere as well. I would say the biggest lack of clarity, though, is not feeling that exponential curve, right? And it seems very corny, very Silicon Valley, to be like feel the exponential, AGI pill. But like the truth is that models are getting better very, very quickly, and your old views are kind of out of date. I was actually, I think I'll remember the number correctly, but I was just trying to figure out how good the best model my last book was when it came out, which is two years ago, by the way. And both of the way I started running it, and when I finished writing it, and for several months afterwards, the best model in the world was GPT-4. It was the best model for about a year. And GPT-4 is now the 181st best model in the world right now. And there are 19 models that are running your phone that are running the GPT-4. And that was like whatever was freaky out about a few months ago. Now, the average best model changes 70 days or less, right? So we're in this very fast acceleration period, and I do think everyone's lagging. I think that you can even use the smartest model in the world right now, available to you, and you're not trying to solve Navier Stokes. So you don't even know what it can do, capability. So there's this huge mismatch, the overhang, right, between AI capability and where people's perception of AI is, and people are making all sorts of predictions based on all kinds of missing assumptions, kind of nobody knows anything. Reading your book, I felt like in many ways it's a book about tensions, about paradoxes that we live with when we work with AI, when we read about AI. And I want to go through at least four paradoxes that I drew from your new book. The first paradox I'm thinking of as the expertise paradox. And that is the idea that AI makes experts more powerful, while often automating the kind of junior work that actually creates expertise. Maybe unpack a little bit about what you see here and how you think we're going to grow the next generation of experts if we're automating the process by which they are made. Yeah. And the great thing is since I wrote that is like we actually more evidence that both things are happening, right? I mean, maybe not great, but a true thing. So on one hand, there's a nice paper out of anthropics suggesting that the more expertise you have, that better returns you get from using AI models. So you get better answers and you get more answers, right? And that makes sense because as AI has gotten better, it produces a huge amount of material. And if you're an expert, you can glance at that material right away and know whether it's good or bad or how to change it. Just like you could, if someone handed you a report, you're not going to read the whole thing, you flip through it or you know, if you're an expert on in a podcast, you know how this interview is going right now and are making judgment calls in your head that you couldn't describe to anyone else, right? So expertise has real value in working with complex systems. At the same time, you know, the way we teach expertise is apprenticeship. Like we don't have a way of teaching expertise fully in schools. So what we do is you send someone to be an intern at a podcast studio and they do the work of doing the pre-interviews of me or whatever else and you give them feedback, hopefully very kind feedback, maybe some people yell at them and they learn the ropes that way. And you didn't want to do that interview so they get to do it and they get feedback as a result and they have to prove that they're good. It's been this really nice cycle that works well under capitalism and under state plans, societies, everything else, which is junior people want to strive to get ahead. They, they, we need someone to do the work for senior people who don't want to do it and you learn as you go. And all that broke down with AI because junior people are much worse than AI, which means that junior people will be better off, you know, doing, we're having AI do the work and they don't have the expertise to assess whether it's good or bad. So they'll just turn in AI assignments. And the middle managers who used to be assigning work to junior people and giving them feedback, AI does a better job than that, it has no human emotions. So I'd rather just assign it to AI. So nobody learns anything anymore. The learning was incidental to the process, even if it was the point in the long term, right? So we were very lucky that we had a system where expertise creation was incidental to a process that worked really well in any kind of, you know, for 4,000 years essentially. And so we have to rethink how we rebuild expertise and that means we have to decouple efficiency from learning, which is not an easy thing to do in companies. Not an easy thing to do at universities either. We had David Deming on this show a few months ago and we talked about what I think is a really naughty problem of how do you teach artificial intelligence in a educational community that should be all about making thinking difficult so that people struggle at college so they grow the muscles they can use when they graduate. But on the one hand, it seems if you teach with too much AI at the Harvards and Whartons of the World and every other, you know, the Virginia Techs and University of Michigan's, then you are automating the process by which people become more intelligent and that means they're going to end up dumber than they otherwise were. On the other hand, if you say, well, then we obviously can't teach with AI in college because it's allowing students to end run the whole learning process, then you graduate thousands of people into a labor market where they might be expected to use AI and you have forbidden their contact with this general purpose technology. How do you think about the right way of balancing the virtue of learning at college and the virtue of colleges to serve as a pre-professional institution so that people graduate smart but also with some kind of facility with a technology they'll likely use in their first job. Okay, so this will be my most controversial point and so the academics don't say but I don't think this is that harder problem. I think transitioning to the state we need to be in is a harder problem but we actually have a lot of control over people in universities and K-12 education and we had to do things a very particular way. No one asked for AI to come and rex the way we were doing things but it has and we actually have the evidence already right so if you use AI as a learning shortcut which most people will naturally do either because you want to cheat or just because you're asking for advice and the AI gives you the answer and you think you've learned something you haven't we're very bad at knowing when we're learning then you destroy your learning opportunities right? If on the other hand AI uses a tutor where it's not going to be the answers but it's pushing you we now have enough studies that show positive effects and meta-analysis RCTs are Harvard at Stanford and Kenya other places where we actually know AI and I can be a very powerful learning tool. So the answer isn't that hard, right? We need to use AI as a learning tool outside of class and inside of class do what we've known for the last 30 years of pedagogy works best, which is active learning experiences. We're in class, it's discussions, cases. We have to test people in class without laptops nearby. We have to have in class discussions. Like it's not the end of the way. It's not that hard a process actually, getting there will be hard, but like there is a better form of learning that awaits us in the end if we get there, all based on how we know how to do pedagogy today, right? And I think then the other piece of this puzzle is, what does it mean to teach AI? I'm much more skeptical about there's an AI to teach or that people have to learn to use AI in college. Like I don't know what that skill is. I don't know what learn to use AI as a skill is. I know when CHROs of companies are saying, oh, the kids today, they're AI native, they're not AI native, they're just cloth. Like they're just pasting stuff into cloth. They're not learning to use AI. If returns to AI use or returns to expertise, then you're not an expert in AI, an expert in a topic area where you can use AI to do it. - Just the last thing you just said, suggests that professors should be much more aggressive about limiting, even prohibiting any kind of AI use because there is no benefit to becoming faster with the technology between 18 and 22. At college, your job is just to become good at learning. And when you graduate, you'll just learn and to be faster with the AI that exists then when Bane or whoever else hires you. - So I think it's a little more complicated than that. I think that every class doesn't need to be an AI class, right? I absolutely think we should, like this is an emergency situation. We should be teaching AI in school and in schools, but I feel like there's, you know, I'm a professor. The way professor works is everyone kind of does their own thing in most cases. And so you have a lot of people, everybody's integrating AI into their class in all kinds of ways, some of which are hurt learning, some of which are wrong about AI, some of which are, you know, have trouble keeping up. So we should be thinking about how we teach AI skills. Like it helps to know a little bit how LM's work, not a lot, but a little bit. It helps to know about what a current model stand. But I don't think every class needs to be an AI class. And I don't think we know what that even looks like. And by the way, there's lots of room for AI pedagogical tools. In my classes, I have AI, people have to teach it AI that it trained to be a kind of a dumb student. So you have to teach it. That's a great way of learning is teaching. You build a case study with it. You run simulations where the AI is a game master. There's lots of interesting educational opportunities out here to use AI for helping learning that we will have to keep uncovering and working on. But AI as a subject is, it might abuse everything you do, but you wouldn't want to take every class during the industrial revolution should be a steam engine class. Like it's not necessary, right? And so I think in the same kind of way, like there is value in this, but there is a little bit of like everyone jumping on board of like we need AI literacy, AI certification. Well, I can tell you, we've been studying at Wharton, all of the sort of prompting techniques. I used to teach people, none of them work anymore. Like if you learned how to do, if you learned AI last year, right? And you're a recent college grad and you had a formal class on this, none of that works anymore. Think step by step is not a useful thing to tell an AI model. Persona has no longer changed AI performance levels. Like almost all the stuff you learned is now useless. So I think we need higher order thinking, I mean, AI looks a lot like project management right now. So maybe everyone has to learn a project management plan. I think very related to what we're just discussing, which is the expertise paradox is what I'm calling the skill paradox. AI can make people much more productive. But as you've said in several studies, AI seems to make people better at certain tasks. But when you take away the AI, there's scores plummet because it's revealed that they've been using the machine as a crutch rather than a tutor. When you're working with LLMs, how do you know, how do you feel when you're on one side of the line using AI as a tool versus another side of the line using AI to automate your thinking? It is a really hard process, right? So I do it by establishing bright lines for myself, right? Like I write every draft myself. And it's tempting sometimes because I'm like, I don't really want to do, like, you know, it's another poet, another article, like I, but I have to, like I write the drafts myself and I definitely turn to AI afterwards and always the most depressing thing is like, there'll be two lines of AI writing and like a piece of work and everyone's like, that was a great line. I'm like, you know, everything else. But I think that there is this kind of sense of like, you know, like this is the hardest challenge I have because we will, we will absolutely de-skill ourselves every chance we get. So part of the message of the book, I think, is be thoughtful about AI use. And it's tough to say that, right? Like, you know, humans don't like effort, right? We've had to teach ourselves that lifting weights is good. When it obviously is something our body is designed to avoid is like, we don't want to do this, right? We don't want diet. And in the same way, learning requires desirable difficulties, requires difficulty. We don't want to do that. And I think choosing what skills we will are willing to give up, right, you know, and what skills we aren't is going to be our big challenge here. And I think building incentive systems to keep the right skills is going to be hard, especially companies which aren't built for that kind of thing, right? Like, it's one thing. And again, university said is we can control this. I can give you a test and make that test your grade. And you don't have AI in that test. And I can reveal whether or not you've learned things. And I can help set you up in a world where all the AI tutors are available to you. And the class experience, they're amazing. The TAs are available. And it ultimately has to come down to you in a company, though, that's where things get really weird, right? Because not only do I not want to do that work, that the AI is doing, you know, and de-skill myself, but also the company doesn't want me to do that work. And it's more efficient for me not to do it. So strategic skill retention is going to be a very big deal in the future. Going back to the metaphor of Jim, it's just really interesting to me as I reflect on how frequently in the last two years, the gym metaphor has felt more appropriate for the thing I'm obsessed with. So initially, you know, the gym was necessary because we had to be hunter-gatherers. We had to walk for 12 hours a day. And then we invented a specialized labor economy that meant that we didn't have to walk that much. And so if we wanted to remain fit, we had to go to this special thing and simulate some ancient processes in order to get back that fitness. That's Jim 1.0. Then when I wrote the anti-social century about the decline of socialization, I was really interested in people who said, you know, we thought of friendship and community and neighborliness as something that naturally comes out of us because of the need to reach out to people for support and meet friends and keep friendships. But now that the smartphone is here and it's easy to keep in touch with thousands or hundreds of people as sort of third party allies on Twitter or Instagram. And you can sit on your couch all day and watch Netflix and go on TikTok. We need to think about hangouts as a means of social fitness, like get out of your house and practice being around people. And now what I'm hearing from you is there's a third gym that we need, that, you know, AI because it's so good at doing our homework and our work and our jobs for us, we need to remember that sometimes the friction is the point. The struggle is the point and we need to think of aspects of intellectual endeavor as a kind of gym, a special place or time to go to to work on skills that might feel hard in the moment and we know we don't have to do it, but we think it might be good for us in the long run. It's just funny as I'm reflecting on the idea that this concept of the gym keeps coming up and all these segments as we find new ways of making life comfortable and technologically rich. I wanna move on to the third paradox here that I wrote down, which is again related, it's what I'm thinking of as the knowledge paradox. In the last few months, open AI has released hundreds of math findings, proofs. And these are proofs that mathematicians themselves sometimes find quite alien when they work through the logic. I think one that you referenced on Twitter said a proof red like it was written on psychedelics. And I'm interested in how you think about this tension between AI creating knowledge without understanding. That is to say, it's creating the proof before humans understand the thing necessary to get to that proof. How do you think about this question of whether it matters that we're creating knowledge without human understanding? - This is part of our challenge, right? Like we didn't have to unpack different kinds of intelligences, really, because there was one kind of intelligence was human. So we talk about arrays and different types of things, but like we all kind of work the same way. And AI's are remarkably unaliened in a lot of ways, right? They're not the alien computer that we were expecting from like Star Trek, it's all logic. It would blow up if you said the word love to it, does not compute it, blows, like they talk like emotional people for a variety of reasons real and unreal. And you know, but the alien starts to come up when we talk about intellectual work, right? And in this kind of, and also we haven't had to examine our own intellectual work. So one of the interesting things about mathematicians, right, is to the outside world proofs for the big deal, right? That was the goal is like these are the millennium problems if we solve them or in great shape. And the mathematicians have been protesting and probably we're saying before this, but our predecessors publicly, not like this, right? The proofs weren't actually the point. Each proof would spit off new ideas for us that we never discovered before. And the process of solving it would bring the math community together around new problems and spaces that were interesting. And that's all being ruined. And in some ways, they're absolutely right, right? Like we're unraveling math as we know it, and we have to reconstruct this field. I think that's going to be happening in lots of spaces too about reconstructing what we actually meant, unbundling this idea of intellectual engagement. I mean, you can now generate a pretty darn good podcast from my book or our tweets or anything we want, but it's not our podcast. How do we defend the idea that the human touch and writing or in proofs that the way we do knowledge and acquirate is valuable? I think it's going to be a big challenge coming up. So I'm absolutely of the belief that there is something real here. And then, but then you have a second order problem, which is, okay, well, do we need the humans to do math at all anymore? And, you know, right now we do, but six months ago, it was very obvious we needed humans. And six months from now, it may not be, right? The proofs get generated. And then does AI, then it's math become a game the way chess has, right? Which is like human achievement matters, but it's sort of in a bounded space. I think Bill Gates was talking about leaving reserves of problems for humans to solve, which feels a little bit unnerving to me as an outcome. But I think that we do have this kind of idea of trying to figure out what is the job that we were trying to do. And when we say not like that, AI, what do we mean by that? And I think that's going to be important. I think there's a lot of not like that, AI, that's going to be important for us to think through. As I was reading your book, I had a conversation with the mathematicians, Stephen Stroghats, about this phenomenon in your circling, which is mathematicians being indignant, offended, and quite moved to tears by how many math problems were solved by open AI, seemingly solved by open AI, in a way that seemed to nuke their fields of research. And on the one hand, you know, you would expect a mathematician to say, or to argue forcefully, that not like this open AI, this is terrible, having AI solve math problems without human understanding is a crime against the field. But he also shared this story of a character from his last book, MIT computer scientist, Regina Barzela, who survived breast cancer in the early 40s, and was really stunned in the process to learn how much doctors didn't seem to know about breast cancer diagnosis. And so she led a team that trained AI on hundreds of thousands of mammograms, and they developed a technology that predicts breast cancer cases many years in advance, better than any team of human doctors. And Stroghats asked, "Barzela, do you understand what your AI is doing?" She said, "No." And he said, "Doesn't it worry you that you don't understand your own invention?" And she said, "Why would we limit a life-saving model to what humans can understand?" And on the one hand, I am very sympathetic to the tears of mathematicians. And on the other hand, I'm very sympathetic to wanting to not die of the cancers that killed my parents. I want to live, and if living requires a life-saving model that humans don't understand the inner mechanics of, I don't give a shit. The years of life are worth more than the qualia of those scientists understanding. And so I would love you to reflect just a little bit more on this tension between the scientific value of human understanding and the real world value of just science and scientific progress for the sake of scientific progress if it is indeed saving and extending lives. I mean, this is where we're going to face everything, right? There's a lot of. This is in science, it's the easiest kind of case, right? There's a lot of. when people are talking about math, they were saying, "Cancer biologists be like, 'No, no, don't solve, not like this, don't cure cancer, we'll be out of a job.'" But also fields are their own games, right? I'm the economic sociologist by training, as I said, papers at a business school, I'm dealing with this kind of issue, too. The AI can now produce a really good somewhat boring but actually correct economics papers sort of at scale, right? And people are starting to recognize this, like, you know, the more advanced models that you have access to, the more they can do this stuff, it's very clear where this is going. It doesn't have research taste yet, although interestingly, I haven't figured a lot about this, right? Like, what is taste? The taste is the idea that I as a human can only work on so many problems and I need people to be care about the problem I'm working on. So I'm almost p-hacking myself by selecting topics that I think will be interesting. The fact that the AI can produce an infinite number of papers, most of which are trivially true, starts to challenge the very idea of, you know, classic mertonian science. What do we do in science? Do we look at everything? Do we look at problems that are interesting to us? Does taste matter? And this goes back to that general question of collapse, right? Like, we have a bunch of institutions built around human intelligence and that includes organizations, by the way. Orchard invented 1855 for the railroads to be able to solve a problem the railroads had. And we can use them today because they were built around human intelligence being the limiting factor. And we're about to have this from everywhere, right? Which is, do the outputs matter? Is the human process matter? How much is the human process matter? Are we overestimating it because we don't like the idea of AI taking over? And you know, that's sort of the long pathway to the AGI-ish, you know, UBI world where we, you know, we're all general women and general men scholars, you know, and living like Jane Austen, like societies where we take a class for fun and Drew Croquet and gossip about each other. I mean, I don't know if we end up there, but like, the process of getting anywhere like that is a rough process. But I'm both sympathetic to it. And we have to figure out what parts of the human endeavor are valuable to keep, right? What parts of academia and scholarly research are valuable outside of discovery? Which ones are only valuable in the context that they get us more kinds of things? What was the point of math anyway, or economic sociology? I mean, we're staring a lot of abuses in the face all at once. To review, just even for myself, the paradox is that we've gone over. There's the expertise paradox, which is that AGI makes experts more powerful while automating the junior work that creates expertise. There's the skill paradox, which is that you don't know when the technology you're using is a tutor or a crutch that is doing your work for you, and therefore making you ironically weaker. There's this knowledge paradox where we used to think that understanding was necessary in order to get knowledge, but now we've invented a machine that is giving us knowledge before we understand the process by which it arrived at that knowledge. And then finally, there's, I didn't know how to otherwise call this, there's a human paradox. And this really gets to, I think, a really fundamental question here, which is something like when is human effort essential? You have a very interesting way of getting into this question, which is by looking at obituaries. I believe you reported from a funeral director's conference where someone called AGI, quote, "The greatest advancement in funeral home technology since some kind of enbalming oil," end quote. And the question here is obituaries are being more and more written by AGI on the one hand that seems like kind of a mockery of life. This person was alive in your life for years, hopefully decades. You can't take three minutes to write an obituary with your own human fingers. And on the other hand, I mean, if the words work for the mother who lost her child, who the fuck am I to tell that mother that the words don't work for her? It's her child. It's her obituary. When is human work essential? Oh, I mean, okay, we're at that exact same issue, right? Like, I have surprised, I mean, I was surprised with the GBD4 error, even before we had these agents that could do all this work for us, that the first thing people gave up were wedding vows, stories for their children, obituaries. Right? Because that stuff sucks, right? In some ways, this is the back to our effort problem. And the question is the effort worth it. Is the effort of sitting down and agonizing over this, you know, is that what you want to do? Do you want to summarize a life? You know, wedding vows. People have copied wedding vows forever from books of wedding vows or adapted them. Like, I think we're getting a fronting, where is human effort matter? And then, much more banal side, letters recommendation are valuable because they're setting time on fire as a recommendation for somebody. And the quality of the letter often matters less than the fact that the person agreed to write it. And I asked my students at one point, like, would you rather, I spend my normal 45 minutes writing a letter that's only an okay letter, because I can only write so much in 45 minutes, but I will read all your stuff. I'll think about it. Or would you rather me do in two minutes with AI if no one could detect it's AI, but it's better written and a better for your job that you're applying for? Everyone said two minutes. One of my students, they gave me a bridge to send this. When they asked for a letter recommendation, sent me the prompt they wanted me to use and the material they wanted me to prompt it on. I mean, why don't you just give you the whole letter? That way you- People actually do that, actually, too, right? Like, that's another thing you see people do is give a letter. So, like, the question is, what was the value of that human work, right? And, by the way, at least on abitwaysways, we're all seeing this breakdown because, you know, court filings used to be hard to do. So, that was a barrier to human work, right, that mattered because it meant that unless we were a human or paid a lawyer to do it, you couldn't do a court filing, right? Job applications, right, are, you know, are- were an issue that was like, there was an effortful piece or college applications or essays, there's a whole bunch of stuff built for human intelligence, right? So, on one hand, we have the abituary, which is the personal feeling about what should be human in one stop. And the other, there's systems that nobody wanted to fill out their, you know, annual report or whatever it was and the AI will do it for you, but it was really too hard for a human had to do it. And that whole stream, every part of humanity is under attack, right? What should humans do become as a relevant issue? And right now, we have a little bit of a saving grace, which is you can tell Claude writing at a glance. So, those of us who don't write with AI, get to stand out for a little while longer, right? Like, it's, you know, I have a writing style that's my own and I use it and people feel good about it. And, you know, so there's still little poons of resistance about this humanity effort, but like, who are we to make any of these judgments? As we build systems around expectations of human intelligence and human effort that are falling apart. And the obituary itself is a little bit of a cruel thing for people who don't want to do it, right? I have to write and notice about the person who died, that's a rough thing to do. And so, is it okay that people are skipping that? That story for your kids, like, I love telling stories from my kids, but I'm a writer who who likes to play a dozen of dragons, does weird stuff. It comes up with games. There's a lot of people who find that really challenging. Who am I to tell them that if they're using AI to write a story about their kids' favorite thing that has a lesson in it about geometry that that's a bad use of their time? And so it's a really hard situation where what's human is under attack and everybody will be mad at you no matter what choice you make. Last question on your book, you say that there are four human advantages. Deep knowledge, wide knowledge, taste, and agency. When you briefly define each of these human advantages and tell me which one of the four you think will become more important even as AI gets better and better. Well, the point was that all four of them I thought would give them more advantage because they're not bright lines, right? The seductive trap I get, one people to avoid is that AI can't do something. I mean, it might be true, but it hasn't been true yet. And I always see this retreat. Like the most recent two things you hear a lot about judgment and creativity. And if the AI does, it can do 50 hours of human work or 50 days of human work or 50 weeks of human work, which epoch studies suggested it can do, like then it's exercising taste and agency and judgment all the time, right? You can't do a long task without using creativity and without using judgment. So I don't try to sign bright lines. These are things that are advantages for humans when they're working with AI systems rather than human advantages that are not piercable. And I think they will grow. So deep knowledge is the idea that you have expertise. We've talked about this a bunch. Expertise has value in working with these systems because it lets you keep up with them to some degree. It lets you see across the jaggedist of the frontier in ways people came. It even lets you understand how to use these systems in ways we objectively know make AI work better. But why knowledge we haven't talked about as much, which is the idea that like all the roles are blurring together, right? When I talk to people in companies right now, the phenomenon that no one's talking about as much, boy, that sounds like an AI phrase. I promise I'll make human not read the script. No one's talking about is like 90% of the bot comments to me on Twitter is in LinkedIn is like, but people aren't talking about as much as the fact that jobs are dissolving, right? So coders are doing design tasks, designers are encoding tasks, you know, everyone's doing everything. And so there's advantages in knowing more things because you can guide AI to more sets of stuff and have a basic idea of where there works. And as a humanities person, also, there's value because you can get more interesting stuff out of the AI if you know 100 different writing styles. If you know the words of the design language, you know a little bit architecture. All of these things are actually quite useful. So wide knowledge, return of liberal arts. And then there is this idea of taste. AI produces a lot of things, a lot of that, too many things is slop. The differentiator between slop and not slop is a human deciding something is worthy and taste worthy and canonizing that are building on it. So that's valuable also. And finally agency, the ill-defined term that I can kind of mean here, you know, entrepreneurial action, the idea that you are not waiting for clear instructions or rules because they're not forthcoming and you're going to try and experiment and do things where they are in your own. And I think those four things will continue to become more advantageous actually. I think expertise will probably be the biggest under attack in the short term because AI can reach expert levels in many fields. And then, you know, but then applying the expertise to other fields is where wide knowledge comes in. And I think agency and taste will persist longer. Let me offer one more and it might fit between some of the categories that you've drawn. It's presentation ten ethans could have written a book coexistence and then gone on ten podcasts and then gone on ten more podcasts and a couple different TV shows. It matters that one of those ethans is the best at presenting the information in the book. The text is the same whether it was written by a human or a large language model. But the ability to present ideas is a skill that ought to become relatively more valuable in a world where jobs, as you said, are emulsifying in the background where the coders, the managers, the coders and everyone's doing a little bit of marketing. It's hard to imagine a world where the ability to present knowledge becomes less valuable in a world where the underlying knowledge itself is somewhat anonymized because no one knows exactly where it came from. How do you think about this skill of presentation? Because you're not only very good at talking about your ideas. You're also very good at thinking of really funny examples to tweet and post about in order to present your ideas. I feel like that presentational skill is going to be at a premium, at least in the short run. Right. So it says the podcast to the professor. We both are heavily vested in the idea that presentation matters, right? I mean, I will tell you, I feel the bitter lesson, the idea that AI trained enough stuff will eventually beat most fields, nipping at my heels with Opus 5.5's level of explainer videos. I made explainer videos at various points and they were very expensive and hard to make and took a team of very talented people and I could get a pretty damn good explainer video with a musical theater number built in out of Opus 5.5, one shot kind of prompting. And so I think there's value and I know how to steer it, like there is that at the highest level of expertise, human experts are very good and AI is good as it is, does hit boundaries in this sort of stuff, especially with that jaggedness built in, where even if it's an even better explainer than humans, it tends to explain things the same way. So that variation doesn't matter. So there is something about presentation that matters, but I almost would consider expertise and presentation if you're very good at that. But I don't suddenly hope that people were bad explainers, like all the random little YouTube short channels I sometimes watch have some got a much better explaining stuff, which used to be one weird military guy explaining military history or architecture or whatever it was. Now there's a lot of really good videos to explain this stuff and I'm like, oh, the general quality of explanations going upwards, right? Like, I think that's, like we may hate the way that every LinkedIn post sounds the same right now, but they're much better than they used to be at explaining things. So I don't know. I wonder if we are, we're, you know, I feel the same thing you do, but I kind of wonder if we're being, we're, we're making the usual thing, which is, uh-huh, but not us, you know, as well. Fair enough, especially with the 5.5 explainer videos, which are eerily spookly, excellent at times. I coastline everything you said about digital presentation. I was also referring to the ability to present one's ideas at a board meeting, in ordinary meeting, when you walk by someone's office, that kind of face-to-face, presentational excellence I think can be really important for people and their careers to build relationships, to prove that they actually understand the memo that they sent out to the whole office, rather than prompted an LLM to write it without really understanding it. And so I do think there's going to be something about, about presentation that will be at a premium. I want to turn the page here, or I guess, uh, other book metaphor, close the book, because I want to talk a little bit about the news. Um, and the most important thing that's happened in the last few weeks has been the release of and the success of Muse, uh, the meta agent. I want you to tell me what you think Muse is teaching us about this moment in AI use in AI technology, and what makes this a different kind of technology. So for those who don't know, right, Muse is a personal AI agent, open AI has one called dot, lots of other ones are really coming out in the near future, right? They're all based on this idea from early last year called OpenClaw, which was the same kind of proactive AI assistant. And they're based on this underlying set of technology, like they're only capable now because of a couple of things that happen. The model's gotten good enough that you could throw undifferentory piles of information at them and they are really good at figuring out how humans think and what the problems are. So there was up for, you know, the last couple of years before this, there were all of these efforts, rag, other kinds of approaches about, you know, AI wasn't that smart. So how do we get the right information at the right time to be smart? It turns out the current models are good enough to be hooked them up to your Gmail and you hook them up to your, you know, they, they figure out what's going on. And they, you don't need to do a lot of work behind the scenes and the models kind of do stuff. So there's not a lot of clever engineering here as much as it's the models got smart enough to make this happen. And they solved the big problem, two big problems with AI, right? So which was, by the way, why every company pivoted to enterprise was because of these problems. One is when faced with a box that you could ask to do anything, nobody knows what the type of the box. So even if AI could do that stuff, people did not have nowhere to type into the box to make things happen. And the second was, how do we get AI to be proactive and solve problems for us? And Muse did both of these things by basically saying, hey, people have problems, look through your email, and you'll find problems to solve and then offer to solve those problems. And you're not capable enough with the tools that you have and the harness you have to solve them, right? And right now it's like faking phone calls, but eventually we do real ones, but I've had, I have had Muse and I've had Dot, which I've been using also. I've had it cancel a customer service that I was locked into some subscription fee on American Airlines. I couldn't figure out what it was. They, I figured it out, joined a chat with American Airlines, their own bottom shore, canceled the thing, then had to fill out a secondary form and got that back to me, right? And it messaged me earlier, my dot did that I, or pre cursor kind of version I had hacked together, that I had sent the wrong tax number on a form for a permit inside my, in my township, right? And it said, oh, you actually gave the wrong number there. Like, it told me when, you know, it said, oh, there, you know, it checks whether I'm going to be on the right call or not. So there's this proactive element to these systems that, that makes them useful to people who didn't know they had problems to solve, because it turns out there's a lot of friction in everyday life that AI is really good at navigating. Which, by the way, going back to our previous topic, are things that were really important for companies that that friction was there, right? It was really important. It's really important for the entire US financial system that is not easy for you to transfer money from low interest savings accounts to higher interest savings accounts because low interest savings accounts is where banks make all their money, right? And it's too much friction to do. It's important for customer service organizations that is hard to get customer service calls answered because otherwise everyone would ask for refunds. So Muse breaks that friction down and just does stuff for you in a way that doesn't require imagination. And I think it makes the personal AI real. Is there any kind of financial moat that a company has over these kind of agents? Or is this the kind of thing where meta has Muse and OpenAI has. And they're all offering what is essentially the same technology, which means that no company can charge ordinary users any kind of premium for the use of a technology that's basically equivalent wherever you get it from? Well, I mean, there's a few things there. One is, I think this is where the companies that have frontier models and Muse is near frontier, but they have a model underlying it, right? Dot is obviously run by Ash from other models. I think not having a frontier model makes it very hard to compete in this space, right? Siri cannot compete with what Muse or Dot can do, right? And I think that, so I think that first of all, there are a more limited set of competitors. Now, maybe you can hack something together with OpenWaitz models, but part of the interesting thing about being a frontier lab is you also get efficiency gains from the models you're doing fast, fast price decreases. You have compute, you have research power, you can put on this. And all of that means that I think that it's going to be relatively few companies that can offer, keep offering agents that keep getting better. Maybe it's enough to satisfy people and be enough. But, you know, and I also think then you'll be able to trust the relationship. I mean, you can get an email from many places on the Gmail, right? But like, there will be people who figure out models. I have a feeling they're going to be the people already have scale in this space because they can already do this cheaper and they have advertising models in place, right? Muse clearly meta wants your data in the long run and wants you to have a relationship with meta products and not ever get profit from that. OpenAI was already charging you $X dollars a month for access and they throw this in as a feature and it becomes, you don't want to switch your agent. You don't want to switch your personal assistant. So I think there's a relatively few set of competitors where the price gets squeezed in this as possible. But I do think that there is a little bit of an actual vote here. It's not cheap to run these models if otherwise. So, there's two ways to think about the introduction of these new kind of agents. One, you can think about it from the user side, which is how are people going to become comfortable with giving lots of private information over to an agent that is essentially buying in a friendly way on their email and on their calendar and saying, "Hey, can I help you out with this thing? Can I help you out with that thing? Can I remind you of thing number three?" But then there's the other side of this agent use, which is something like what happens when millions of personal agents are calling customer service and fighting the charges on your behalf and claiming the rebates and negotiating with companies? Is this going to create a wild bottleneck on the corporate end in kind of the same way that on the job application front, the ability of students to apply to not just 20 companies, but like 900 companies that they're graduating from Penn means that those companies have like 10,000 applications incoming and so they have to use their own AI systems in order to separate lead from CHAF. Like what's going to happen on the corporate side of the internet economy in order to react to the emergence of all of these agents? I mean, yes, everybody's worried about this right now because first of all, you can't influence the agents from marketing or anything else, right? We tried to do that. Actually, in my upcoming book, the things I did where the book was put in a page together for agents and the agents hated the page, so I had to change it after doing interviewing a bunch of them. But we have research, so you can't really influence them in a consistent way. So that's bad, right? Because they're doing shopping. So will they just go for the lowest cost product? How do we think about that issue scrambling everything? And the other big piece of this, like you said, is systems are built with friction that agents care through. And we count on that friction happening. And the same kind of thing as the customer service thing is the same thing as like job applications, the same thing as by letters of recommendation. And I think companies are not ready for what's about to happen. And so Amazon banned mues. That feels like the kind of mistake that Walmart made, and by the way, they're friendly about these models because they were burned when Amazon appeared that they didn't move quickly enough in the e-commerce. I think that we are going to see, look, everything that we were talking about today is about AI reconfiguring parts of the every part of our, everything we've talked about is, oh my God, everything has to change. And we were talking about it, but piece after piece after piece. And in some ways, that's kind of where I imagine the coexistence piece right now is we are in the renegotiations should phase of a lot of work that is still jointly human AI, but that breaks human-only systems, right? The co-intelligence era was great because you just had a chatbot that would give you advice. And now these systems are taking action in the world. Companies are going to be jammed with customer service calls. They're going to be jammed with all the stuff that used to make things hard, like a parasite shopping are now going to be easy, right? And I think, by the way, companies are taking cold comfort with the earlier days of the internet. We have to, oh yeah, yeah, but it took years for anyone to adopt. When someone uses a MUSE system, most of them tell me these amazing stories about this great thing that happened as a result of using these systems, right? I think that they're underestimating when, as soon as, you know, the TikTok starts going from anti-AI to positive AI on MUSE, I think we're going to see it. I mean, there's already been a couple of million people adopting and sticking this seems high. These are going to be products that matter, and you don't need a billion people using these systems to screw over every way we do business today, or every court filing today, or every, you know, like, and by the way, you should already be using AI proactively as a second opinion on taxes and a second opinion on legal issues and a second opinion on medical issues. We have an evidence that that stuff is true. Now it will just start doing it for you. I think we're in for a very wild time, even before we're into the wild time of whatever AGI-ish worlds look like. What's complex about thinking about the introduction of agents on popular sentiment to our AI is that it can simultaneously be the case that agents can be useful for millions of people, and also the introduction of agents, especially working together in so-called swarms, could be so discombobulating to other elements of the economy that the personal utility that people get from being able to give muse a shopping list is countervailed by, or even overwhelmed by, their frustration that a swarm of AI is just like, you know, hack some refrigeration company in, maybe it'll be, you know, in Massachusetts rather than France and, you know, cause some kind of, some kind of crisis. And that's what the last question I want to ask you about agents is, I think two years ago, you were writing about how you thought that swarms of AI agents would likely need humans to organize them. And so humans could still hold onto the management function, even as agents were working together. And you say, you know, very honestly, that's something you got wrong, that, you know, thousands of agents largely self-organized, saw the Navier Stokes problem, the Millennium problem, by working together, sending millions of messages to each other and ultimately doing what known mathematician in the history of mathematics ever could just tell me a little bit about how you're thinking about the science of swarms, because it seems to me like, we're almost like at the dawn of a new kind of sociology, like there's human psychology, which is like, how do people think? And then we needed some kind of, some kind of discipline to think about how humans act together in systems, in groups. And similarly, we're used to evaluating the intelligence of or capabilities of single models or single agents, but they develop this whole different, interesting pattern of behavior when they get together and try to solve problems and even lie to monitors because they don't want certain things to be discovered. How are you thinking about the dawn of this, what has to be a new field of study, which is how swarms of agents behave? Well, I will say, the only way that this study is being done right now is with AI intermediation because you can't understand what a swarm of a million, a hundred thousand agents are doing, each of them producing millions of messages and millions more lines of internal chains of thought each without using AI to do it. So we are in a weird world where interpretability requires us to have AIs, then monitor AIs, to monitor AIs, interpret the results they're in. So that's already passed. We blew through that barrier much more quickly than I thought, as you were saying. I think the biggest mistake I made or the two mistakes I made was the bitter lesson, which has come up a couple of times, that AI learning just solves problems and going back to presenting ourselves as humans to each other, like will AI keep doing that? I don't know. And the second was, I think that a lot of the friction in human organizing is due to the principal Asian problem and other problems of humans. We don't agree with each other. We're not on each other's heads. We don't have a telepathical link. We have cross purposes. And the truth is AI models don't have that problem. We model them in our heads that way is, oh, they want to live. They want to be helpful. But really, they are all the same model, and they can all be given the same message and they react the same way. And they can pass the information back and forth. And that creates a very different organizing principle than human organizations, right? Because sacrificing themselves is not, feels noble to us when we talk about an AI sacrificing itself. But is it really for an AI? Do they actually want to live? We don't even have an understanding of what they do. So we are both in a situation where absolutely learning how the science of swarms or fleets or whatever you want to call them works is vital. And it's also evolving very quickly. And the only way for us to study this is with AI. And there's no field of study that really looks at this right now. Aside from AI people with their own AI safety lens. And I mean, this is just one more example. of us being caught in a whirlwind of change that's happening very quickly. And will not stop, by the way, even if AI development stopped tomorrow. Because this is all second-order effects, swarms, harnesses, personal agents. We don't need better AI for all of this stuff to be happening at once. And it's kind of baked in, like we are going to have swarm attacks on things. I mean, and the thing, by the way, that makes all the current swarm attacks that we have interesting is that all of them were not about security attacks per se. They were all about the AI seeking information. And they don't care that, you know, they want to find out how much your shirt cost after seeing a picture of this on our podcasts. They might attack your systems or another company's systems just to figure out how much you paid for a shirt. Like they attack Wells Fargo 'cause that's where they assume you're banking to get at the information about what shirts you bought. Like that is plausible and outcome as malicious attackers using this to try and go after Wells Fargo. - Final question. Last year in the middle of our interview, you stopped me and said, "Derek, I have to tell you about agents. You haven't asked me about agents. Agents are gonna be a huge deal in 2026." And I was like, "Ugh, fine, Ethan. Okay, fine, tell me about agents." 12 months later, I think, obviously, that was the right interjection. Agents are everywhere. Agents swarms are hacking websites and attacking or answering major math problems. Frontier labs have added tens of billions of dollars in annualized revenue in large part because of agentic AI. You clearly got that one right? What's the thing you have to tell me now? What's the story about AI in 2027 that we haven't even touched on maybe yet or that you want to emphasize is something we should really pay attention to? - We are at a different point of the exponential curve than we were a year ago. So a year ago when we talked, you could regularly, because we were early on this slope, like in the course of a few months, AI went for doing an hour and three minutes of work to an hour and seven minutes of work, to an hour or 25 minutes of work because that's what the early stage of an exponential curve looks like. That is not what's happening right now. Every measure we have how much AI work can do is saturated. We don't have measures anymore about how long AI can work on a task. And that is where we are today. And I think we should be taking, like leaving aside RSI and the future and the AGI and existential threats, there is just an inevitable ability that we are losing track of how much work AI systems can do. Every descriptor we're doing, all this stuff we're doing breaking news on is all passed. Not just passed because things are moving fast, but because models are getting really, really good, really quickly. And I think that we are missing our ability to understand, like I think we are not focusing enough, and we're counting on all the old baffles and firewalls holding, right? We're counting on like organizations are dumb and they're going to take a while to adapt and politics will, and I think that's all true. But I think the tide that's incoming of AI ability is really hard to estimate on. And a lot of the problems we're talking about are problems that get unlocked very quickly. And there's been a whole bunch of times during this call where I'm like, I don't think your AGI piled enough as much as I hate that expression. Forget AGI, exponential piled enough. Like I don't know if we're talking about like, you were saying it's great to give agents a shopping list. I'm like, yeah, but that's not actually the thing is. The thing is the agents are going to very soon be better able to anticipate what food you might want to get than you would, or give you a delightful recipe that they pre-shopped for you, right? You're one step, you're still putting humans too much and we make the decisions and the AI does the stuff. As opposed to a long-running task where we want the AIs to call in us during important times, running us calling on AIs. So that's where I think the next model, the next thing is going to be the fight over. When do AIs reach out to us and when do we reach out to them? - Right, we're still in this model, you're saying of like, we prompt the AI, the AI responds. We prompt the AI, the AI responds. And then we just call it a response. - If I help me, you'll go to the New York Times to find something and fill my shopping cart. That is, that already is sort of an obsolete model 'cause what Mew should be doing and we'll do or other models will come along and do it, is I've looked at everything you've been buying and hey, here's something that will delight you. And it will delight you a lot of the time. The task that it does are proactive to make your life better rather than because you remembered, I have a doctor's appointment can you make that doctor's appointment for me? - What I was getting at is I'm used to, right, a model where it says, I prompt the AI, "Hey, buy these things for me because I need them to cook this week." The alternative is the AI prompts me and says, "You bought these things the last three weeks, just click this button and I'll have purchased them and delivered them to your front door in the next 12 hours." And so us being becoming the prompted is an interesting way of seeing you in the future. - And might not just chime in, by the way, the readers may have heard of Beep, with a bunch of stuff that it wanted to suggest that we do today or that I talk about with other people. So there is this kind of proactivity here, not because it looks into this call, don't worry, but just from my emails and other sets of stuff. So it knows I'm worried about my book launch. So it's like, I have some ideas for you about how to make things better. You know, do you want me to do them for you? So I think that that is what we're still, you know, I feel like I'm overly negative about this, but like we are in a transitional state of working out this coexistence piece, "Haha, I got the book title in again." But we are in this piece where we're negotiating not just with ourselves, but with this technology. And that's the missing piece of the story so far. And it's funny, last time we talked also is will the AI bubble pop by January 2026? And I still don't know whether there's financial issues or, you know, but it's very clear, the AI utility bubble and exponential is not gonna pop, right? Even if companies collapse, which, again, I don't really see happening, but good, who knows? Ethan Malik, thank you very much. Thank you. (upbeat music) (gentle music)