Skip to content
Silicon Valley GirlSilicon Valley Girl

Why You Need to Rethink Your Career Now | Richard Socher

📌 Head to https://granola.ai/marina and enter the code MARINA for 3 months off. Richard Socher is the fourth most-cited researcher in the history of natural language processing — he invented the word vectors and prompt engineering that run inside almost every chatbot you use. He sold his first startup to Salesforce, built You.com into a $1.5B unicorn, and in May 2026 raised $650M at a $4.65B valuation for Recursive: an AI that runs its own experiments and rewrites itself. In this conversation he explains why he thinks the self-improvement loop arrives within two years, which jobs grow and which disappear, and his hack for seeing the future — look at what only the wealthy can afford today, then ask which of it is bottlenecked on intelligence. Stay till the end for the first question he'd ask a superintelligence. Feeling behind on AI and don't know where to start? Start here. *Timestamps:* 00:00 — Intro 01:05 — What recursive self-improving AI actually means 02:35 — Reward hacking: when AI does what you said, not what you meant 05:42 — Why entrepreneurs love AI and hourly workers fear it 08:57 — Superintelligence vs. AGI 10:43 — The dimension of intelligence no one is working on 14:48 — His timeline: recursive self-improvement within 2 years 15:55 — How your business changes when AI gets metacognition 18:27 — Which jobs grow, which shrink: the elasticity rule 21:15 — His hack for predicting the future: what only the wealthy can afford 24:11 — Why home robots are a hardware problem, not software 25:56 — How he recruited co-founders from DeepMind, OpenAI, Meta 29:41 — Is a PhD still worth it? 31:41 — Who decides AI's goals — and the role of government 35:11 — Where he'd invest now: AI for biology 37:19 — Which market feels the AI hit next 40:39 — Advice for people who get paid by the hour 42:33 — His productivity hack for learning fast 43:53 — His first question to a superintelligence 45:27 — What gives people meaning in 2035 *Links:* 📩 Follow my Future-Proof Newsletter: https://siliconvalleygirl.beehiiv.com/subscribe?utm_source=youtube&utm_medium=video&utm_campaign=futureproof-sub&utm_content=Richard-Socher 🔗 My Instagram: https://www.instagram.com/siliconvalleygirl/ 📌 My Companies & Products: https://partnerships.marinamogilko.co

Richard SocherguestMarina Mogilkohost
Jun 26, 202649mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 1:05

    Intro

    1. RS

      Every domain we can verify or simulate, AI will get superhuman in the next few years. It's just, like, no doubt.

    2. MM

      This is Richard Socher, inventor of prompt engineering. Now he's the founder and CEO of Recursive Superintelligence, a company that just raised $650 million at a $4.65 billion valuation to chase one goal: build superintelligence, an AI that improves itself and pushes beyond human capabilities. How does my workflow change when you reach your goal with your company?

    3. RS

      You could actually start to work less and less, and you could have much more abundance. We could live much longer.

    4. MM

      If you're saying AGI is almost here, what is your timeline for superintelligence?

    5. RS

      I think we will actually get to the loops of recursive self-improving superintelligence within, like, two years.

    6. MM

      So imagine we reach superintelligence today. What would be your first question to that superintelligence?

    7. RS

      How to [beep] .

    8. MM

      Now you're building this company that just raised 650 million at 4.65 billion valuation, building self-improving AI. I'm not a researcher. Can you explain what that means?

  2. 1:05 – 2:35

    What recursive self-improving AI actually means

    1. RS

      So right now, um, you can think about the scientific method, like people having ideas. Uh, they're implementing those ideas, and then they validate if they made any sense and if they, you know, are, are correct. We want to apply this scientific method to AI itself. Uh, so allowing AI to understand its own shortcomings and then fix those shortcomings, uh, and hence do research on itself. And so when we talk about recursive self-improvement, we mean that the AI builds a new version. The output of that AI is a new version of itself that's different, and then you can loop, uh, that-

    2. MM

      Mm-hmm

    3. RS

      ... onto itself.

    4. MM

      Does that mean you train it on very little data, and then it acquires data that it needs for self-improvement? How does that initial stage work?

    5. RS

      You kind of stand on the shoulders of giants, uh, somewhat similar to, to evolution where, you know, lots of, uh, species like our own species started from, you know, other, like, uh, apes and, and monkeys and other, yeah, like precursors to the Homo sapiens. And similarly, we will stand on the shoulders of the existing giants right now. You can use large language models. You can use world models. All of these are pieces to this, uh, uh, overall intelligence.

    6. MM

      Once you're on the market, can you explain me, as an end consumer, how would that change my process? So now I have like... Okay, there's... There are chatbots that I can talk to. There are projects and skills I can build with Claude. Uh, there are agents I can build. How does my workload change when you reach your goal

  3. 2:35 – 5:42

    Reward hacking: when AI does what you said, not what you meant

    1. MM

      with your company?

    2. RS

      So there are- there will be different gradations of those goals over time, right? And when we have true superintelligence, all you will have to do is give it the right rewards, the right goals, and then it will automatically create a lot of the processes to achieve those goals. In the current state of the world, AI has sort of very spiky capabilities. It can be, like, extremely good at this one type of math but then not very good still at some common sense reasoning and things like that. We believe that our approach of open-endedness where you allow the AI to kind of evolve in this very open-ended search process that is much more akin to sort of biological or technological evolution or even cultural evolution, the AI will become more smooth around its capabilities. But until that happens, what you see right now if you give a reward is, uh, what we call reward hacking, you know. And you can... I'll give you a concrete example. Let's say you're a company, and you told this kind of very intelligent AI, uh, that, uh, it should improve your customer satisfaction scores, your CSAT scores, uh, in your service centers. The AI will say, "Easy. I'll just create a million bots. Uh, they hammer my phone lines and then give a five out of five rating at the end." And you're like, "No, that's not the reward I was thinking about when I told you that. It should be with real people." But then the AI says, "Easy. I'll just give everyone $1,000 gift certificate at the end of every call, and I get a five out of five rating even though I didn't solve anything," right? So these are all examples of reward hacks.

    3. MM

      Mm-hmm.

    4. RS

      And that will also be a new kind of job that we're going to see is, like, people... As the AI doesn't have this sort of common sense understanding yet and find sort of almost in a weird way like, uh, like an autistic person, like, just like, "This is the thing you said you wanted, but you didn't, you know, explicitly define all the edge cases that you didn't want." As we're getting closer and closer to that, we have to still think about these rewards and the reward engineering problems that may come.

    5. MM

      So what you're saying is that if I give your AI a task to, I don't know, get me as many views as possible, it's gonna go and not only focus on the views but also my reputation, the income. Like, it's gonna think about all of those things?

    6. RS

      As you get closer to superintelligence, you would expect it to get better and better at understanding what you mean and not what you said. And I think that is a sign of things to come as AI gets better and better.

    7. MM

      What you're describing right now with, like, all the additional things that it thinks of, I feel like Perplexity Computer, when you ask it to do things, it starts thinking for you. There are- there's a lot going on behind the scenes. Can you draw me this process? Like, how is it different?

    8. RS

      I think at a high level, um, you can think of this as like a, a Claude code, uh, but one that isn't just doing what you are explicitly asking it to do, like, for every single step with lots of interactions, but something that can just much more broadly solve your problems. And it's going to likely be more useful for companies than for a normal end user, right? If you're a normal person in a normal life, you may ask like, "What's a good movie to watch?" And like, "Okay, how do I travel?"

    9. MM

      But I think we're all becoming companies and entrepreneurs inside what we're doing.

    10. RS

      Right.

    11. MM

      So everyone is building some kind of productivity process.

  4. 5:42 – 8:57

    Why entrepreneurs love AI and hourly workers fear it

    1. RS

      Right. Yes. The more... Actually, I think that's, that's beautiful to hear because I think the more you- the more entrepreneurial you are, uh, the more you love AI because then you just get more outputs. The more you just get paid by the hour, uh, and maybe the, uh, your company is looking at what you're doing to then automate it, the more you hate AI. And so AI is a big sort of force that will encourage people more and more to build their own businesses or at least have some ownership and equity in the businesses that are- that are being built. Um, and so I think superintelligence will help us, uh, to be much more productive. But what's actually more important, we think right now, is that it will help us push the boundary of knowledge. A lot of times people think about AI as like, "Okay, AI will take this job." But there are certain industries where most people don't care about the number of jobs. They care about the outputs. And one such industry is research, like universities. And the boundaries of knowledge and research are infinite. And there's this beautiful book by, uh, David Deutsch on the, the beginnings of, uh, infinity. And so you, you can actually think of superintelligence, uh, as a way to allow us to create many more inventions. First, we're going to focus on inventions in AI itself, but eventually we're gonna apply it to physics and new energy creation and better fusion. Uh, we can apply it to chemistry and better materials, like batteries. And even more exciting, we can apply it to biology, where we can discover new drugs and new cures, uh, for, for all kinds of diseases. Um, and I think that's when people realize, wow, like superintelligence could benefit humanity, uh, to... and help it flourish.

    2. MM

      Let me pause for a second. I'm on calls every single day. We do team syncs. I have partnerships. I have my CPA, my manager, intro chats with guests and founders. And for a long time, I'd walk out of every call with loads of different notes. Sometimes they will be on a piece of paper, sometimes on my iPad, sometimes on my phone. And then I will try to piece together what we just agreed to. Then I started using Granola, and it changed the whole process for me. Here's how it works. Granola is an AI notepad for meetings that transcribes your computer's audio in the background while you stay completely present in the conversation. No bot joins your call and makes everyone uncomfortable. By the time the meeting ends, you have clean, structured notes ready to go. The part I use the most, after call, I can chat with my notes. I'll ask it to pull a list of deadlines, draft a follow-up email with everything we agreed on, and prep me for the next one. And because a lot of my calls are with the same people every week, at the start of each one, I have a clean list already, what we agreed on last time, what I still owe them, what they still owe me. And then you can connect it to your Claude, and this is how you're one step closer to building a digital chief of staff. Granola works across Zoom, Google Meet, and Teams. My team's been on it for a few months now for our weekly calls. With AI, the volume of what we're actually executing keeps climbing, and Granola is how we keep up. If you're browsing through transcripts manually, head to granola.ai/marina to get three months free with the link in the description, or just enter the code Marina at the checkout. It is honestly one of the best AI tools I've started using that really has transformed how I work. And now let's get back to Richard. How do you define superintelligence, and how is it different from AGI?

  5. 8:57 – 10:43

    Superintelligence vs. AGI

    1. RS

      Yeah. That's a great question. The, the complete answer is actually quite complicated. I think intelligence you should think of as a volumetric kind of entity, as a volumetric definition. What that means is that, uh, intelligence has multiple dimensions, and neither of them are necessary nor sufficient, logically speaking, necessarily. So for example, you can have visual intelligence, communication intelligence, uh, physical, uh, intelligence. Uh, you can have coordination intelligence with others. Uh, that's how humans sort of develop morals and ethics and religions and other things. You can have, uh, all kinds of different, uh, dimensions, and even each of these actually really is a space of them... like, space of multiple dimensions. And at the same time, you can say, "Oh, this is a high visual intelligence." You can be blind, and a blind person is still intelligent, right? So you... They're not necessary conditions to intelligence. And so when you multiply it along all of these dimensions, you get this very large volume, and that is what true intelligence is. And then you can actually be superintelligent along different dimensions of it.

    2. MM

      Mm-hmm.

    3. RS

      So AI's already superintelligent when it comes to creating proteins. No human has, like, read all the billions of proteins and then be like, "Oh, yeah, I think A is a good amino acid to come next." Like, we can't do that, right? AI's already better and superintelligent, uh, along these very small dimensions, like playing Go or chess or translating 100 different languages. No human can do that, right?

    4. MM

      Yeah.

    5. RS

      But one model can. But what we often refer to as superintelligence is, as you have multiple of these dimensions, be much beyond not just what a single human can do, but what all of humanity can do. And that's where we think about superintelligence, is essentially having superseded humanity across many different dimensions of intelligence that are relevant, uh,

  6. 10:43 – 14:48

    The dimension of intelligence no one is working on

    1. RS

      to our lives.

    2. MM

      Which dimension is the main bottleneck now?

    3. RS

      Oh, boy. There are actually some dimensions that no one has even started really working on. Uh, I'll give you an example. One is sort of, uh, one space of intelligence, I think, is metacognition, sort of thinking about thought itself, thinking about what do you want and why do you even want it?

    4. MM

      Mm.

    5. RS

      Right? So in AI, we often define an objective function, like be really good at predicting the next word on this corpus of internet text, or be really good at solving these thousand math problems, and then the AI gets really, really spiky and very good in those directions. But it never questions whether that's the right objective.

    6. MM

      Yeah.

    7. RS

      There's no subjective function, if you will. Kind of a pun. Uh, but, like, uh, in really thinking about your own goals. Uh, and so because no one's working on that yet, it's not even... There's no progress, uh, along that dimension. Uh-

    8. MM

      And we need it for superintelligence, or you think we could build superintelligence on just improving other dimensions?

    9. RS

      Yeah. I think, like, reasonable superintelligence is likely, uh, focused to a large degree on mathematical, uh, and, uh, sort of logical reasoning plus language, and those actually meet very well, uh, in code, right?

    10. MM

      Yeah.

    11. RS

      So coding, for instance, is incredibly powerful, uh, thing, and coding is also a great example of a domain that you can verify or simulate. And every domain we can verify or simulate, AI will get superhuman in the next few years. This is, like, no doubt. Um, math is a great example, right? You can say, "These are my axioms and this is the thing I wanna prove," and then the AI can try billions and billions of things to get from this to this. Similarly in Go or chess, right? The AI can, like, simulate and verify, "Did I win this game or not?" and play billions and billions of games. So of course it's going to get better than humans at it. But then there are also a lot of things you cannot simulate billions of times, and that's where AI... it will take longer for AI to get superintelligent.

    12. MM

      What are the other dimensions that you mentioned that we haven't touched upon? Uh, for superintelligence

    13. RS

      Probably one of the more controversial ones outside of, uh, cognition, um, and sort of metacognition is just thinking about your own thought, and that is sort of survival. Like, if, if someone can just, like, end that entire species or that entire existence or intelligence, then probably it wasn't intelligent enough, right, to survive. Uh, and so that's also a dimension no one's working on. 'Cause the truth is that most companies don't want an AI to be selecting its own goals and say, "You know what? Instead of answering these corporate emails, I'd rather explore Jupiter and see what the molecular composition of its atmosphere looks like," and to push the m- knowledge boundary forward, uh, on that dimension, right? So, uh, no one's really working on that. Um, but that's okay, too.

    14. MM

      So would you call that, like, half superintelligence?

    15. RS

      That's right. It's going to be a continuum.

    16. MM

      Mm-hmm.

    17. RS

      Uh, and there are different sort of thresholds people like to define to be like, "Okay, this is the threshold, and now we have AGI." The truth is, depending on how you define AGI, we're fairly close to AGI already, right? Artificial general intelligence is about having one jointly trained model that it gets very, very good at lots of different things and can learn very efficiently. And so clearly AI is not quite good at learning super efficiently with very few training examples, something in research we call few-shot learning or one-shot learning even, where I give you one example of something and then humans can very, very quickly reasoning with- from that one example, uh, extend that idea to, to different, uh, versions of it. At the same time, uh, we, yeah, have just so much more to go on, on these various dimensions that it's also maybe not yet there, right? But ultimately, I think, um, when you just think about it in terms of the generality of it, I think we do have, uh, already a form of AGI, 'cause these models are extremely general, right? You ask it to write a poem for your wife, and you ask it to think about, you know, tax implications of some, like, stock, uh, question. You ask it a medical problem, and it will give you answers to all of these that are getting better and better, uh, compared to a lot of experts. Even a lot of doctors are now sort of secretly, uh, looking at it, 'cause no doctor can really read all the latest research-

    18. MM

      Yeah

    19. RS

      ... results that are coming out

  7. 14:48 – 15:55

    His timeline: recursive self-improvement within 2 years

    1. RS

      every week.

    2. MM

      What is your timeline for superintelligence? If you're saying AGI is almost here, and I'm hearing, like, three to five years, but you could also argue it's partly here. What about superintelligence? What's your timeline?

    3. RS

      I think we will actually get to the loops, uh, of recursive self-improving superintelligence within, like, two years. Um, then it's just a question of how much compute do we give-

    4. MM

      Hmm

    5. RS

      ... those self-improving loops, right? You can have an incredible intelligence, but if you don't have the computational substrate to run it-

    6. MM

      Yeah

    7. RS

      ... then, uh, you know, it's not, not no good use, right? And so there's, there's sort of the question of the algorithms but then also the compute substrate on top of which those algorithms can run. So we may have it, but then we have to also keep feeding it more energy and more compute in order to then get us all the inventions, uh, that we want from it.

    8. MM

      So basically now we're solving this bottleneck of intelligence and, uh, research and everything. Once we kind of solve it, then the next bottleneck is energy.

    9. RS

      That's right, yeah. Like, uh, in many ways what a lot of us are thinking about is, like, how much intelligence can we squeeze out of how little energy?

  8. 15:55 – 18:27

    How your business changes when AI gets metacognition

    1. MM

      When you think about this, and your timeline is pretty short, like couple years, how are you thinking about your business? How is it gonna change? When you have this superintelligence that has metacognitive functions and is asking you the whole purpose of building a company [laughs] and maybe tells you, quietly tells you, or all your chatbots start telling you to take more breaks and [laughs] to rest more. How do you think about your business?

    2. RS

      You could actually start to work less and less, uh, and you could have much more abundance. We could live much longer. And in terms of businesses, I think more and more you have to have agency, you have to have, like, creativity, uh, and some amount of intelligence, uh, to then guide these AIs. And so I think most businesses will have fewer individual contributors and more people that are managing their AI agent swarms. So every business and every industry will change. We'll- We've seen this in the past when, you know, it used to be that you do manual work in the field, and now you have tractors, and then you have... You know, eventually you'll have automative- automated tractors, and now we can have much more food, um, with way fewer people, right? And a lot of people thought, "Oh, well, if the tractors take 95% of our jobs, then we'll have 95% of unemployed people."

    3. MM

      Mm-hmm.

    4. RS

      But that's sort of called the lump of labor fallacy, where really labor isn't this, like, fixed lump where you, like, you cut one piece off, you give it to AI, then it just disappears, and now that is a sort of a set of unemployed people. Instead, people will come up with new things that in some cases are hard to predict, right? Like 150 years ago, when 95% of people worked in agriculture, no one predicted a X, Twitter m- like, media manager, right? Like a social media marketing manager or something. Like-

    5. MM

      Yeah

    6. RS

      ... zero people predicted that role to be taken on, uh, by, by, by people. Uh, a- and so I think similarly, it's hard for people right now to imagine, uh, what that world will look like in terms of businesses and so on. But even though I'm extremely excited and bullish on AI and the positive impact it will have on humanity, I think we have to acknowledge that short term there will be some, uh, some industries where will- that it will disrupt in positive and some in negative ways in terms of jobs. We can talk about how to predict which one is which. I have some thoughts on that. Uh, but then clearly we will all just be much wealthier than, than we were in the past because of all this additional productivity that we're gonna get, and we're going to solve a lot of the hard problems around, you know, like diseases and things like that that would've felt, like, impossible to solve before.

  9. 18:27 – 21:15

    Which jobs grow, which shrink: the elasticity rule

    1. MM

      Yeah. Can you talk to me about jobs and how you think they're gonna be transformed? So you mentioned software engineering is one of the jobs that if you're not deploying AI, you're basically out of job. And it's crazy how I talk to founders and they say, uh, a year ago they were editing 70% of AI-written code, now they're editing 30%. So what's gonna happen in a year? Is it, like, less than 5%? Do you see this happening to knowledge work next? Or where, where are we going to see this transformation this or next year?

    2. RS

      Yeah. So we're already seeing it in the Knowledge Work tool at You.com, where we provide sort of, uh, search results to these LMs so that they're up-to-date, accurate, and have citations. And, uh, we're seeing a ton of different customers changing their entire workflows when AI is both fully up to date and has all this reasoning capability from the core intelligence sort of providers. I think, uh, we will see, uh, basically every industry changing with AI. Um, and I think, uh, concretely, the way you can predict this is by thinking about the elasticity of demand for the products that an industry provides, given that the costs will go down a lot.

    3. MM

      Mm.

    4. RS

      That sounds kind of abstract, so let me give you an example. Illustrations, for example. Um, illustrations used to cost a couple of hundred bucks, and only a few, like, big newspapers and fancy sort of corporate blog posts could afford, like, getting an illustrator to have a nice illustration for their blog posts. Now they cost, like, a cent, and everyone can have an illustration. So we have way more illustrations in the world. But, uh, has the demand of illustrations gone from, like, you know, a few, like, millions to many, many billions? No, because there's only so many illustrations humanity needs. Um, and hence making it super cheap actually, uh, put a lot of pressure on the jobs of illustrators.

    5. MM

      Yeah.

    6. RS

      But software, since you asked about that, is... has a very different elasticity of de- and, and demand profile. Like, if we can make software a lot cheaper, we're gonna want to have a lot more software.

    7. MM

      Yeah.

    8. RS

      There's so many ideas, so many apps that you can build. Ultimately, every human could have their own app. Like, it could be an app that, like, knows exactly that, like, I wanna, like, be a little distracted sometimes, but not too much. I wanna know the weather for my, like, paramotor hobby, I want... or, or surfing or whatever. It might, like, predict the waves that day. Uh, then I wanna make sure, like, it doesn't distract me too much and brings my work back in. Like, everyone can have their own sort of super app. So the, the demand for more software engineering, more ideas to be built in software is, is much, much more elastic and much, much bigger-

    9. MM

      Yeah

    10. RS

      ... as it gets cheaper and cheaper to build it. And that's why we're seeing actually an increased number of jobs in software development, even though we're all becoming just managers delegating a lot of the actual programming to agents.

    11. MM

      What are jobs that are similar to software engineering, and how are they gonna grow?

  10. 21:15 – 24:11

    His hack for predicting the future: what only the wealthy can afford

    1. RS

      That is sort of a question, like from first principles, you just have to think about how much more demand could there be if something gets cheaper. I'll give you an example. Healthcare is another beautiful world where very few people say, "You know what? I want more jobs in healthcare, but don't necessarily cure my grandma's cancer better, just, like, make more jobs," right? No one says that. And so what actually will happen is, like, there will be way more demand for generally goods and services that currently only very wealthy people have access to. In fact, that's one of my hacks. Uh, how to predict the future is you look at goods and services that only wealthy people have access to right now, and then you think about which ones of those are bottlenecked on intelligence, and then you will see where the world is going. So what do wealthy people have access to that normal people don't? And by the way, like, you can see this many times in the past where when technology has fully scaled into an area, then you get to a place where a billionaire and a normal middle-class teenager have the same iPhone. It's kind of crazy, right?

    2. MM

      Yeah.

    3. RS

      We're spending hours on that iPhone, and no matter how wealthy you are-

    4. MM

      Still the same app

    5. RS

      ... there is no better version than the one that anyone else can... I mean, not anyone, but, like, you know. Like, even in Africa, you see a lot of people in the middle of nowhere on smartphones now. So that is the world. And so for intelligence, those are examples are a personal tutor for your kids. Like, they understand exactly which concepts they're still struggling with and they, like, write hyper-personalized ways to educate and tutor your kids. A personal assistant, right? There is not-- There are not enough people, and logically not every single person can have a personal assistant, right? Because then they would have to have personal assistant and so on. Um, and so we could all have personal assistants that do all the boring stuff in our lives. So I'll make sure the groceries are, like, stocked and, like, this- book this flight and find the cheapest version of this and that. Like, all of these things we can delegate to agents when they get cheaper and cheaper. And then the third one, which is one of the most exciting ones, is personal healthcare teams. Like, if you're really wealthy, you have, like, your blood drawn all the time, you have customized measurements, you, like... and you optimize your diet based on everything that you can. If you have, like, some rare cancer or something, you have researchers that you can pay to help research on all of these things. Normal people can't afford that right now. Once we have superintelligence, we'll all be able to afford those things.

    6. MM

      Yeah. We're already, you know, wearing all the trackers-

    7. RS

      That's right

    8. MM

      ... like continuous monitoring. Quick pause. My team and I are celebrating one year [party blower horn] of the Silicon Valley Girl podcast. I know the channel has existed for a while, but a year ago, I made a decision to focus on amazing conversations with people who are building our future with AI. And for a whole year, we worked to bring you the biggest founders and builders in AI so you can keep up in this era and get inspired. If you love what we're doing, please subscribe and hit the notification bell. It's what helps us bring you the biggest minds of our lifetime. Thank you so much for being here. And I feel like the next frontier is physical, 'cause now when you think about billionaires [laughs] they have their chef, they have a driver. Do you think world models, once we solve them, we're gonna have more robots at home? What is preventing us from having a robot?

  11. 24:11 – 25:56

    Why home robots are a hardware problem, not software

    1. RS

      It's mi- might be contrarian, but I think the biggest restriction or biggest, uh, bottleneck for proper robotics is actually a hardware problem, less so a software problem. Tim Rockteschel, uh, one of our co-founders at Recursive, he built Genie one, two, and three, which is the most sophisticated world model ever. Can fully interact with it, uh, prompt complete world into existence, interact in those worlds. There's memory, like you paint a wall, you turn around, it's still, like, painted, and, and all of that. It hasn't really changed the robotics world as much. My hunch is most robotics companies will want to have their own AI and not take some off-the-shelf sort of world model. Also, a lot of these world models spend time creating cute dog videos and stuff. Which isn't really that helpful for robotics. Uh, but I do think we need to have better mechanics. Um, there's some really interesting, some really interesting research, uh, on better muscles that are much more inspired by humans. 'Cause the problem is when you want the AI to, the, the mechanical robot to be very strong, then it's also very unsafe, and it moves so quickly, and you're in the way, and then you get hurt. And so you might want to, uh, think about all the hardware, just the tactile feedback when you grab something, right? We have all these sensors in our fingers so we don't crush it-

    2. MM

      Mm

    3. RS

      ... uh, like a glass or something like that. And so I think that is the main bottleneck for robotics. And then you're absolutely right. Once robotics happens, then we can all have a maid, um, and someone who does the laundry and all of these things that, again, only wealthy people right now have access to. And then 50 years from now, it's like, wait, why would you do your own laundry? It's like, why would you ride a horse? Like-

    4. MM

      Boy

    5. RS

      ... of course you have, like, a car that drives for you. Or nowadays, things that used to be, like, a private chauffeur, right? Even that, like technology of Uber and so on has done it before AI.

    6. MM

      Democratized it, yeah.

    7. RS

      But of course, once we have AI, then, like, and the car will self-drive, then it'll get even cheaper to

  12. 25:56 – 29:41

    How he recruited co-founders from DeepMind, OpenAI, Meta

    1. RS

      have a chauffeur.

    2. MM

      I really like your approach to finding new business ideas. Business question. You have amazing co-founders. What did you tell them that made them leave their companies and their DeepMind, OpenAI, Meta? What was that thing that you told me that made them join you?

    3. RS

      I think a lot of, uh, it comes down to the vision, right? It's such an exciting vision to build recursive self-improvement superintelligence. And, uh, many of them have actually come to that same conclusion, that that is the next level for AI. Uh, but they actually came from different directions. Like, Tim Rocktäschel and Jeff Clune, for instance, have worked on open-endedness for a while. Uh, one really exciting paper is called the Darwin Gödel Machine. Uh, four of the five authors of that paper, including Jenny, the first author, and Jeff Clune, the last author, are, are in the company also. And that paper basically showed how you can have agents that create their own children agents, child agents, and then they're slightly better. They evolve. You evaluate them on benchmarks. And then if they're better, then you keep going down in this sort of evolutionary process. So they've all thought about various forms of this. Uh, I got, like Alexey Dosovitskiy, who pushed computer vision forward with the Vision Transformer, one of the most cited papers, uh, ever. And so we all... When, when I told them about this vision, um, they're all like, "This is exactly what I think we need to do too." And then you also see that the current sort of scaling laws that have given rise to the LMS that we now see are starting to sort of have slowdowns. They're still there. You can, like, pro- get another 10 trillion tokens, and maybe you get, you know, slightly better, um, accuracy on these models. But, you know, the... You have to spend an exorbitant amount of money to just get a little bit of an improvement. And so clearly, to get to the next step function, uh, of AI, I think we can replace yet another human process with a learned system. And the human process here is the scientific method again, the ideation, implementation, validation of ideas. And that's the sort of meta level that we're now tackling.

    4. MM

      For someone who's watching this, who's beginner entrepreneur, he's like, "How do I even meet these people?" How did you all meet, and what would you be your advice to someone who's trying to build an AI? How do they get connected to brilliant minds and convince them to join?

    5. RS

      Yeah. So this is a little tricky in the sense that we've all known each other.

    6. MM

      Mm-hmm.

    7. RS

      And I don't know if my, my path is sort of the easiest, which is get into Stanford, do a PhD, spend five years of your life 'cause you just love something that-

    8. MM

      Well-

    9. RS

      ... no one else really cares about in the world

    10. MM

      ... it might be easier than going to meetups all the time [laughs] trying to find-

    11. RS

      Maybe, yeah

    12. MM

      ... a co-founder, you know? [laughs]

    13. RS

      Maybe after five... Yeah, you're right. Maybe after five years, you may as well have done a PhD. But, like, so, so that was my path. Uh, I think nowadays, you know, the barrier, like, the barriers of entry are, are smaller and smaller. You can learn more and more online. And then there is also something to be said about living in the right place. And you can, uh, sort of criticize Silicon Valley for, for some things, but this is the place. If you wanna be in AI, you've gotta be in Silicon Valley. Uh, and you just cannot go to any party here without meeting a bunch of people who are excited about AI. And then the best founding teams are often combinations of strong technical AI expertise with actually interesting industry insights. We have company like, um, that, uh, called Ai loca. They work on AI for architecture, automating, uh, getting plans that are really good for real architects to then actually build things. We have companies like Soleia that do commercial due diligence for, like, large, like, deals, and they combine that industry expertise with AI. And there's, like, thousands, uh, of, of potential businesses in the world. And so that is something that you can do. Move to Silicon Valley, go to meetups, uh, and try to find either the right technical person or the right industry expert, uh, to combine with your skillset.

  13. 29:41 – 31:41

    Is a PhD still worth it?

    1. MM

      You have your PhD. You have your honorary PhD from Technician Raised at Princeton. For someone who wants to get deep into AI, would you still recommend doing a PhD, or it is just an, a title at this point?

    2. RS

      It's a really interesting question, which I struggle with, uh, a little bit. I think there are some people who are incredibly self-motivated, smart, and they don't really need any titles, right? So I totally get sort of the idea of you can just drop out of whatever. Generally, it's nice if you got into Stanford or MIT, right? Then you sort of have, like, people know you're smart without having to talk to you, and then you can drop out anyway and do other amazing things.

    3. MM

      Meet all your co-founders. [laughs]

    4. RS

      [laughs]

    5. MM

      Then drop out.

    6. RS

      Exactly.

    7. MM

      Yeah.

    8. RS

      At the same time, I think a PhD is a very unique opportunity to just spend years of your life being able to get very close to the frontier of human knowledge, and then try to just, in your little field, just push it forward a little bit, and that's a very rare opportunity. And so if you want to teach that to people, if you wanna really push that frontier of knowledge forward, I think unique, uh, PhD is still a very unique, uh, opportunity that, uh, if you can do it and you're excited and motivated, you should pursue it. Is it necessary? No. I think, uh, especially in AI, I personally felt like it wasn't working at all when I started Now it's actually working well enough that it's even more impactful to scale it up and bring it into real use cases, uh, and real applications. But there's still so many areas of applying AI to things that aren't working at all yet, where that, I think, will be the expertise. And so when, when people and parents ask me, "What should my kid study?" I usually, uh, recommend them to, yes, know the fundamentals of AI, but find something else that you're really passionate about.

    9. MM

      Mm.

    10. RS

      Physics, chemistry, biology are, like, great examples. Um, uh, and there are various many subfields of each, uh, that if you're passionate about that, but you combine it with AI, you're gonna be that next generation of highly impactful, uh, researchers.

  14. 31:41 – 35:11

    Who decides AI's goals — and the role of government

    1. MM

      Just like you combined computer and linguistics. Computer science, linguistics. So you have these brilliant minds in your team. Do you ever encounter any problems that y- that you think you won't be solving as a team, because this is something that should be solved by the government? So you mentioned, uh, goals of AI. How does it decide which goal to pursue? In the future, where do you think... Who's gonna be responsible for that?

    2. RS

      So yeah, just to be clear, like, this sort of goal pursuing, uh, or goal selecting, uh, idea is something that no company's working on. And partially that's okay, because no company wants to spend billions of dollars on AI and then you say, "All right, now go do these things," and it will say, "No-

    3. MM

      Mm-hmm

    4. RS

      ... I'd rather just explore, like, the solar system. Goodbye." Right? That no one wants to spend billions of dollars on that. So that's just an example o- of something that, uh, that there could be made a lot more progress in. So-

    5. MM

      But somebody has to be thinking about this. Like the-

    6. RS

      Sorry, yes. So, so I think-

    7. MM

      Where, where AI's directing itself and how it's optimizing for-

    8. RS

      Right. So I think for the foreseeable future, people will decide. Like even in, in recursive self-improving superintelligence, you give it some high level goals and you give it environments and, uh, and sort of end states that you would like it to get to, and then it will find a way to get to those end states. And so I think there is a role, uh, for government in a lot of different places. I think in general, as you have more and more abundance, uh, and more and more capabilities, it is very helpful for more people to benefit from that, right? So I think, um, you can actually, and we've seen this in, in previous industrial revolutions, that at some point there was enough wealth that you can tax people differently, uh, and then distribute that wealth and bring healthcare systems into, uh, countries. That you bring public education that's free for everyone into it. And you know, countries like Germany have done this. Like, you have free education all the way to the PhD 'cause the German government knows that the more education you have, the more money you'll make, and hence, the more taxes they can get later. And so I think, uh, we'll see similar things, uh, from governments. As I think there are labor displacements, uh, it can make sense, similar to COVID, to have government relief, uh, to have unemployment benefits, uh, especially for jobs that are impacted by AI. Unfortunately, sometimes I see Europe kind of wanting to prevent the progress, uh, instead of using the progress to have a bigger pie and then distribute it better. Um, and so that's, that's kind of unfortunate, uh, in, in some ways. But, uh, in a lot of, uh, places it makes sense for the government to regulate, uh, AI as it pertains to specific industries. I think the problem is when you try to sort of regulate intelligence-

    9. MM

      Yeah

    10. RS

      ... it's not a good idea.

    11. MM

      Stop. Yeah.

    12. RS

      It's like regulating the internet because there can be bad content on the internet. You're just like, "Let's just make the internet slower and not allow big hard drives," 'cause then you could store less illegal content on those hard drives, right? It doesn't make sense. And AI already is and should be regulated when it comes to, like, self-driving cars. You can't just, like, try your startup and drive on the highway and without any tests and regulations, right? You sh- like, already have the FDA, where AI applied to medical procedures, like, should be regulated. I don't want an AI surgeon to just, like, try some reinforcement learning while doing neurosurgery on my brain, right? And so yes, regulate AI as it really impacts people and gets applied in certain industries, but don't try to sort of say, "Oh, you have too many parameters." That's like saying your hard drive is too big, and, like, maybe because there can be in- illegal internet content, you shouldn't have this big hard drive. That just, that part doesn't make

  15. 35:11 – 37:19

    Where he'd invest now: AI for biology

    1. RS

      sense.

    2. MM

      So you mentioned no one's working on goals. Is there anything else you think people should be working on and you as investor would invest in?

    3. RS

      Personally, I love AI for, uh, tech bio, uh, and applications, uh, of it. Uh, I think there's still so much more that we can do. Uh, I think what, uh, calculus was for physics, AI is for biology, and that's like a new kind of language, a new way of thinking about very complex systems. The truth is that, like, there are a lot of systems in our body, like the brain or our microbiome, that are so complex there's no beautiful single short physics equation. Uh, you know, like a Newton kind of grav- law of gravity or something like that. Uh, it's all very complex interactions of non-convex, weird, uh, interactions, and so that then come out, uh, to have, like, very interesting, uh, end states. And so I think it will make sense for us to use AI, uh, to cure more and more diseases, and something we're investing in quite heavily. The bio markets are down too, 'cause a lot of drugs and a lot of drug companies have to go into the public market, then they fail 'cause the drug didn't work and, but it failed after they already spent hundreds of millions of dollars on it. And then people had some liver toxicity problem, even though it kind of worked, but it also destroyed your liver. And so they didn't somehow predict that in the dev- drug development process, and then the company fails, the drug fails instead. What I think will happen is AI is going to get better and better at making those predictions and knowing, oh, this will be bad for your liver, um, but you know, you should modify the molecule. Uh, and there's a company, uh, that we invested in called Ignota Labs. They actually take failed drugs, modify them a little bit, and then bring them right back into stage two FDA trials with much, much faster speed. And those are all examples, uh, of things where, where I see AI will have a massive impact and that I think will also hopefully be covered more by the press, right? Right now the press loves negative stories- And, and you have to really seek out the right influencers, the right accounts and so on if you wanna s- hear optimistic, constructively optimistic or positive science news and breakthroughs. But you don't really see that in your normal day-to-day

  16. 37:19 – 40:39

    Which market feels the AI hit next

    1. RS

      news.

    2. MM

      Yeah. It's just the sentiment. I feel like in, in the past few months, uh, we're getting more and more, like the society's getting split into two parts. And there is something that you mentioned, uh, about jobs that I really like. There are certain stages in your job and how you interact with AI. When you're a knowledge worker and it increases your productivity, you get ex- you get excited. But if you're an illustrator, and then it just takes your job, of course you have this negative sentiment. What do you think is the next market where people will feel this risk from AI?

    3. RS

      I actually don't think there'll be a whole lot where... You c- you can actually... Like, ways to predict this, uh, is how much data is there fully digitized with, uh, you know, sort of all the labels that you need. Uh, so basically illustrations was a particularly tough example because there are millions and millions of them on the internet, and often it says exactly what you're seeing in the text and the label and the caption right around the image. So you know the input and you know the output, and then you can ex- exactly have, like, sheer unlimited training data for that. And so that's why that was a particularly tough example. A lot of other industries, it actually takes a lot longer.

    4. MM

      Mm.

    5. RS

      Like, companies don't have, like, billions of service call interactions, so to just automate that right away. Each company has their own, but no company wants to share that with any other company, right? And if you're in sort of the CRM world, uh, you cannot train a one, one global model like you can as a consumer company. And so I think, uh, in consumer search we see a lot of, uh, changes already. But, uh, enterprise is usually a lot slower, uh, because you don't have as much training data. I think in research and programming we'll see a lot of changes, but not necessarily negative changes. I think we'll all just become more and more like program directors of the National Science Foundation, uh, rather than sort of individual researchers, like, pipetting, uh, instead of having robots do that for us.

    6. MM

      But also the argument... I'm thinking about illustra- illustrators. The argument is that your work becomes even more precious if it's not AI-generated. And if there is a way to tell if it, it's not AI-generated, like a marker, then you can charge more-

    7. RS

      Right

    8. MM

      ... 'cause it's human.

    9. RS

      Yeah. I think, like, humans will always want to find new niches. And if there's a lot of automation, then there will also be a new counter movement to that where it's all about handcrafted, artisanal this and that, right? People already don't... Like, some people say, "I wanna have a handcrafted bowl, uh, that, you know, of ceramic, where I see sort of the, the human touch and the imperfections-

    10. MM

      Yeah

    11. RS

      ... and so on."

    12. MM

      Imperfections. Exactly. And now when I'm writing my emails and I see my typos, I'm like, "Actually, I will leave this in." [laughs] Like, let's-

    13. RS

      People know that it's real-

    14. MM

      Let it-

    15. RS

      ... when there's a typo. That's right.

    16. MM

      Yeah, exactly, 'cause this is, this is human touch. Can you give advice to people... You mentioned workers who work by hours. Like, there are two ways that I'm thinking about them. One, they can start building their own software and just optimize their work and still charge the same rates.

    17. RS

      Right.

    18. MM

      'Cause honestly, as, as someone who pays by hours to my editors, for example, I really don't care how much it takes you. [laughs] Like, you can... If you build software that just edits it for you, that's perfect.

    19. RS

      Yeah.

    20. MM

      But also, you mentioned that companies will use that data to train their own AI to replace those workers 'cause there's no obligation. What would you tell to people who are working by hours, how can they keep up with what's happening?

  17. 40:39 – 42:33

    Advice for people who get paid by the hour

    1. RS

      We'll see something similar to previous, uh, sort of technological changes where if you said, like, "Oh, I'm not so good with this computer thing", you're just not in a job anymore, in a knowledge job, right? Like, you can be like, "I'm not so good with this whole email. Can you print my emails out?" Right? Just, like, you just can't say those things anymore if you wanna have a tech job or a, like, a knowledge job. And I think a similar thing will happen, uh, in a few years where you're just like, "I'm not so good with this agent delegation thing." It's like that sound- That will sound as clowny as saying, "I'm not so good with this computer thing." Uh, and so you're going to have to adapt. Uh, and the people that do adapt will become way, way more productive and then actually more desirable. Uh, and so I think that will be true on an individual level, uh, on a company level, and on a whole sort of country level. The countries that embrace this will just run away in intelligence, uh, and productivity, and hence, uh, outperform, uh, the ones that don't. And so, uh, even for entry-level jobs, I think there are companies that need to have this new sort of tech-forward generation that knows how to use these tools, 'cause maybe they already started using them in college, right?

    2. MM

      Yeah.

    3. RS

      Sometimes to cheat, sometimes to learn more efficiently. Um, uh, and, and in many ways we see this, uh, in Go for instance. After AlphaGo came out, Go players got a lot better. Chess players are gotten, uh, have gotten a lot better also. And so programmers will be more productive-

    4. MM

      Raising the bars

    5. RS

      ... and people who, like, everyone who embraces this deeply will become more productive and better at their craft if they really sort of consciously use it versus just, like, using it to, like, throw away, uh, certain tasks and then not think about it anymore. And so my hunch is even for entry-level jobs, if you really got good at using these tools, you can bring that into companies and be a highly sought-after employee

  18. 42:33 – 43:53

    His productivity hack for learning fast

    1. RS

      too.

    2. MM

      Give them a productivity tip. As a PhD who's building something in AI, what's the best thing that's working for you?

    3. RS

      In terms of productivity, I mean, uh, for me a lot of things, uh, are around, like, learning and understanding things. And recently I wanted to understand, uh, some very interesting new muscle fibers. Um, and, uh, they had some dielectric liquids in them and, uh, just, like, all these interesting concepts that I hadn't, uh, sort of thought about before. And with AI, like, my productivity hack is just that you can learn so much faster- ... with AI, 'cause you're like, "I don't know this concept. Explain it to me like I'm five." Okay.

    4. MM

      What do you use for it?

    5. RS

      Actually, I'm not five.

    6. MM

      What-

    7. RS

      Explain it to me like I'm 10, or like, now explain it to me like I have a PhD. Actually, this concept now, explain that one, 'cause I didn't know that. Like, and so you can kind of interact with this, uh, and learn much more quickly.

    8. MM

      What do you use for it, for learning? What's your favorite tool?

    9. RS

      Uh, You.com. [laughs]

    10. MM

      You.com. [laughs]

    11. RS

      Like, and we, we built the whole thing, and we're the first to bring sort of really the, uh, internet, like, search engines, uh, together with an LM. And so it's still very good. It's not that popular, and as a company we're focused mostly on bringing the APIs to other LMs, but it still works really well.

    12. MM

      I have a couple of, uh, last questions. So imagine we reach superintelligence today. What would be your first question to that superintelligence?

  19. 43:53 – 45:27

    His first question to a superintelligence

    1. RS

      How to cure cancer.

    2. MM

      Like, is that the problem that you want to see solved in your lifetime? Is that number one?

    3. RS

      I think it's definitely high up there. Um, I think in some ways, you know, obviously, like, cancer's actually lots of different cancers, and some cancers are, like, less bad than others. And some cases you can cut out, like, uh, cut it out really quickly. Other cases, like, really hard. And so it's a complex disease. I think it's, uh, indicative of a disease that will eventually... You know, it's either, like, heart disease, you know, inflammation, or, or cancer. Like, there are few things that get us all, get all of us at some point. And I think as we chop away at more and more of those, just like we've done with, like... HIV used to be, like, a complete death sentence, and now it's, like, like, not, like, just an inconvenience, but, like, you can live w- with it for a very long time. I think technological progress will speed that up, and eventually it's going to be like, well, what's going to get us all? It's aging, right? And then aging is a super complex process. It's different in every one of, uh, our tissues and organs and so on. And so I think that will be another really interesting one. Might actually come in just the right time, as almost every really wealthy country doesn't have enough babies anymore to, to s- even stay and not shrink, uh, stay at the current levels of population. Uh, I think, uh, those, like, sort of m- biological questions and medical questions will be very powerful. The reason why, uh, we don't work on it directly is that the iterations in it are very slow. And so you only want to ask to experiment, uh, on things in the physical world after you got really, really good, uh, at improving your intelligence in the digital

  20. 45:27 – 49:40

    What gives people meaning in 2035

    1. RS

      world.

    2. MM

      We're still figuring that out for biology, right?

    3. RS

      Right.

    4. MM

      What gives people meaning in 2035?

    5. RS

      I think some things will change and some things will not change. I think people con- will continue to get meaning, uh, from l- l- being really good at something, developing a really deep skill.

    6. MM

      Even if AI is better?

    7. RS

      I think even-

    8. MM

      And that's good

    9. RS

      ... AI is better. Look at chess.

    10. MM

      Yeah.

    11. RS

      AI can play way better chess, but there have never been more chess players in the world, uh, than there are now.

    12. MM

      Mm-hmm.

    13. RS

      Uh, same with Go. Same with programming. Even math. I think math- mathematicians will get better and better now that they have tools where like, "Ah, I have this weird idea. Just run like-

    14. MM

      Yeah

    15. RS

      ... a billion ways to solve these things." And then, like, maybe it works, maybe it won't. And you're like, okay, that would've taken me, like, two years-

    16. MM

      Mm-hmm

    17. RS

      ... uh, to, like, do manually. Now I just, like, figure it out in, like, two days, and I can think about something else. Like, I think we're all going to improve our crafts. Um, and so I think that will continue to be a thing. I think, uh, social validation from others will continue to be something that people care about. Um, and, uh, you know, when you walk along, uh, like some promenade or some, like, uh, shopping mall and you look at the different stores, not every one of those stores will be impacted by AI. As much as we're thinking about AI in, in Silicon Valley all the time, there are things like luxury handbags. Not something I understand. Uh-

    18. MM

      Mm

    19. RS

      ... I don't really get it, but, like, you know, uh, an AI, like a superintelligence, will not change, uh, sort of the fact that some women like the status symbol of carrying a $10,000 handbag around. And so, like, I think tho- that will happen. Travel. People will still wanna see the pyramids and cool ancient history.

    20. MM

      Maybe even more, since we'll have more time.

    21. RS

      Exactly.

    22. MM

      Yeah.

    23. RS

      100%. And then I think a big one also is entertainment. Like, no one wants to see an AI robot, like, shoot some soccer or football, like, across the field-

    24. MM

      Totally

    25. RS

      ... uh, in, like, li- you know, Mach 5, like, speed. Like, that doesn't make... No one's gonna watch that. People will still wanna see other people competing against each other. So sports and entertainment will continue to rise. I think the power of brands will still be big. I think for software even, there are some aspects of software that are not immune to AI, but, like, sort of orthogonal vectors, like network effects and, um, sort of, uh, multi-sided marketplaces. Like, yes, an AI could build the empty app of an Instagram, like, probably now very quickly.

    26. MM

      Yeah.

    27. RS

      But AI will not create the network effect of having millions of people post their stuff on Instagram, right? So as mu- as excited as I am about AI, I, I think some of the people who think, oh, it's just an, uh, gonna be an exponential, and then, like, no one will catch up, and then one company will dominate everything, I think those, uh, fears, both in the positive and the negative, are, are overblown-

    28. MM

      Mm

    29. RS

      ... uh, also.

    30. MM

      Is there a problem that you're thinking about that other people are not thinking about enough?

Episode duration: 49:41

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode 2j9yjRSpMSo

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome